1. Introduction

Welcome to the official Infinispan User Guide. This comprehensive document will guide you through every last detail of Infinispan.

1.1. What is Infinispan ?

Infinispan is a distributed in-memory key/value data store with optional schema. It can be used both as an embedded Java library and as a language-independent service accessed remotely over a variety of protocols (Hot Rod, REST, and Memcached). It offers advanced functionality such as transactions, events, querying and distributed processing as well as numerous integrations with frameworks such as the JCache API standard, CDI, Hibernate, WildFly, Spring Cache, Spring Session, Lucene, Spark and Hadoop.

1.2. Why use Infinispan ?

1.2.1. As a local cache

The primary use for Infinispan is to provide a fast in-memory cache of frequently accessed data. Suppose you have a slow data source (database, web service, text file, etc): you could load some or all of that data in memory so that it’s just a memory access away from your code. Using Infinispan is better than using a simple ConcurrentHashMap, since it has additional useful features such as expiration and eviction.

1.2.2. As a clustered cache

If your data doesn’t fit in a single node, or you want to invalidate entries across multiple instances of your application, Infinispan can scale horizontally to several hundred nodes.

1.2.3. As a clustering building block for your applications

If you need to make your application cluster-aware, integrate Infinispan and get access to features like topology change notifications, cluster communication and clustered execution.

1.2.4. As a remote cache

If you want to be able to scale your caching layer independently from your application, or you need to make your data available to different applications, possibly even using different languages / platforms, use Infinispan Server and its various clients.

1.2.5. As a data grid

Data you place in Infinispan doesn’t have to be temporary: use Infinispan as your primary store and use its powerful features such as transactions, notifications, queries, distributed execution, distributed streams, analytics to process data quickly.

1.2.6. As a geographical backup for your data

Infinispan supports replication between clusters, allowing you to backup your data across geographically remote sites.

2. The Embedded CacheManager

The CacheManager is Infinispan’s main entry point. You use a CacheManager to

  • configure and obtain caches

  • manage and monitor your nodes

  • execute code across a cluster

  • more…​

Depending on whether you are embedding Infinispan in your application or you are using it remotely, you will be dealing with either an EmbeddedCacheManager or a RemoteCacheManager. While they share some methods and properties, be aware that there are semantic differences between them. The following chapters focus mostly on the embedded implementation. For details on the remote implementation refer to Hot Rod Java Client.

CacheManagers are heavyweight objects, and we foresee no more than one CacheManager being used per JVM (unless specific setups require more than one; but either way, this would be a minimal and finite number of instances).

The simplest way to create a CacheManager is:

EmbeddedCacheManager manager = new DefaultCacheManager();

which starts the most basic, local mode, non-clustered cache manager with no caches. CacheManagers have a lifecycle and the default constructors also call start(). Overloaded versions of the constructors are available, that do not start the CacheManager, although keep in mind that CacheManagers need to be started before they can be used to create Cache instances.

Once constructed, CacheManagers should be made available to any component that require to interact with it via some form of application-wide scope such as JNDI, a ServletContext or via some other mechanism such as an IoC container.

When you are done with a CacheManager, you must stop it so that it can release its resources:


This will ensure all caches within its scope are properly stopped, thread pools are shutdown. If the CacheManager was clustered it will also leave the cluster gracefully.

2.1. Configuration

Infinispan offers both declarative and programmatic configuration.

2.1.1. Configuring caches declaratively

Declarative configuration comes in a form of XML document that adheres to a provided Infinispan configuration XML schema.

Every aspect of Infinispan that can be configured declaratively can also be configured programmatically. In fact, declarative configuration, behind the scenes, invokes the programmatic configuration API as the XML configuration file is being processed. One can even use a combination of these approaches. For example, you can read static XML configuration files and at runtime programmatically tune that same configuration. Or you can use a certain static configuration defined in XML as a starting point or template for defining additional configurations in runtime.

There are two main configuration abstractions in Infinispan: global and cache.

Global configuration

Global configuration defines global settings shared among all cache instances created by a single EmbeddedCacheManager. Shared resources like thread pools, serialization/marshalling settings, transport and network settings, JMX domains are all part of global configuration.

Cache configuration

Cache configuration is specific to the actual caching domain itself: it specifies eviction, locking, transaction, clustering, persistence etc. You can specify as many named cache configurations as you need. One of these caches can be indicated as the default cache, which is the cache returned by the CacheManager.getCache() API, whereas other named caches are retrieved via the CacheManager.getCache(String name) API.

Whenever they are specified, named caches inherit settings from the default cache while additional behavior can be specified or overridden. Infinispan also provides a very flexible inheritance mechanism, where you can define a hierarchy of configuration templates, allowing multiple caches to share the same settings, or overriding specific parameters as necessary.

Embedded and Server configuration use different schemas, but we strive to maintain them as compatible as possible so that you can easily migrate between the two.

One of the major goals of Infinispan is to aim for zero configuration. A simple XML configuration file containing nothing more than a single infinispan element is enough to get you started. The configuration file listed below provides sensible defaults and is perfectly valid.

<infinispan />

However, that would only give you the most basic, local mode, non-clustered cache manager with no caches. Non-basic configurations are very likely to use customized global and default cache elements.

Declarative configuration is the most common approach to configuring Infinispan cache instances. In order to read XML configuration files one would typically construct an instance of DefaultCacheManager by pointing to an XML file containing Infinispan configuration. Once the configuration file is read you can obtain reference to the default cache instance.

EmbeddedCacheManager manager = new DefaultCacheManager("my-config-file.xml");
Cache defaultCache = manager.getCache();

or any other named instance specified in my-config-file.xml.

Cache someNamedCache = manager.getCache("someNamedCache");

The name of the default cache is defined in the <cache-container> element of the XML configuration file, and additional caches can be configured using the <local-cache>,<distributed-cache>,<invalidation-cache> or <replicated-cache> elements.

The following example shows the simplest possible configuration for each of the cache types supported by Infinispan:

   <cache-container default-cache="local">
      <transport cluster="mycluster"/>
      <local-cache name="local"/>
      <invalidation-cache name="invalidation" mode="SYNC"/>
      <replicated-cache name="repl-sync" mode="SYNC"/>
      <distributed-cache name="dist-sync" mode="SYNC"/>
Cache configuration templates

As mentioned above, Infinispan supports the notion of configuration templates. These are full or partial configuration declarations which can be shared among multiple caches or as the basis for more complex configurations.

The following example shows how a configuration named local-template is used to define a cache named local.

   <cache-container default-cache="local">
      <!-- template configurations -->
      <local-cache-configuration name="local-template">
         <expiration interval="10000" lifespan="10" max-idle="10"/>

      <!-- cache definitions -->
      <local-cache name="local" configuration="local-template" />

Templates can inherit from previously defined templates, augmenting and/or overriding some or all of the configuration elements:

   <cache-container default-cache="local">
      <!-- template configurations -->
      <local-cache-configuration name="base-template">
         <expiration interval="10000" lifespan="10" max-idle="10"/>

      <local-cache-configuration name="extended-template" configuration="base-template">
         <expiration lifespan="20"/>
            <object size="2000"/>

      <!-- cache definitions -->
      <local-cache name="local" configuration="base-template" />
      <local-cache name="local-bounded" configuration="extended-template" />

In the above example, base-template defines a local cache with a specific expiration configuration. The extended-template configuration inherits from base-template, overriding just a single parameter of the expiration element (all other attributes are inherited) and adds a memory element. Finally, two caches are defined: local which uses the base-template configuration and local-bounded which uses the extended-template configuration.

Be aware that for multi-valued elements (such as properties) the inheritance is additive, i.e. the child configuration will be the result of merging the properties from the parent and its own.
Cache configuration wildcards

An alternative way to apply templates to caches is to use wildcards in the template name, e.g. basecache*. Any cache whose name matches the template wildcard will inherit that configuration.

        <local-cache-configuration name="basecache*">
            <expiration interval="10500" lifespan="11" max-idle="11"/>
        <local-cache name="basecache-1"/>
        <local-cache name="basecache-2"/>

Above, caches basecache-1 and basecache-2 will use the basecache* configuration. The configuration will also be applied when retrieving undefined caches programmatically.

If a cache name matches multiple wildcards, i.e. it is ambiguous, an exception will be thrown.
Declarative configuration reference

For more details on the declarative configuration schema, refer to the configuration reference. If you are using XML editing tools for configuration writing you can use the provided Infinispan schema to assist you.

2.1.2. Configuring caches programmatically

Programmatic Infinispan configuration is centered around the CacheManager and ConfigurationBuilder API. Although every single aspect of Infinispan configuration could be set programmatically, the most usual approach is to create a starting point in a form of XML configuration file and then in runtime, if needed, programmatically tune a specific configuration to suit the use case best.

EmbeddedCacheManager manager = new DefaultCacheManager("my-config-file.xml");
Cache defaultCache = manager.getCache();

Let’s assume that a new synchronously replicated cache is to be configured programmatically. First, a fresh instance of Configuration object is created using ConfigurationBuilder helper object, and the cache mode is set to synchronous replication. Finally, the configuration is defined/registered with a manager.

Configuration c = new ConfigurationBuilder().clustering().cacheMode(CacheMode.REPL_SYNC).build();

String newCacheName = "repl";
manager.defineConfiguration(newCacheName, c);
Cache<String, String> cache = manager.getCache(newCacheName);

The default cache configuration (or any other cache configuration) can be used as a starting point for creation of a new cache. For example, lets say that infinispan-config-file.xml specifies a replicated cache as a default and that a distributed cache is desired with a specific L1 lifespan while at the same time retaining all other aspects of a default cache. Therefore, the starting point would be to read an instance of a default Configuration object and use ConfigurationBuilder to construct and modify cache mode and L1 lifespan on a new Configuration object. As a final step the configuration is defined/registered with a manager.

EmbeddedCacheManager manager = new DefaultCacheManager("infinispan-config-file.xml");
Configuration dcc = manager.getDefaultCacheConfiguration();
Configuration c = new ConfigurationBuilder().read(dcc).clustering().cacheMode(CacheMode.DIST_SYNC).l1().lifespan(60000L).build();
String newCacheName = "distributedWithL1";
manager.defineConfiguration(newCacheName, c);
Cache<String, String> cache = manager.getCache(newCacheName);

As long as the base configuration is the default named cache, the previous code works perfectly fine. However, other times the base configuration might be another named cache. So, how can new configurations be defined based on other defined caches? Take the previous example and imagine that instead of taking the default cache as base, a named cache called "replicatedCache" is used as base. The code would look something like this:

EmbeddedCacheManager manager = new DefaultCacheManager("infinispan-config-file.xml");
Configuration rc = manager.getCacheConfiguration("replicatedCache");
Configuration c = new ConfigurationBuilder().read(rc).clustering().cacheMode(CacheMode.DIST_SYNC).l1().lifespan(60000L).build();
String newCacheName = "distributedWithL1";
manager.defineConfiguration(newCacheName, c);
Cache<String, String> cache = manager.getCache(newCacheName);

Refer to CacheManager , ConfigurationBuilder , Configuration , and GlobalConfiguration javadocs for more details.

ConfigurationBuilder Programmatic Configuration API

While the above paragraph shows how to combine declarative and programmatic configuration, starting from an XML configuration is completely optional. The ConfigurationBuilder fluent interface style allows for easier to write and more readable programmatic configuration. This approach can be used for both the global and the cache level configuration. GlobalConfiguration objects are constructed using GlobalConfigurationBuilder while Configuration objects are built using ConfigurationBuilder. Let’s look at some examples on configuring both global and cache level options with this API:

Configuring the transport

One of the most commonly configured global option is the transport layer, where you indicate how an Infinispan node will discover the others:

GlobalConfiguration globalConfig = new GlobalConfigurationBuilder().transport()
        .addProperty("configurationFile", "jgroups-tcp.xml")
Using a custom JChannel

If you want to construct the JGroups JChannel by yourself, you can do so.

The JChannel must not be already connected.
GlobalConfigurationBuilder global = new GlobalConfigurationBuilder();
JChannel jchannel = new JChannel();
// Configure the jchannel to your needs.
JGroupsTransport transport = new JGroupsTransport(jchannel);
new DefaultCacheManager(global.build());
Enabling JMX MBeans and statistics

Sometimes you might also want to enable collection of global JMX statistics at cache manager level or get information about the transport. To enable global JMX statistics simply do:

GlobalConfiguration globalConfig = new GlobalConfigurationBuilder()

Please note that by not enabling (or by explicitly disabling) global JMX statistics your are just turning off statistics collection. The corresponding MBean is still registered and can be used to manage the cache manager in general, but the statistics attributes do not return meaningful values.

Further options at the global JMX statistics level allows you to configure the cache manager name which comes handy when you have multiple cache managers running on the same system, or how to locate the JMX MBean Server:

GlobalConfiguration globalConfig = new GlobalConfigurationBuilder()
    .mBeanServerLookup(new JBossMBeanServerLookup())
Configuring the thread pools

Some of the Infinispan features are powered by a group of the thread pool executors which can also be tweaked at this global level. For example:

GlobalConfiguration globalConfig = new GlobalConfigurationBuilder()

You can not only configure global, cache manager level, options, but you can also configure cache level options such as the cluster mode:

Configuration config = new ConfigurationBuilder()

Or you can configure eviction and expiration settings:

Configuration config = new ConfigurationBuilder()
Configuring transactions and locking

An application might also want to interact with an Infinispan cache within the boundaries of JTA and to do that you need to configure the transaction layer and optionally tweak the locking settings. When interacting with transactional caches, you might want to enable recovery to deal with transactions that finished with an heuristic outcome and if you do that, you will often want to enable JMX management and statistics gathering too:

Configuration config = new ConfigurationBuilder()
    .transactionManagerLookup(new GenericTransactionManagerLookup())
Configuring cache stores

Configuring Infinispan with chained cache stores is simple too:

Configuration config = new ConfigurationBuilder()
Advanced programmatic configuration

The fluent configuration can also be used to configure more advanced or exotic options, such as advanced externalizers:

GlobalConfiguration globalConfig = new GlobalConfigurationBuilder()
    .addAdvancedExternalizer(998, new PersonExternalizer())
    .addAdvancedExternalizer(999, new AddressExternalizer())

Or, add custom interceptors:

Configuration config = new ConfigurationBuilder()
    .interceptor(new FirstInterceptor()).position(InterceptorConfiguration.Position.FIRST)
    .interceptor(new LastInterceptor()).position(InterceptorConfiguration.Position.LAST)
    .interceptor(new FixPositionInterceptor()).index(8)
    .interceptor(new AfterInterceptor()).after(NonTransactionalLockingInterceptor.class)
    .interceptor(new BeforeInterceptor()).before(CallInterceptor.class)

For information on the individual configuration options, please check the configuration guide.

2.1.3. Configuration Migration Tools

The configuration format of Infinispan has changed since schema version 6.0 in order to align the embedded schema with the one used by the server. For this reason, when upgrading to schema 7.x or later, you should use the configuration converter included in the all distribution. Simply invoke it from the command-line passing the old configuration file as the first parameter and the name of the converted file as the second parameter.

To convert on Unix/Linux/macOS:

bin/config-converter.sh oldconfig.xml newconfig.xml

on Windows:

bin\config-converter.bat oldconfig.xml newconfig.xml
If you wish to help write conversion tools from other caching systems, please contact infinispan-dev.

2.1.4. Clustered Configuration

Infinispan uses JGroups for network communications when in clustered mode. Infinispan ships with pre-configured JGroups stacks that make it easy for you to jump-start a clustered configuration.

Using an external JGroups file

If you are configuring your cache programmatically, all you need to do is:

GlobalConfiguration gc = new GlobalConfigurationBuilder()
   .addProperty("configurationFile", "jgroups.xml")

and if you happen to use an XML file to configure Infinispan, just use:

     <stack-file name="external-file" path="jgroups.xml"/>
  <cache-container default-cache="replicatedCache">
    <transport stack="external-file" />
    <replicated-cache name="replicatedCache"/>



In both cases above, Infinispan looks for jgroups.xml first in your classpath, and then for an absolute path name if not found in the classpath.

Use one of the pre-configured JGroups files

Infinispan ships with a few different JGroups files (packaged in infinispan-core.jar) which means they will already be on your classpath by default. All you need to do is specify the file name, e.g., instead of jgroups.xml above, specify /default-configs/default-jgroups-tcp.xml.

The configurations available are:

  • default-jgroups-udp.xml - Uses UDP as a transport, and UDP multicast for discovery. Usually suitable for larger (over 100 nodes) clusters or if you are using replication or invalidation. Minimises opening too many sockets.

  • default-jgroups-tcp.xml - Uses TCP as a transport and UDP multicast for discovery. Better for smaller clusters (under 100 nodes) only if you are using distribution, as TCP is more efficient as a point-to-point protocol

  • default-jgroups-ec2.xml - Uses TCP as a transport and S3_PING for discovery. Suitable on Amazon EC2 nodes where UDP multicast isn’t available.

  • default-jgroups-kubernetes.xml - Uses TCP as a transport and KUBE_PING for discovery. Suitable on Kubernetes and OpenShift nodes where UDP multicast is not always available.

Tuning JGroups settings

The settings above can be further tuned without editing the XML files themselves. Passing in certain system properties to your JVM at startup can affect the behaviour of some of these settings. The table below shows you which settings can be configured in this way. E.g.,

$ java -cp ... -Djgroups.tcp.port=1234 -Djgroups.tcp.address=
Table 1. default-jgroups-udp.xml

System Property





IP address to use for multicast (both for communications and discovery). Must be a valid Class D IP address, suitable for IP multicast.



Port to use for multicast socket




Specifies the time-to-live (TTL) for IP multicast packets. The value here refers to the number of network hops a packet is allowed to make before it is dropped



Table 2. default-jgroups-tcp.xml

System Property





IP address to use for the TCP transport.



Port to use for TCP socket




IP address to use for multicast (for discovery). Must be a valid Class D IP address, suitable for IP multicast.



Port to use for multicast socket




Specifies the time-to-live (TTL) for IP multicast packets. The value here refers to the number of network hops a packet is allowed to make before it is dropped



Table 3. default-jgroups-ec2.xml

System Property





IP address to use for the TCP transport.



Port to use for TCP socket




The Amazon S3 access key used to access an S3 bucket




The Amazon S3 secret key used to access an S3 bucket




Name of the Amazon S3 bucket to use. Must be unique and must already exist



Table 4. default-jgroups-kubernetes.xml

System Property





IP address to use for the TCP transport.




Port to use for TCP socket



Further reading

JGroups also supports more system property overrides, details of which can be found on this page: SystemProps

In addition, the JGroups configuration files shipped with Infinispan are intended as a jumping off point to getting something up and running, and working. More often than not though, you will want to fine-tune your JGroups stack further to extract every ounce of performance from your network equipment. For this, your next stop should be the JGroups manual which has a detailed section on configuring each of the protocols you see in a JGroups configuration file.

2.2. Obtaining caches

After you configure the CacheManager, you can obtain and control caches.

Invoke the getCache() method to obtain caches, as follows:

Cache<String, String> myCache = manager.getCache("myCache");

The preceding operation creates a cache named myCache, if it does not already exist, and returns it.

Using the getCache() method creates the cache only on the node where you invoke the method. In other words, it performs a local operation that must be invoked on each node across the cluster. Typically, applications deployed across multiple nodes obtain caches during initialization to ensure that caches are symmetric and exist on each node.

Invoke the createCache() method to create caches dynamically across the entire cluster, as follows:

Cache<String, String> myCache = manager.administration().createCache("myCache", "myTemplate");

The preceding operation also automatically creates caches on any nodes that subsequently join the cluster.

Caches that you create with the createCache() method are ephemeral by default. If the entire cluster shuts down, the cache is not automatically created again when it restarts.

Use the PERMANENT flag to ensure that caches can survive restarts, as follows:

Cache<String, String> myCache = manager.administration().withFlags(AdminFlag.PERMANENT).createCache("myCache", "myTemplate");

For the PERMANENT flag to take effect, you must enable global state and set a configuration storage provider.

For more information about configuration storage providers, see GlobalStateConfigurationBuilder#configurationStorage().

2.3. Clustering Information

The EmbeddedCacheManager has quite a few methods to provide information as to how the cluster is operating. The following methods only really make sense when being used in a clustered environment (that is when a Transport is configured).

2.3.1. Member Information

When you are using a cluster it is very important to be able to find information about membership in the cluster including who is the owner of the cluster.

The getMembers() method returns all of the nodes in the current cluster.

The getCoordinator() method will tell you which one of the members is the coordinator of the cluster. For most intents you shouldn’t need to care who the coordinator is. You can use isCoordinator() method directly to see if the local node is the coordinator as well.

2.3.2. Other methods

This method provides you access to the underlying Transport that is used to send messages to other nodes. In most cases a user wouldn’t ever need to go to this level, but if you want to get Transport specific information (in this case JGroups) you can use this mechanism.

The stats provided here are coalesced from all of the active caches in this manager. These stats can be useful to see if there is something wrong going on with your cluster overall.

3. The Cache API

3.1. The Cache interface

Infinispan’s Caches are manipulated through the Cache interface.

A Cache exposes simple methods for adding, retrieving and removing entries, including atomic mechanisms exposed by the JDK’s ConcurrentMap interface. Based on the cache mode used, invoking these methods will trigger a number of things to happen, potentially even including replicating an entry to a remote node or looking up an entry from a remote node, or potentially a cache store.

For simple usage, using the Cache API should be no different from using the JDK Map API, and hence migrating from simple in-memory caches based on a Map to Infinispan’s Cache should be trivial.

3.1.1. Performance Concerns of Certain Map Methods

Certain methods exposed in Map have certain performance consequences when used with Infinispan, such as size() , values() , keySet() and entrySet() . Specific methods on the keySet, values and entrySet are fine for use please see their Javadoc for further details.

Attempting to perform these operations globally would have large performance impact as well as become a scalability bottleneck. As such, these methods should only be used for informational or debugging purposes only.

It should be noted that using certain flags with the withFlags method can mitigate some of these concerns, please check each method’s documentation for more details.

For more performance tips, have a look at our Performance Guide.

3.1.2. Mortal and Immortal Data

Further to simply storing entries, Infinispan’s cache API allows you to attach mortality information to data. For example, simply using put(key, value) would create an immortal entry, i.e., an entry that lives in the cache forever, until it is removed (or evicted from memory to prevent running out of memory). If, however, you put data in the cache using put(key, value, lifespan, timeunit) , this creates a mortal entry, i.e., an entry that has a fixed lifespan and expires after that lifespan.

In addition to lifespan , Infinispan also supports maxIdle as an additional metric with which to determine expiration. Any combination of lifespans or maxIdles can be used.

3.1.3. Expiration and Mortal Data

See Expiration for more information about using mortal data with Infinispan.

3.1.4. putForExternalRead operation

Infinispan’s Cache class contains a different 'put' operation called putForExternalRead . This operation is particularly useful when Infinispan is used as a temporary cache for data that is persisted elsewhere. Under heavy read scenarios, contention in the cache should not delay the real transactions at hand, since caching should just be an optimization and not something that gets in the way.

To achieve this, putForExternalRead acts as a put call that only operates if the key is not present in the cache, and fails fast and silently if another thread is trying to store the same key at the same time. In this particular scenario, caching data is a way to optimise the system and it’s not desirable that a failure in caching affects the on-going transaction, hence why failure is handled differently. putForExternalRead is considered to be a fast operation because regardless of whether it’s successful or not, it doesn’t wait for any locks, and so returns to the caller promptly.

To understand how to use this operation, let’s look at basic example. Imagine a cache of Person instances, each keyed by a PersonId , whose data originates in a separate data store. The following code shows the most common pattern of using putForExternalRead within the context of this example:

// Id of the person to look up, provided by the application
PersonId id = ...;

// Get a reference to the cache where person instances will be stored
Cache<PersonId, Person> cache = ...;

// First, check whether the cache contains the person instance
// associated with with the given id
Person cachedPerson = cache.get(id);

if (cachedPerson == null) {
   // The person is not cached yet, so query the data store with the id
   Person person = dataStore.lookup(id);

   // Cache the person along with the id so that future requests can
   // retrieve it from memory rather than going to the data store
   cache.putForExternalRead(id, person);
} else {
   // The person was found in the cache, so return it to the application
   return cachedPerson;

Please note that putForExternalRead should never be used as a mechanism to update the cache with a new Person instance originating from application execution (i.e. from a transaction that modifies a Person’s address). When updating cached values, please use the standard put operation, otherwise the possibility of caching corrupt data is likely.

3.2. The AdvancedCache interface

In addition to the simple Cache interface, Infinispan offers an AdvancedCache interface, geared towards extension authors. The AdvancedCache offers the ability to inject custom interceptors, access certain internal components and to apply flags to alter the default behavior of certain cache methods. The following code snippet depicts how an AdvancedCache can be obtained:

AdvancedCache advancedCache = cache.getAdvancedCache();

3.2.1. Flags

Flags are applied to regular cache methods to alter the behavior of certain methods. For a list of all available flags, and their effects, see the Flag enumeration. Flags are applied using AdvancedCache.withFlags() . This builder method can be used to apply any number of flags to a cache invocation, for example:

advancedCache.withFlags(Flag.CACHE_MODE_LOCAL, Flag.SKIP_LOCKING)
   .put("hello", "world");

3.2.2. Custom Interceptors

The AdvancedCache interface also offers advanced developers a mechanism with which to attach custom interceptors. Custom interceptors allow developers to alter the behavior of the cache API methods, and the AdvancedCache interface allows developers to attach these interceptors programmatically, at run-time. See the AdvancedCache Javadocs for more details.

For more information on writing custom interceptors, see Custom Interceptors.

3.3. Listeners and Notifications

Infinispan offers a listener API, where clients can register for and get notified when events take place. This annotation-driven API applies to 2 different levels: cache level events and cache manager level events.

Events trigger a notification which is dispatched to listeners. Listeners are simple POJO s annotated with @Listener and registered using the methods defined in the Listenable interface.

Both Cache and CacheManager implement Listenable, which means you can attach listeners to either a cache or a cache manager, to receive either cache-level or cache manager-level notifications.

For example, the following class defines a listener to print out some information every time a new entry is added to the cache:

public class PrintWhenAdded {

  public void print(CacheEntryCreatedEvent event) {
    System.out.println("New entry " + event.getKey() + " created in the cache");


For more comprehensive examples, please see the Javadocs for @Listener.

3.3.1. Cache-level notifications

Cache-level events occur on a per-cache basis, and by default are only raised on nodes where the events occur. Note in a distributed cache these events are only raised on the owners of data being affected. Examples of cache-level events are entries being added, removed, modified, etc. These events trigger notifications to listeners registered to a specific cache.

Please see the Javadocs on the org.infinispan.notifications.cachelistener.annotation package for a comprehensive list of all cache-level notifications, and their respective method-level annotations.

Please refer to the Javadocs on the org.infinispan.notifications.cachelistener.annotation package for the list of cache-level notifications available in Infinispan.
Cluster Listeners

The cluster listeners should be used when it is desirable to listen to the cache events on a single node.

To do so all that is required is set to annotate your listener as being clustered.

@Listener (clustered = true)
public class MyClusterListener { .... }

There are some limitations to cluster listeners from a non clustered listener.

  1. A cluster listener can only listen to @CacheEntryModified, @CacheEntryCreated, @CacheEntryRemoved and @CacheEntryExpired events. Note this means any other type of event will not be listened to for this listener.

  2. Only the post event is sent to a cluster listener, the pre event is ignored.

Event filtering and conversion

All applicable events on the node where the listener is installed will be raised to the listener. It is possible to dynamically filter what events are raised by using a KeyFilter (only allows filtering on keys) or CacheEventFilter (used to filter for keys, old value, old metadata, new value, new metadata, whether command was retried, if the event is before the event (ie. isPre) and also the command type).

The example here shows a simple KeyFilter that will only allow events to be raised when an event modified the entry for the key Only Me.

public class SpecificKeyFilter implements KeyFilter<String> {
    private final String keyToAccept;

    public SpecificKeyFilter(String keyToAccept) {
      if (keyToAccept == null) {
        throw new NullPointerException();
      this.keyToAccept = keyToAccept;

    boolean accept(String key) {
      return keyToAccept.equals(key);

cache.addListener(listener, new SpecificKeyFilter("Only Me"));

This can be useful when you want to limit what events you receive in a more efficient manner.

There is also a CacheEventConverter that can be supplied that allows for converting a value to another before raising the event. This can be nice to modularize any code that does value conversions.

The mentioned filters and converters are especially beneficial when used in conjunction with a Cluster Listener. This is because the filtering and conversion is done on the node where the event originated and not on the node where event is listened to. This can provide benefits of not having to replicate events across the cluster (filter) or even have reduced payloads (converter).
Initial State Events

When a listener is installed it will only be notified of events after it is fully installed.

It may be desirable to get the current state of the cache contents upon first registration of listener by having an event generated of type @CacheEntryCreated for each element in the cache. Any additionally generated events during this initial phase will be queued until appropriate events have been raised.

This only works for clustered listeners at this time. ISPN-4608 covers adding this for non clustered listeners.
Duplicate Events

It is possible in a non transactional cache to receive duplicate events. This is possible when the primary owner of a key goes down while trying to perform a write operation such as a put.

Infinispan internally will rectify the put operation by sending it to the new primary owner for the given key automatically, however there are no guarantees in regards to if the write was first replicated to backups. Thus more than 1 of the following write events (CacheEntryCreatedEvent, CacheEntryModifiedEvent & CacheEntryRemovedEvent) may be sent on a single operation.

If more than one event is generated Infinispan will mark the event that it was generated by a retried command to help the user to know when this occurs without having to pay attention to view changes.

public class MyRetryListener {
  public void entryModified(CacheEntryModifiedEvent event) {
    if (event.isCommandRetried()) {
      // Do something

Also when using a CacheEventFilter or CacheEventConverter the EventType contains a method isRetry to tell if the event was generated due to retry.

3.3.2. Cache manager-level notifications

Cache manager-level events occur on a cache manager. These too are global and cluster-wide, but involve events that affect all caches created by a single cache manager. Examples of cache manager-level events are nodes joining or leaving a cluster, or caches starting or stopping.

Please see the Javadocs on the org.infinispan.notifications.cachemanagerlistener.annotation package for a comprehensive list of all cache manager-level notifications, and their respective method-level annotations.

3.3.3. Synchronicity of events

By default, all notifications are dispatched in the same thread that generates the event. This means that you must write your listener such that it does not block or do anything that takes too long, as it would prevent the thread from progressing. Alternatively, you could annotate your listener as asynchronous , in which case a separate thread pool will be used to dispatch the notification and prevent blocking the event originating thread. To do this, simply annotate your listener such:

@Listener (sync = false)
public class MyAsyncListener { .... }
Asynchronous thread pool

To tune the thread pool used to dispatch such asynchronous notifications, use the <listener-executor /> XML element in your configuration file.

3.4. Asynchronous API

In addition to synchronous API methods like Cache.put() , Cache.remove() , etc., Infinispan also has an asynchronous, non-blocking API where you can achieve the same results in a non-blocking fashion.

These methods are named in a similar fashion to their blocking counterparts, with "Async" appended.  E.g., Cache.putAsync() , Cache.removeAsync() , etc.  These asynchronous counterparts return a Future containing the actual result of the operation.

For example, in a cache parameterized as Cache<String, String>, Cache.put(String key, String value) returns a String. Cache.putAsync(String key, String value) would return a Future<String>.

3.4.1. Why use such an API?

Non-blocking APIs are powerful in that they provide all of the guarantees of synchronous communications - with the ability to handle communication failures and exceptions - with the ease of not having to block until a call completes.  This allows you to better harness parallelism in your system.  For example:

Set<Future<?>> futures = new HashSet<Future<?>>();
futures.add(cache.putAsync(key1, value1)); // does not block
futures.add(cache.putAsync(key2, value2)); // does not block
futures.add(cache.putAsync(key3, value3)); // does not block

// the remote calls for the 3 puts will effectively be executed
// in parallel, particularly useful if running in distributed mode
// and the 3 keys would typically be pushed to 3 different nodes
// in the cluster

// check that the puts completed successfully
for (Future<?> f: futures) f.get();

3.4.2. Which processes actually happen asynchronously?

There are 4 things in Infinispan that can be considered to be on the critical path of a typical write operation. These are, in order of cost:

  • network calls

  • marshalling

  • writing to a cache store (optional)

  • locking

As of Infinispan 4.0, using the async methods will take the network calls and marshalling off the critical path.  For various technical reasons, writing to a cache store and acquiring locks, however, still happens in the caller’s thread.  In future, we plan to take these offline as well.  See this developer mail list thread about this topic.

3.4.3. Notifying futures

Strictly, these methods do not return JDK Futures, but rather a sub-interface known as a NotifyingFuture .  The main difference is that you can attach a listener to a NotifyingFuture such that you could be notified when the future completes.  Here is an example of making use of a notifying future:

FutureListener futureListener = new FutureListener() {

   public void futureDone(Future future) {
      try {
      } catch (Exception e) {
         // Future did not complete successfully
cache.putAsync("key", "value").attachListener(futureListener);

3.4.4. Further reading

The Javadocs on the Cache interface has some examples on using the asynchronous API, as does this article by Manik Surtani introducing the API.

3.5. Invocation Flags

An important aspect of getting the most of Infinispan is the use of per-invocation flags in order to provide specific behaviour to each particular cache call. By doing this, some important optimizations can be implemented potentially saving precious time and network resources. One of the most popular usages of flags can be found right in Cache API, underneath the putForExternalRead() method which is used to load an Infinispan cache with data read from an external resource. In order to make this call efficient, Infinispan basically calls a normal put operation passing the following flags: FAIL_SILENTLY , FORCE_ASYNCHRONOUS , ZERO_LOCK_ACQUISITION_TIMEOUT

What Infinispan is doing here is effectively saying that when putting data read from external read, it will use an almost-zero lock acquisition time and that if the locks cannot be acquired, it will fail silently without throwing any exception related to lock acquisition. It also specifies that regardless of the cache mode, if the cache is clustered, it will replicate asynchronously and so won’t wait for responses from other nodes. The combination of all these flags make this kind of operation very efficient, and the efficiency comes from the fact this type of putForExternalRead calls are used with the knowledge that client can always head back to a persistent store of some sorts to retrieve the data that should be stored in memory. So, any attempt to store the data is just a best effort and if not possible, the client should try again if there’s a cache miss.

3.5.1. Examples

If you want to use these or any other flags available, which by the way are described in detail the Flag enumeration , you simply need to get hold of the advanced cache and add the flags you need via the withFlags() method call. For example:

Cache cache = ...
   .put("local", "only");

It’s worth noting that these flags are only active for the duration of the cache operation. If the same flags need to be used in several invocations, even if they’re in the same transaction, withFlags() needs to be called repeatedly. Clearly, if the cache operation is to be replicated in another node, the flags are carried over to the remote nodes as well.

Suppressing return values from a put() or remove()

Another very important use case is when you want a write operation such as put() to not return the previous value. To do that, you need to use two flags to make sure that in a distributed environment, no remote lookup is done to potentially get previous value, and if the cache is configured with a cache loader, to avoid loading the previous value from the cache store. You can see these two flags in action in the following example:

Cache cache = ...
   .put("local", "only")

For more information, please check the Flag enumeration javadoc.

3.6. Tree API Module

Infinispan’s tree API module offers clients the possibility of storing data using a tree-structure like API. This API is similar to the one provided by JBoss Cache, hence the tree module is perfect for those users wanting to migrate their applications from JBoss Cache to Infinispan, who want to limit changes their codebase as part of the migration. Besides, it’s important to understand that Infinispan provides this tree API much more efficiently than JBoss Cache did, so if you’re a user of the tree API in JBoss Cache, you should consider migrating to Infinispan.

3.6.1. What is Tree API about?

The aim of this API is to store information in a hierarchical way. The hierarchy is defined using paths represented as Fqn or fully qualified names , for example: /this/is/a/fqn/path or /another/path . In the hierarchy, there’s a special path called root which represents the starting point of all paths and it’s represented as: /

Each FQN path is represented as a node where users can store data using a key/value pair style API (i.e. a Map). For example, in /persons/john , you could store information belonging to John, for example: surname=Smith, birthdate=05/02/1980…​etc.

Please remember that users should not use root as a place to store data. Instead, users should define their own paths and store data there. The following sections will delve into the practical aspects of this API.

3.6.2. Using the Tree API


For your application to use the tree API, you need to import infinispan-tree.jar which can be located in the Infinispan binary distributions, or you can simply add a dependency to this module in your pom.xml:


Replace ${version.infinispan} with the appropriate version of Infinispan.

3.6.3. Creating a Tree Cache

The first step to use the tree API is to actually create a tree cache. To do so, you need to create an Infinispan Cache as you’d normally do, and using the TreeCacheFactory , create an instance of TreeCache . A very important note to remember here is that the Cache instance passed to the factory must be configured with invocation batching. For example:

import org.infinispan.config.Configuration;
import org.infinispan.tree.TreeCacheFactory;
import org.infinispan.tree.TreeCache;
Configuration config = new Configuration();
Cache cache = new DefaultCacheManager(config).getCache();
TreeCache treeCache = TreeCacheFactory.createTreeCache(cache);

3.6.4. Manipulating data in a Tree Cache

The Tree API effectively provides two ways to interact with the data:

Via TreeCache convenience methods: These methods are located within the TreeCache interface and enable users to store , retrieve , move , remove …​etc data with a single call that takes the Fqn , in String or Fqn format, and the data involved in the call. For example:

treeCache.put("/persons/john", "surname", "Smith");


import org.infinispan.tree.Fqn;
Fqn johnFqn = Fqn.fromString("persons/john");
Calendar calendar = Calendar.getInstance();
calendar.set(1980, 5, 2);
treeCache.put(johnFqn, "birthdate", calendar.getTime()));

Via Node API: It allows finer control over the individual nodes that form the FQN, allowing manipulation of nodes relative to a particular node. For example:

import org.infinispan.tree.Node;
TreeCache treeCache = ...
Fqn johnFqn = Fqn.fromElements("persons", "john");
Node<String, Object> john = treeCache.getRoot().addChild(johnFqn);
john.put("surname", "Smith");


Node persons = treeCache.getRoot().addChild(Fqn.fromString("persons"));
Node<String, Object> john = persons.addChild(Fqn.fromString("john"));
john.put("surname", "Smith");

Or even:

Fqn personsFqn = Fqn.fromString("persons");
Fqn johnFqn = Fqn.fromRelative(personsFqn, Fqn.fromString("john"));
Node<String, Object> john = treeCache.getRoot().addChild(johnFqn);
john.put("surname", "Smith");

A node also provides the ability to access its parent or children . For example:

Node<String, Object> john = ...
Node persons = john.getParent();


Set<Node<String, Object>> personsChildren = persons.getChildren();

3.6.5. Common Operations

In the previous section, some of the most used operations, such as addition and retrieval, have been shown. However, there are other important operations that are worth mentioning, such as remove:

You can for example remove an entire node, i.e. /persons/john , using:


Or remove a child node, i.e. persons that a child of root, via:


You can also remove a particular key/value pair in a node:

Node john = treeCache.getRoot().getChild(Fqn.fromElements("persons", "john"));

Or you can remove all data in a node with:

Node john = treeCache.getRoot().getChild(Fqn.fromElements("persons", "john"));

Another important operation supported by Tree API is the ability to move nodes around in the tree. Imagine we have a node called "john" which is located under root node. The following example is going to show how to we can move "john" node to be under "persons" node:

Current tree structure:


Moving trees from one FQN to another:

Node john = treeCache.getRoot().addChild(Fqn.fromString("john"));
Node persons = treeCache.getRoot().getChild(Fqn.fromString("persons"));
treeCache.move(john.getFqn(), persons.getFqn());

Final tree structure:


3.6.6. Locking in the Tree API

Understanding when and how locks are acquired when manipulating the tree structure is important in order to maximise the performance of any client application interacting against the tree, while at the same time maintaining consistency.

Locking on the tree API happens on a per node basis. So, if you’re putting or updating a key/value under a particular node, a write lock is acquired for that node. In such case, no write locks are acquired for parent node of the node being modified, and no locks are acquired for children nodes.

If you’re adding or removing a node, the parent is not locked for writing. In JBoss Cache, this behaviour was configurable with the default being that parent was not locked for insertion or removal.

Finally, when a node is moved, the node that’s been moved and any of its children are locked, but also the target node and the new location of the moved node and its children. To understand this better, let’s look at an example:

Imagine you have a hierarchy like this and we want to move c/ to be underneath b/:

     /     \
     a     c
     |     |
     b     e

The end result would be something like this:

     /     \
     d     c

To make this move, locks would have been acquired on:

  • /a/b - because it’s the parent underneath which the data will be put

  • /c and /c/e - because they’re the nodes that are being moved

  • /a/b/c and /a/b/c/e - because that’s new target location for the nodes being moved

3.6.7. Listeners for tree cache events

The current Infinispan listeners have been designed with key/value store notifications in mind, and hence they do not map to tree cache events correctly. Tree cache specific listeners that map directly to tree cache events (i.e. adding a child…​etc) are desirable but these are not yet available. If you’re interested in this type of listeners, please follow this issue to find out about any progress in this area.

3.7. Functional Map API

Infinispan includes an experimental API for interacting with your data which takes advantage of the functional programming additions and improved asynchronous programming capabilities available in Java 8.

Infinispan’s Functional Map API is a distilled map-like asynchronous API which uses functions to interact with data.

3.7.1. Asynchronous and Lazy

Being an asynchronous API, all methods that return a single result, return a CompletableFuture which wraps the result, so you can use the resources of your system more efficiently by having the possibility to receive callbacks when the CompletableFuture has completed, or you can chain or compose them with other CompletableFuture.

For those operations that return multiple results, the API returns instances of a ​Traversable interface which offers a lazy pull-style API for working with multiple results. ​Traversable ,​ being a lazy pull-style API, can still be asynchronous underneath since the user can decide to work on the traversable at a later stage, and the ​Traversable implementation itself can decide when to compute those results.

3.7.2. Function transparency

Since the content of the functions is transparent to Infinispan, the API has been split into 3 interfaces for read­-only ( R​eadOnlyMap )​, read­-write ( R​eadWriteMap )​ and write­-only ( W​riteOnlyMap )​ operations respectively, in order to provide hints to the Infinispan internals on the type of work needed to support functions.

3.7.3. Constructing Functional Maps

To construct any of the read-only, write-only or read-write map instances, an Infinispan AdvancedCache is required, which is retrieved from the Cache Manager, and using the AdvancedCache , static method factory methods are used to create R​eadOnlyMap , R​eadWriteMap or W​riteOnlyMap

import org.infinispan.commons.api.functional.FunctionalMap.*;
import org.infinispan.functional.impl.*;

AdvancedCache<String, String> cache = ...

FunctionalMapImpl<String, String> functionalMap = FunctionalMapImpl.create(cache);
ReadOnlyMap<String, String> readOnlyMap = ReadOnlyMapImpl.create(functionalMap);
WriteOnlyMap<String, String> writeOnlyMap = WriteOnlyMapImpl.create(functionalMap);
ReadWriteMap<String, String> readWriteMap = ReadWriteMapImpl.create(functionalMap);
At this stage, the Functional Map API is experimental and hence the way FunctionalMap, ReadOnlyMap, WriteOnlyMap and ReadWriteMap are constructed is temporary.

3.7.4. Read-Only Map API

Read-only operations have the advantage that no locks are acquired for the duration of the operation. Here’s an example on how to the equivalent operation for Map.get(K):

import org.infinispan.commons.api.functional.EntryView.*;
import org.infinispan.commons.api.functional.FunctionalMap.*;

ReadOnlyMap<String, String> readOnlyMap = ...
CompletableFuture<Optional<String>> readFuture = readOnlyMap.eval("key1", ReadEntryView::find);

Read-only map also exposes operations to retrieve multiple keys in one go:

import org.infinispan.commons.api.functional.EntryView.*;
import org.infinispan.commons.api.functional.FunctionalMap.*;
import org.infinispan.commons.api.functional.Traversable;

ReadOnlyMap<String, String> readOnlyMap = ...

Set<String> keys = new HashSet<>(Arrays.asList("key1", "key2"));
Traversable<String> values = readOnlyMap.evalMany(keys, ReadEntryView::get);

Finally, read-only map also exposes methods to read all existing keys as well as entries, which include both key and value information.

Read-Only Entry View

The function parameters for read-only maps provide the user with a read-only entry view to interact with the data in the cache, which include these operations:

  • key() method returns the key for which this function is being executed.

  • find() returns a Java 8 Optional wrapping the value if present, otherwise it returns an empty optional. Unless the value is guaranteed to be associated with the key, it’s recommended to use find() to verify whether there’s a value associated with the key.

  • get() returns the value associated with the key. If the key has no value associated with it, calling get() throws a NoSuchElementException. get() can be considered as a shortcut of ReadEntryView.find().get() which should be used only when the caller has guarantees that there’s definitely a value associated with the key.

  • findMetaParam(Class<T> type) allows metadata parameter information associated with the cache entry to be looked up, for example: entry lifespan, last accessed time…​etc. See Metadata Parameter Handling to find out more.

3.7.5. Write-Only Map API

Write-only operations include operations that insert or update data in the cache and also removals. Crucially, a write-only operation does not attempt to read any previous value associated with the key. This is an important optimization since that means neither the cluster nor any persistence stores will be looked up to retrieve previous values. In the main Infinispan Cache, this kind of optimization was achieved using a local-only per-invocation flag, but the use case is so common that in this new functional API, this optimization is provided as a first-class citizen.

Using write-only map API , an operation equivalent to javax.cache.Cache (JCache) 's void returning put can be achieved this way, followed by an attempt to read the stored value using the read-only map API:

import org.infinispan.commons.api.functional.EntryView.*;
import org.infinispan.commons.api.functional.FunctionalMap.*;

WriteOnlyMap<String, String> writeOnlyMap = ...
ReadOnlyMap<String, String> readOnlyMap = ...

CompletableFuture<Void> writeFuture = writeOnlyMap.eval("key1", "value1",
   (v, view) -> view.set(v));
CompletableFuture<String> readFuture = writeFuture.thenCompose(r ->
   readOnlyMap.eval("key1", ReadEntryView::get));

Multiple key/value pairs can be stored in one go using evalMany API:

WriteOnlyMap<String, String> writeOnlyMap = ...

Map<K, String> data = new HashMap<>();
data.put("key1", "value1");
data.put("key2", "value2");
CompletableFuture<Void> writerAllFuture = writeOnlyMap.evalMany(data, (v, view) -> view.set(v));
writerAllFuture.thenAccept(x -> "Write completed");

To remove all contents of the cache, there are two possibilities with different semantics. If using evalAll each cached entry is iterated over and the function is called with that entry’s information. Using this method also results in listeners being invoked. See functional listeners for more information.

WriteOnlyMap<String, String> writeOnlyMap = ...

CompletableFuture<Void> removeAllFuture = writeOnlyMap.evalAll(WriteEntryView::remove);
removeAllFuture.thenAccept(x -> "All entries removed");

The alternative way to remove all entries is to call truncate operation which clears the entire cache contents in one go without invoking any listeners and is best-effort:

WriteOnlyMap<String, String> writeOnlyMap = ...

CompletableFuture<Void> truncateFuture = writeOnlyMap.truncate();
truncateFuture.thenAccept(x -> "Cache contents cleared");
Write-Only Entry View

The function parameters for write-only maps provide the user with a write-only entry view to modify the data in the cache, which include these operations:

  • set(V, MetaParam.Writable…​) method allows for a new value to be associated with the cache entry for which this function is executed, and it optionally takes zero or more metadata parameters to be stored along with the value. See Metadata Parameter Handling for more information.

  • remove() method removes the cache entry, including both value and metadata parameters associated with this key.

3.7.6. Read-Write Map API

The final type of operations we have are read­write operations, and within this category CAS-like (Compare­And­Swap) operations can be found. This type of operations require previous value associated with the key to be read and for locks to be acquired before executing the function. The vast majority of operations within ConcurrentMap and JCache APIs fall within this category, and they can easily be implemented using the read-write map API . Moreover, with read-write map API , you can make CAS­like comparisons not only based on value equality but based on metadata parameter equality such as version information, and you can send back previous value or boolean instances to signal whether the CAS­like comparison succeeded.

Implementing a write operation that returns the previous value associated with the cache entry is easy to achieve with the read-write map API:

import org.infinispan.commons.api.functional.EntryView.*;
import org.infinispan.commons.api.functional.FunctionalMap.*;

ReadWriteMap<String, String> readWriteMap = ...

CompletableFuture<Optional<String>> readWriteFuture = readWriteMap.eval("key1", "value1",
   (v, view) -> {
      Optional<V> prev = rw.find();
      return prev;

ConcurrentMap.replace(K, V, V) is a replace function that compares the value present in the map and if it’s equals to the value passed in as first parameter, the second value is stored, returning a boolean indicating whether the replace was successfully completed. This operation can easily be implemented using the read-write map API:

ReadWriteMap<String, String> readWriteMap = ...

String oldValue = "old-value";
CompletableFuture<Boolean> replaceFuture = readWriteMap.eval("key1", "value1", (v, view) -> {
   return view.find().map(prev -> {
      if (prev.equals(oldValue)) {
         return true; // previous value present and equals to the expected one
      return false; // previous value associated with key does not match
   }).orElse(false); // no value associated with this key
replaceFuture.thenAccept(replaced -> System.out.printf("Value was replaced? %s%n", replaced));
The function in the example above captures oldValue which is an external value to the function which is valid use case.

Read-write map API contains evalMany and evalAll operations which behave similar to the write-only map offerings, except that they enable previous value and metadata parameters to be read.

Read-Write Entry View

The function parameters for read-write maps provide the user with the possibility to query the information associated with the key, including value and metadata parameters, and the user can also use this read-write entry view to modify the data in the cache.

The operations are exposed by read-write entry views are a union of the operations exposed by read-only entry views and write-only entry views.

3.7.7. Metadata Parameter Handling

Metadata parameters provide extra information about the cache entry, such as version information, lifespan, last accessed/used time…​etc. Some of these can be provided by the user, e.g. version, lifespan…​etc, but some others are computed internally and can only be queried, e.g. last accessed/used time.

The functional map API provides a flexible way to store metadata parameters along with an cache entry. To be able to store a metadata parameter, it must extend MetaParam.Writable interface, and implement the methods to allow the internal logic to extra the data. Storing is done via the set(V, MetaParam.Writable…​) method in the write-only entry view or read-write entry view function parameters.

Querying metadata parameters is available via the findMetaParam(Class) method available via read-write entry view or read-only entry views or function parameters.

Here is an example showing how to store metadata parameters and how to query them:

import java.time.Duration;
import org.infinispan.commons.api.functional.EntryView.*;
import org.infinispan.commons.api.functional.FunctionalMap.*;
import org.infinispan.commons.api.functional.MetaParam.*;

WriteOnlyMap<String, String> writeOnlyMap = ...
ReadOnlyMap<String, String> readOnlyMap = ...

CompletableFuture<Void> writeFuture = writeOnlyMap.eval("key1", "value1",
   (v, view) -> view.set(v, new MetaLifespan(Duration.ofHours(1).toMillis())));
CompletableFuture<MetaLifespan> readFuture = writeFuture.thenCompose(r ->
   readOnlyMap.eval("key1", view -> view.findMetaParam(MetaLifespan.class).get()));

If the metadata parameter is generic, for example MetaEntryVersion<T> , retrieving the metadata parameter along with a specific type can be tricky if using .class static helper in a class because it does not return a Class<T> but only Class, and hence any generic information in the class is lost:

ReadOnlyMap<String, String> readOnlyMap = ...

CompletableFuture<String> readFuture = readOnlyMap.eval("key1", view -> {
   // If caller depends on the typed information, this is not an ideal way to retrieve it
   // If the caller does not depend on the specific type, this works just fine.
   Optional<MetaEntryVersion> version = view.findMetaParam(MetaEntryVersion.class);
   return view.get();

When generic information is important the user can define a static helper method that coerces the static class retrieval to the type requested, and then use that helper method in the call to findMetaParam:

class MetaEntryVersion<T> implements MetaParam.Writable<EntryVersion<T>> {
   public static <T> T type() { return (T) MetaEntryVersion.class; }

ReadOnlyMap<String, String> readOnlyMap = ...

CompletableFuture<String> readFuture = readOnlyMap.eval("key1", view -> {
   // The caller wants guarantees that the metadata parameter for version is numeric
   // e.g. to query the actual version information
   Optional<MetaEntryVersion<Long>> version = view.findMetaParam(MetaEntryVersion.type());
   return view.get();

Finally, users are free to create new instances of metadata parameters to suit their needs. They are stored and retrieved in the very same way as done for the metadata parameters already provided by the functional map API.

3.7.8. Invocation Parameter

Per-invocation parameters are applied to regular functional map API calls to alter the behaviour of certain aspects. Adding per invocation parameters is done using the withParams(Param<?>…​) method.

Param.FutureMode tweaks whether a method returning a CompletableFuture will span a thread to invoke the method, or instead will use the caller thread. By default, whenever a call is made to a method returning a CompletableFuture , a separate thread will be span to execute the method asynchronously. However, if the caller will immediately block waiting for the CompletableFuture to complete, spanning a different thread is wasteful, and hence Param.FutureMode.COMPLETED can be passed as per-invocation parameter to avoid creating that extra thread. Example:

import org.infinispan.commons.api.functional.EntryView.*;
import org.infinispan.commons.api.functional.FunctionalMap.*;
import org.infinispan.commons.api.functional.Param.*;

ReadOnlyMap<String, String> readOnlyMap = ...
ReadOnlyMap<String, String> readOnlyMapCompleted = readOnlyMap.withParams(FutureMode.COMPLETED);
Optional<String> readFuture = readOnlyMapCompleted.eval("key1", ReadEntryView::find).get();

Param.PersistenceMode controls whether a write operation will be propagated to a persistence store. The default behaviour is for all write-operations to be propagated to the persistence store if the cache is configured with a persistence store. By passing PersistenceMode.SKIP as parameter, the write operation skips the persistence store and its effects are only seen in the in-memory contents of the cache. PersistenceMode.SKIP can be used to implement an Cache.evict() method which removes data from memory but leaves the persistence store untouched:

import org.infinispan.commons.api.functional.EntryView.*;
import org.infinispan.commons.api.functional.FunctionalMap.*;
import org.infinispan.commons.api.functional.Param.*;

WriteOnlyMap<String, String> writeOnlyMap = ...
WriteOnlyMap<String, String> skiPersistMap = writeOnlyMap.withParams(PersistenceMode.SKIP);
CompletableFuture<Void> removeFuture = skiPersistMap.eval("key1", WriteEntryView::remove);

Note that there’s no need for another PersistenceMode option to skip reading from the persistence store, because a write operation can skip reading previous value from the store by calling a write-only operation via the WriteOnlyMap.

Finally, new Param implementations are normally provided by the functional map API since they tweak how the internal logic works. So, for the most part of users, they should limit themselves to using the Param instances exposed by the API. The exception to this rule would be advanced users who decide to add new interceptors to the internal stack. These users have the ability to query these parameters within the interceptors.

3.7.9. Functional Listeners

The functional map offers a listener API, where clients can register for and get notified when events take place. These notifications are post-event, so that means the events are received after the event has happened.

The listeners that can be registered are split into two categories: write listeners and read-write listeners.

Write Listeners

Write listeners enable user to register listeners for any cache entry write events that happen in either a read-write or write-only functional map.

Listeners for write events cannot distinguish between cache entry created and cache entry modify/update events because they don’t have access to the previous value. All they know is that a new non-null entry has been written.

However, write event listeners can distinguish between entry removals and cache entry create/modify-update events because they can query what the new entry’s value via ReadEntryView.find() method.

Adding a write listener is done via the WriteListeners interface which is accessible via both ReadWriteMap.listeners() and WriteOnlyMap.listeners() method.

A write listener implementation can be defined either passing a function to onWrite(Consumer<ReadEntryView<K, V>>) method, or passing a WriteListener implementation to add(WriteListener<K, V>) method. Either way, all these methods return an AutoCloseable instance that can be used to de-register the function listener:

import org.infinispan.commons.api.functional.EntryView.*;
import org.infinispan.commons.api.functional.FunctionalMap.*;
import org.infinispan.commons.api.functional.Listeners.WriteListeners.WriteListener;

WriteOnlyMap<String, String> woMap = ...

AutoCloseable writeFunctionCloseHandler = woMap.listeners().onWrite(written -> {
   // `written` is a ReadEntryView of the written entry
   System.out.printf("Written: %s%n", written.get());
AutoCloseable writeCloseHanlder = woMap.listeners().add(new WriteListener<String, String>() {
   public void onWrite(ReadEntryView<K, V> written) {
      System.out.printf("Written: %s%n", written.get());

// Either wrap handler in a try section to have it auto close...
try(writeFunctionCloseHandler) {
   // Write entries using read-write or write-only functional map API
// Or close manually
Read-Write Listeners

Read-write listeners enable users to register listeners for cache entry created, modified and removed events, and also register listeners for any cache entry write events.

Entry created, modified and removed events can only be fired when these originate on a read-write functional map, since this is the only one that guarantees that the previous value has been read, and hence the differentiation between create, modified and removed can be fully guaranteed.

Adding a read-write listener is done via the ReadWriteListeners interface which is accessible via ReadWriteMap.listeners() method.

If interested in only one of the event types, the simplest way to add a listener is to pass a function to either onCreate , onModify or onRemove methods. All these methods return an AutoCloseable instance that can be used to de-register the function listener:

import org.infinispan.commons.api.functional.EntryView.*;
import org.infinispan.commons.api.functional.FunctionalMap.*;

ReadWriteMap<String, String> rwMap = ...
AutoCloseable createClose = rwMap.listeners().onCreate(created -> {
   // `created` is a ReadEntryView of the created entry
   System.out.printf("Created: %s%n", created.get());
AutoCloseable modifyClose = rwMap.listeners().onModify((before, after) -> {
   // `before` is a ReadEntryView of the entry before update
   // `after` is a ReadEntryView of the entry after update
   System.out.printf("Before: %s%n", before.get());
   System.out.printf("After: %s%n", after.get());
AutoCloseable removeClose = rwMap.listeners().onRemove(removed -> {
   // `removed` is a ReadEntryView of the removed entry
   System.out.printf("Removed: %s%n", removed.get());
AutoCloseable writeClose = woMap.listeners().onWrite(written -> {
   // `written` is a ReadEntryView of the written entry
   System.out.printf("Written: %s%n", written.get());
// Either wrap handler in a try section to have it auto close...
try(createClose) {
   // Create entries using read-write functional map API
// Or close manually

If listening for two or more event types, it’s better to pass in an implementation of ReadWriteListener interface via the ReadWriteListeners.add() method. ReadWriteListener offers the same onCreate/onModify/onRemove callbacks with default method implementations that are empty:

import org.infinispan.commons.api.functional.EntryView.*;
import org.infinispan.commons.api.functional.FunctionalMap.*;
import org.infinispan.commons.api.functional.Listeners.ReadWriteListeners.ReadWriteListener;

ReadWriteMap<String, String> rwMap = ...
AutoCloseable readWriteClose = rwMap.listeners.add(new ReadWriteListener<String, String>() {
   public void onCreate(ReadEntryView<String, String> created) {
      System.out.printf("Created: %s%n", created.get());

   public void onModify(ReadEntryView<String, String> before, ReadEntryView<String, String> after) {
      System.out.printf("Before: %s%n", before.get());
      System.out.printf("After: %s%n", after.get());

   public void onRemove(ReadEntryView<String, String> removed) {
      System.out.printf("Removed: %s%n", removed.get());
AutoCloseable writeClose = rwMap.listeners.add(new WriteListener<String, String>() {
   public void onWrite(ReadEntryView<K, V> written) {
      System.out.printf("Written: %s%n", written.get());

// Either wrap handler in a try section to have it auto close...
try(readWriteClose) {
   // Create/update/remove entries using read-write functional map API
// Or close manually

3.7.10. Marshalling of Functions

Running functional map in a cluster of nodes involves marshalling and replication of the operation parameters under certain circumstances.

To be more precise, when write operations are executed in a cluster, regardless of read-write or write-only operations, all the parameters to the method and the functions are replicated to other nodes.

There are multiple ways in which a function can be marshalled. The simplest way, which is also the most costly option in terms of payload size, is to mark the function as Serializable :

import org.infinispan.commons.api.functional.EntryView.*;
import org.infinispan.commons.api.functional.FunctionalMap.*;

WriteOnlyMap<String, String> writeOnlyMap = ...

// Force a function to be Serializable
Consumer<WriteEntryView<String>> function =
   (Consumer<WriteEntryView<String>> & Serializable) wv -> wv.set("one");

CompletableFuture<Void> writeFuture = writeOnlyMap.eval("key1", function);

Infinispan provides overloads for all functional methods that make lambdas passed directly to the API serializable by default; the compiler automatically selects this overload if that’s possible. Therefore you can call

WriteOnlyMap<String, String> writeOnlyMap = ...
CompletableFuture<Void> writeFuture = writeOnlyMap.eval("key1", wv -> wv.set("one"));

without doing the cast described above.

A more economical way to marshall a function is to provide an Infinispan Externalizer for it:

import org.infinispan.commons.api.functional.EntryView.*;
import org.infinispan.commons.api.functional.FunctionalMap.*;
import org.infinispan.commons.marshall.Externalizer;
import org.infinispan.commons.marshall.SerializeFunctionWith;

WriteOnlyMap<String, String> writeOnlyMap = ...

// Force a function to be Serializable
Consumer<WriteEntryView<String>> function = new SetStringConstant<>();
CompletableFuture<Void> writeFuture = writeOnlyMap.eval("key1", function);

@SerializeFunctionWith(value = SetStringConstant.Externalizer0.class)
class SetStringConstant implements Consumer<WriteEntryView<String>> {
   public void accept(WriteEntryView<String> view) {

   public static final class Externalizer0 implements Externalizer<Object> {
      public void writeObject(ObjectOutput oo, Object o) {
         // No-op
      public Object readObject(ObjectInput input) {
         return new SetStringConstant<>();

To help users take advantage of the tiny payloads generated by Externalizer-based functions, the functional API comes with a helper class called org.infinispan.commons.marshall.MarshallableFunctions which provides marshallable functions for some of the most commonly user functions.

In fact, all the functions required to implement ConcurrentMap and JCache using the functional map API have been defined in MarshallableFunctions. For example, here is an implementation of JCache’s boolean putIfAbsent(K, V) using functional map API which can be run in a cluster:

import org.infinispan.commons.api.functional.EntryView.*;
import org.infinispan.commons.api.functional.FunctionalMap.*;
import org.infinispan.commons.marshall.MarshallableFunctions;

ReadWriteMap<String, String> readWriteMap = ...

CompletableFuture<Boolean> future = readWriteMap.eval("key1,
future.thenAccept(stored -> System.out.printf("Value was put? %s%n", stored));

3.7.11. Use Cases for Functional API

This new API is meant to complement existing Key/Value Infinispan API offerings, so you’ll still be able to use ConcurrentMap or JCache standard APIs if that’s what suits your use case best.

The target audience for this new API is either:

  • Distributed or persistent caching/in­memory­data­grid users that want to benefit from CompletableFuture and/or Traversable for async/lazy data grid or caching data manipulation. The clear advantage here is that threads do not need to be idle waiting for remote operations to complete, but instead these can be notified when remote operations complete and then chain them with other subsequent operations.

  • Users who want to go beyond the standard operations exposed by ConcurrentMap and JCache, for example, if you want to do a replace operation using metadata parameter equality instead of value equality, or if you want to retrieve metadata information from values and so on.

3.8. Encoding

3.8.1. Overview

Encoding is the data conversion operation done by Infinispan caches before storing data, and when reading back from storage.

It allows dealing with a certain data format during API calls (map, listeners, stream, etc) while the format effectively stored is different.

The data conversions are handled by instances of org.infinispan.commons.dataconversion.Encoder :

public interface Encoder {

    * Convert data in the read/write format to the storage format.
    * @param content data to be converted, never null.
    * @return Object in the storage format.
   Object toStorage(Object content);

    * Convert from storage format to the read/write format.
    * @param content data as stored in the cache, never null.
    * @return data in the read/write format
   Object fromStorage(Object content);

     * Returns the {@link MediaType} produced by this encoder or null if the storage format is not known.
   MediaType getStorageFormat();

3.8.2. Default encoders

Infinispan automatically picks the Encoder depending on the cache configuration. The table below shows which internal Encoder is used for several configurations:

Mode Configuration Encoder Description




Passthrough encoder, no conversion done




Use the Infinispan internal marshaller to convert to byte[]. May delegate to the configured marshaller in the cache manager.




Use the Infinispan internal marshaller to convert to byte[], except for primitives and String.




Store byte[]s directly as received by remote clients

3.8.3. Overriding programmatically

Is is possible to override programmatically the encoding used for both keys and values, by calling the .withEncoding() method variants from AdvancedCache.

Example, consider the following cache configured as OFF_HEAP:

// Read and write POJO, storage will be byte[] since for
// OFF_HEAP the GlobalMarshallerEncoder is used internally:
cache.put(1, new Pojo())
Pojo value = cache.get(1)

// Get the content in its stored format by overriding
// the internal encoder with a no-op encoder (IdentityEncoder)
Cache<?,?> rawContent = cache.getAdvancedCache().withValueEncoding(IdentityEncoder.class)
byte[] marshalled = rawContent.get(1)

The override can be useful if any operation in the cache does not require decoding, such as counting number of entries, or calculating the size of byte[] of an OFF_HEAP cache.

3.8.4. Defining custom Encoders

A custom encoder can be registered in the EncoderRegistry.

Ensure that the registration is done in every node of the cluster, before starting the caches.

Consider a custom encoder used to compress/decompress with gzip:

public class GzipEncoder implements Encoder {

   public Object toStorage(Object content) {
      assert content instanceof String;
      return compress(content.toString());

   public Object fromStorage(Object content) {
      assert content instanceof byte[];
      return decompress((byte[]) content);

   private byte[] compress(String str) {
      try (ByteArrayOutputStream baos = new ByteArrayOutputStream();
           GZIPOutputStream gis = new GZIPOutputStream(baos)) {
         return baos.toByteArray();
      } catch (IOException e) {
         throw new RuntimeException("Unabled to compress", e);

   private String decompress(byte[] compressed) {
      try (GZIPInputStream gis = new GZIPInputStream(new ByteArrayInputStream(compressed));
           BufferedReader bf = new BufferedReader(new InputStreamReader(gis, "UTF-8"))) {
         StringBuilder result = new StringBuilder();
         String line;
         while ((line = bf.readLine()) != null) {
         return result.toString();
      } catch (IOException e) {
         throw new RuntimeException("Unable to decompress", e);

   public MediaType getStorageFormat() {
      return MediaType.parse("application/gzip");

   public boolean isStorageFormatFilterable() {
      return false;

   public short id() {
      return 10000;

It can be registered by:

GlobalComponentRegistry registry = cacheManager.getGlobalComponentRegistry();
EncoderRegistry encoderRegistry = registry.getComponent(EncoderRegistry.class);
encoderRegistry.registerEncoder(new GzipEncoder());

And then be used to write and read data from a cache:

AdvancedCache<String, String> cache = ...

// Decorate cache with the newly registered encoder, without encoding keys (IdentityEncoder)
// but compressing values
AdvancedCache<String, String> compressingCache = (AdvancedCache<String, String>) cache.withEncoding(IdentityEncoder.class, GzipEncoder.class);

// All values will be stored compressed...
compressingCache.put("297931749", "0412c789a37f5086f743255cfa693dd5");

// ... but API calls deals with String
String value = compressingCache.get("297931749");

// Bypassing the value encoder to obtain the value as it is stored
Object value = compressingCache.withEncoding(IdentityEncoder.class).get("297931749");

// value is a byte[] which is the compressed value

3.8.5. MediaType

A Cache can optionally be configured with a org.infinispan.commons.dataconversion.MediaType for keys and values. By describing the data format of the cache, Infinispan is able to convert data on the fly during cache operations.

The MediaType configuration is more suitable when storing binary data. When using server mode, it’s common to have a MediaType configured and clients such as REST or Hot Rod reading and writing in different formats.

The data conversion between MediaType formats are handled by instances of org.infinispan.commons.dataconversion.Transcoder

public interface Transcoder {

    * Transcodes content between two different {@link MediaType}.
    * @param content         Content to transcode.
    * @param contentType     The {@link MediaType} of the content.
    * @param destinationType The target {@link MediaType} to convert.
    * @return the transcoded content.
   Object transcode(Object content, MediaType contentType, MediaType destinationType);

    * @return all the {@link MediaType} handled by this Transcoder.
   Set<MediaType> getSupportedMediaTypes();


      <key media-type="application/x-java-object; type=java.lang.Integer"/>
      <value media-type="application/xml; charset=UTF-8"/>


ConfigurationBuilder cfg = new ConfigurationBuilder();

Overriding the MediaType Programmatically

It’s possible to decorate the Cache with a different MediaType, allowing cache operations to be executed sending and receiving different data formats.


DefaultCacheManager cacheManager = new DefaultCacheManager();

// The cache will store POJO for keys and values
ConfigurationBuilder cfg = new ConfigurationBuilder();

cacheManager.defineConfiguration("mycache", cfg.build());

Cache<Integer, Person> cache = cacheManager.getCache("mycache");

cache.put(1, new Person("John","Doe"));

// Wraps cache using 'application/x-java-object' for keys but JSON for values
Cache<Integer, byte[]> jsonValuesCache = (Cache<Integer, byte[]>) cache.getAdvancedCache().withMediaType("application/x-java-object", "application/json");

byte[] json = jsonValuesCache.get(1);

Will return the value in JSON format:

Most Transcoders are installed when server mode is used; when using library mode, an extra dependency, org.infinispan:infinispan-server-core should be added to the project.
Transcoders and Encoders

Usually there will be none or only one data conversion involved in a cache operation:

  • No conversion by default on caches using in embedded or server mode;

  • Encoder based conversion for embedded caches without MediaType configured, but using OFF_HEAP or BINARY;

  • Transcoder based conversion for caches used in server mode with multiple REST and Hot Rod clients sending and receiving data in different formats. Those caches will have MediaType configured describing the storage.

But it’s possible to have both encoders and transcoders being used simultaneously for advanced use cases.

Consider an example, a cache that stores marshalled objects (with jboss marshaller) content but for security reasons a transparent encryption layer should be added in order to avoid storing "plain" data to an external store. Clients should be able to read and write data in multiple formats.

This can be achieved by configuring the cache with the the MediaType that describes the storage regardless of the encoding layer:

ConfigurationBuilder cfg = new ConfigurationBuilder();

The transparent encryption can be added by decorating the cache with a special Encoder that encrypts/decrypts with storing/retrieving, for example:

public class Scrambler implements Encoder {

   Object toStorage(Object content) {
      // Encrypt data

   Object fromStorage(Object content) {
      // Decrypt data

   MediaType getStorageFormat() {
      return "application/scrambled";


To make sure all data written to the cache will be stored encrypted, it’s necessary to decorate the cache with the Encoder above and perform all cache operations in this decorated cache:

Cache<?,?> secureStorageCache = cache.getAdvancedCache().withEncoding(Scrambler.class).put(k,v);

The capability of reading data in multiple formats can be added by decorating the cache with the desired MediaType:

// Obtain a stream of values in XML format from the secure cache

Internally, Infinispan will first apply the encoder fromStorage operation to obtain the entries, that will be in "application/x-jboss-marshalling" format and then apply a successive conversion to "application/xml" by using the adequate Transcoder.

4. Eviction and Data Container

Infinispan supports eviction of entries, such that you do not run out of memory. Eviction is typically used in conjunction with a cache store, so that entries are not permanently lost when evicted, since eviction only removes entries from memory and not from cache stores or the rest of the cluster.

Infinispan supports storing data in a few different formats. Data can be stored as the object iself, binary as a byte[], and off-heap which stores the byte[] in native memory.

Passivation is also a popular option when using eviction, so that only a single copy of an entry is maintained - either in memory or in a cache store, but not both. The main benefit of using passivation over a regular cache store is that updates to entries which exist in memory are cheaper since the update doesn’t need to be made to the cache store as well.
Eviction occurs on a local basis, and is not cluster-wide. Each node runs an eviction thread to analyse the contents of its in-memory container and decide what to evict. Eviction does not take into account the amount of free memory in the JVM as threshold to starts evicting entries. You have to set size attribute of the eviction element to be greater than zero in order for eviction to be turned on. If size is too large you can run out of memory. The size attribute will probably take some tuning in each use case.

4.1. Enabling Eviction

Eviction is configured by adding the <memory /> element to your <*-cache /> configuration sections or using MemoryConfigurationBuilder API programmatic approach.

All cache entry are evicted by piggybacking on user threads that are hitting the cache.

4.1.1. Eviction strategy

Strategies control how the eviction is handled.

The possible choices are


Eviction is not enabled and it is assumed that the user will not invoke evict directly on the cache. If passivation is enabled this will cause aa warning message to be emitted. This is the default strategy.


This strategy is just like <b>NONE</b> except that it assumes the user will be invoking evict directly. This way if passivation is enabled no warning message is logged.


This strategy will actually evict "old" entries to make room for incoming ones.

Eviction is handled by Caffeine utilizing the TinyLFU algorithm with an additional admission window. This was chosen as provides high hit rate while also requiring low memory overhead. This provides a better hit ratio than LRU while also requiring less memory than LIRS.


This strategy actually prevents new entries from being created by throwing a ContainerFullException. This strategy only works with transactional caches that always run with 2 phase commit, that is no 1 phase commit or synchronization optimizations allowed.

4.1.2. Eviction types

Eviction type applies only when the size is set to something greater than 0. The eviction type below determines when the container will decide to remove entries.


This type of eviction will remove entries based on how many there are in the cache. Once the count of entries has grown larger than the size then an entry will be removed to make room.


This type of eviction will estimate how much each entry will take up in memory and will remove an entry when the total size of all entries is larger than the configured size. This type does not work with OBJECT storage type below.

4.1.3. Storage type

Infinispan allows the user to configure in what form their data is stored. Each form supports the same features of Infinispan, however eviction can be limited for some forms. There are currently three storage formats that Infinispan provides, they are:


Stores the keys and values as objects in the Java heap Only COUNT eviction type is supported.


Stores the keys and values as a byte[] in the Java heap. This will use the configured marshaller for the cache if there is one. Both COUNT and MEMORY eviction types are supported.


Stores the keys and values in native memory outside of the Java heap as bytes. The configured marshaller will be used if the cache has one. Both COUNT and MEMORY eviction types are supported.

Both BINARY and OFF-HEAP violate equality and hashCode that they are dictated by the resulting byte[] they generate instead of the object instance.

4.1.4. More defaults

By default when no <memory /> element is specified, no eviction takes place, OBJECT storage type is used, and a strategy of NONE is assumed.

In case there is an memory element, this table describes the behaviour of eviction based on information provided in the xml configuration ("-" in Supplied size or Supplied strategy column means that the attribute wasn’t supplied)

Supplied size Example Eviction behaviour


<memory />

no eviction as an object


<memory> <object strategy="MANUAL" /> </memory>

no eviction as an object and won’t log warning if passivation is enabled

> 0

<memory> <object size="100" /> </memory>

eviction takes place and stored as objects

> 0

<memory> <binary size="100" eviction="MEMORY"/> </memory>

eviction takes place and stored as a binary removing to make sure memory doens’t go higher than 100

> 0

<memory> <off-heap size="100" /> </memory>

eviction takes place and stored in off-heap

> 0

<memory> <off-heap size="100" strategy="EXCEPTION" /> </memory>

entries are stored in off-heap and if 100 entries are in container exceptions will be thrown for additional


<memory> <object size="0" /> </memory>

no eviction

< 0

<memory> <object size="-1" /> </memory>

no eviction

4.2. Expiration

Similar to, but unlike eviction, is expiration. Expiration allows you to attach lifespan and/or maximum idle times to entries. Entries that exceed these times are treated as invalid and are removed. When removed expired entries are not passivated like evicted entries (if passivation is turned on).

Unlike eviction, expired entries are removed globally - from memory, cache stores, and cluster-wide.

By default entries created are immortal and do not have a lifespan or maximum idle time. Using the cache API, mortal entries can be created with lifespans and/or maximum idle times. Further, default lifespans and/or maximum idle times can be configured by adding the <expiration /> element to your <*-cache /> configuration sections.

When an entry expires it resides in the data container or cache store until it is accessed again by a user request. An expiration reaper is also available to check for expired entries and remove them at a configurable interval of milliseconds.

You can enable the expiration reaper declaratively with the reaper-interval attribute or programmatically with the enableReaper method in the ExpirationConfigurationBuilder class.

  • The expiration reaper cannot be disabled when a cache store is present.

  • When using a maximum idle time in a clustered cache, you should always enable the expiration reaper. For more information, see Clustered Max Idle.

4.2.1. Difference between Eviction and Expiration

Both Eviction and Expiration are means of cleaning the cache of unused entries and thus guarding the heap against OutOfMemory exceptions, so now a brief explanation of the difference.

With eviction you set maximal number of entries you want to keep in the cache and if this limit is exceeded, some candidates are found to be removed according to a choosen eviction strategy (LRU, LIRS, etc…​). Eviction can be setup to work with passivation, which is eviction to a cache store.

With expiration you set time criteria for entries to specify how long you want to keep them in the cache.


Specifies how long entries can remain in the cache before they expire. The default value is -1, which is unlimited time.

maximum idle time

Specifies how long entries can remain idle before they expire. An entry in the cache is idle when no operation is performed with the key. The default value is -1, which is unlimited time.

4.3. Expiration details

  1. Expiration is a top-level construct, represented in the configuration as well as in the cache API.

  2. While eviction is local to each cache instance , expiration is cluster-wide . Expiration lifespan and maxIdle values are replicated along with the cache entry.

  3. Maximum idle times for cache entries require additional network messages in clustered environments. For this reason, setting maxIdle in clustered caches can result in slower operation times.

  4. Expiration lifespan and maxIdle are also persisted in CacheStores, so this information survives eviction/passivation.

4.3.1. Maximum Idle Expiration

Maximum idle expiration has different behavior in local and clustered cache environments.

Maximum idle expiration, max-idle, does not currently work with entries stored in off-heap memory. Likewise, max-idle does not work if caches use cache stores as a persistence layer.

Local Max Idle

In local cache mode, Infinispan expires entries with the maxIdle configuration when:

  • accessed directly (Cache.get()).

  • iterated upon (Cache.size()).

  • the expiration reaper thread runs.

Clustered Max Idle

In clustered cache modes, when clients read entries that have max-idle expiration values, Infinispan sends touch commands to all owners. This ensures that the entries have the same relative access time across the cluster.

When nodes detect that an entry reaches the maximum idle time, Infinispan removes it from the cache and does not return the entry to the client that requested it.

Before using max-idle with clustered cache modes, you should review the following points:

  • Cache.get() does not return until the touch commands complete. This synchronous behavior increases latency of client requests.

  • Clustered max-idle also updates the "recently accessed" metadata for cache entries on all owners, which Infinispan uses for eviction.

  • With scattered cache mode, Infinispan sends touch commands to all nodes, not just primary and backup owners.

  • Iteration across a clustered cache returns entries that might be expired with the maximum idle time. This behavior ensures performance because no remote invocations are performed during the iteration. However this does not refresh any expired entries, which are removed by the expiration reaper or when accessed directly (Cache.get()).

  • Clustered caches should always use the expiration reaper with the maxIdle configuration.

  • When using maxIdle expiration with exception-based eviction, entries that are expired but not removed from the cache count towards the size of the data container.

4.3.2. Configuration

Eviction and Expiration may be configured using the programmatic or declarative XML configuration. This configuration is on a per-cache basis. Valid eviction/expiration-related configuration elements are:

<!-- Eviction -->
   <object size="2000"/>
<!-- Expiration -->
<expiration lifespan="1000" max-idle="500" interval="1000" />

Programmatically, the same would be defined using:

Configuration c = new ConfigurationBuilder()

4.3.3. Memory Based Eviction Configuration

Memory based eviction may require some additional configuration options if you are using your own custom types (as Infinispan is normally used). In this case Infinispan cannot estimate the memory usage of your classes and as such you are required to use storeAsBinary when memory based eviction is used.

<!-- Enable memory based eviction with 1 GB/>
   <binary size="1000000000" eviction="MEMORY"/>
Configuration c = new ConfigurationBuilder()

4.3.4. Default values

Eviction is disabled by default. Default values are used:

  • size: -1 is used if not specified, which means unlimited entries.

  • 0 means no entries, and the eviction thread will strive to keep the cache empty.

Expiration lifespan and maxIdle both default to -1, which means that entries will be created immortal by default. This can be overridden per entry with the API.

4.3.5. Using expiration

Expiration allows you to set either a lifespan or a maximum idle time on each key/value pair stored in the cache. This can either be set cache-wide using the configuration, as described above, or it can be defined per-key/value pair using the Cache interface. Any values defined per key/value pair overrides the cache-wide default for the specific entry in question.

For example, assume the following configuration:

<expiration lifespan="1000" />
// this entry will expire in 1000 millis
cache.put("pinot noir", pinotNoirPrice);

// this entry will expire in 2000 millis
cache.put("chardonnay", chardonnayPrice, 2, TimeUnit.SECONDS);

// this entry will expire 1000 millis after it is last accessed
cache.put("pinot grigio", pinotGrigioPrice, -1,
          TimeUnit.SECONDS, 1, TimeUnit.SECONDS);

// this entry will expire 1000 millis after it is last accessed, or
// in 5000 millis, which ever triggers first
cache.put("riesling", rieslingPrice, 5,
          TimeUnit.SECONDS, 1, TimeUnit.SECONDS);

4.4. Expiration designs

Central to expiration is an ExpirationManager.

The purpose of the ExpirationManager is to drive the expiration thread which periodically purges items from the DataContainer. If the expiration thread is disabled (wakeupInterval set to -1) expiration can be kicked off manually using ExprationManager.processExpiration(), for example from another maintenance thread that may run periodically in your application.

The expiration manager processes expirations in the following manner:

  1. Causes the data container to purge expired entries

  2. Causes cache stores (if any) to purge expired entries

5. Persistence

Persistence allows configuring external (persistent) storage engines complementary to the default in memory storage offered by Infinispan. An external persistent storage might be useful for several reasons:

  • Increased Durability. Memory is volatile, so a cache store could increase the life-span of the information store in the cache.

  • Write-through. Interpose Infinispan as a caching layer between an application and a (custom) external storage engine.

  • Overflow Data. By using eviction and passivation, one can store only the "hot" data in memory and overflow the data that is less frequently used to disk.

The integration with the persistent store is done through the following SPI: CacheLoader, CacheWriter, AdvancedCacheLoader and AdvancedCacheWriter (discussed in the following sections).

These SPIs allow for the following features:

  • Alignment with JSR-107. The CacheWriter and CacheLoader interface are similar to the the loader and writer in JSR 107. This should considerably help writing portable stores across JCache compliant vendors.

  • Simplified Transaction Integration. All necessary locking is handled by Infinispan automatically and implementations don’t have to be concerned with coordinating concurrent access to the store. Even though concurrent writes on the same key are not going to happen (depending locking mode in use), implementors should expect operations on the store to happen from multiple/different threads and code the implementation accordingly.

  • Parallel Iteration. It is now possible to iterate over entries in the store with multiple threads in parallel.

  • Reduced Serialization. This translates in less CPU usage. The new API exposes the stored entries in serialized format. If an entry is fetched from persistent storage for the sole purpose of being sent remotely, we no longer need to deserialize it (when reading from the store) and serialize it back (when writing to the wire). Now we can write to the wire the serialized format as read from the storage directly.

5.1. Configuration

Stores (readers and/or writers) can be configured in a chain. Cache read operation looks at all of the specified CacheLoader s, in the order they are configured, until it finds a valid and non-null element of data. When performing writes all cache CacheWriter s are written to, except if the ignoreModifications element has been set to true for a specific cache writer.

Implementing both a CacheWriter and CacheLoader

Store providers should implement both the CacheWriter and the CacheLoader interfaces. Stores that do this are considered both for reading and writing (assuming read-only=false) data.

This is the configuration of a custom(not shipped with infinispan) store:
   <local-cache name="myCustomStore">
      <persistence passivation="false">
            fetch-state="false" preload="true" shared="false"
            purge="true" read-only="false" singleton="false" segmented="true">

            <write-behind modification-queue-size="123" thread-pool-size="23" />

            <property name="myProp">${system.property}</property>

Parameters that you can use to configure persistence are as follows:


Sets the maximum number of attempts to start each configured CacheWriter/CacheLoader. If the attempts to start are not successful, an exception is thrown and the cache does not start.


Specifies the time, in milliseconds, to wait between connection attempts on startup. A negative or zero value means no wait between connection attempts.


Specifies the time, in milliseconds, between availability checks to determine if the PersistenceManager is available. In other words, this interval sets how often stores/loaders are polled via their org.infinispan.persistence.spi.CacheWriter#isAvailable or org.infinispan.persistence.spi.CacheLoader#isAvailable implementation. If a single store/loader is not available, an exception is thrown during cache operations.


Enables passivation. The default value is false (boolean).

This property has a significant impact on Infinispan interactions with the loaders. See Cache Passivation for more information.


Defines the class of the store and must implement CacheLoader, CacheWriter, or both.


Fetches the persistent state of a cache when joining a cluster. The default value is false (boolean).

The purpose of this property is to retrieve the persistent state of a cache and apply it to the local cache store of a node when it joins a cluster. Fetching the persistent state does not apply if a cache store is shared because it accesses the same data as the other stores.

This property can be true for one configured cache loader only. If more than one cache loader fetches the persistent state, a configuration exception is thrown when the cache service starts.


Pre-loads data into memory from the cache loader when the cache starts. The default value is false (boolean).

This property is useful when data in the cache loader is required immediately after startup to prevent delays with cache operations when the data is loaded lazily. This property can provide a "warm cache" on startup but it impacts performance because it affects start time.

Pre-loading data is done locally, so any data loaded is stored locally in the node only. The pre-loaded data is not replicated or distributed. Likewise, Infinispan pre-loads data only up to the maximum configured number of entries in eviction.


Determines if the cache loader is shared between cache instances. The default value is false (boolean).

This property prevents duplicate writes of data to the cache loader by different cache instances. An example is where all cache instances in a cluster use the same JDBC settings for the same remote, shared database.


Configures a cache store to segment data. The default value is false (boolean).

If true the cache store stores data in buckets. The hash.numSegments property configures how many buckets there are for storing data.

Depending on the cache store implementation, segmenting data can cause slower write operations. However, performance improves for other cache operations. See Segmented Stores for more information.


Prevents new data from being persisted to the cache store. The default value is false (boolean).


Empties the specified cache loader at startup. The default value is false (boolean). This property takes effect only if the read-only property is set to false.


Sets the maximum size of a batch to insert of delete from the cache store. The default value is #{AbstractStore-maxBatchSize}.

If the value is less than 1, no upper limit applies to the number of operations in a batch.


Asynchronously persists data to the cache store. The default value is false (boolean). See Asynchronous Write-Behind for more information.


Enables one node in the cluster, the coordinator, to store modifications. The default value is false (boolean).

Whenever data enters a node, it is replicated or distributed to keep the in-memory state of the caches synchronized. The coordinator is responsible for pushing that state to disk.

If true you must set this property on all nodes in the cluster. Only the coordinator of the cluster persists data. However, all nodes must have this property enabled to prevent other nodes from persisting data.

You cannot enable the singleton property if the cache store is shared.

You can define additional attributes in the properties section to configure specific aspects of each cache loader, such as the myProp attribute in the previous example.

Other cache loaders with more complex configurations also include additional properties. See the following JDBC cache store configuration for examples.

The preceding configuration applies a generic cache store implementation. However, the default Infinispan store implementation has a more complex configuration schema, in which the properties section is replaced with XML attributes:

<persistence passivation="false">
   <!-- note that class is missing and is induced by the fileStore element name -->
           shared="false" preload="true"
      <write-behind thread-pool-size="5" />

The same configuration can be achieved programmatically:

   ConfigurationBuilder builder = new ConfigurationBuilder();

5.2. Cache Passivation

A CacheWriter can be used to enforce entry passivation and activation on eviction in a cache. Cache passivation is the process of removing an object from in-memory cache and writing it to a secondary data store (e.g., file system, database) on eviction. Cache activation is the process of restoring an object from the data store into the in-memory cache when it’s needed to be used. In order to fully support passivation, a store needs to be both a CacheWriter and a CacheLoader. In both cases, the configured cache store is used to read from the loader and write to the data writer.

When an eviction policy in effect evicts an entry from the cache, if passivation is enabled, a notification that the entry is being passivated will be emitted to the cache listeners and the entry will be stored. When a user attempts to retrieve a entry that was evicted earlier, the entry is (lazily) loaded from the cache loader into memory. When the entry has been loaded a notification is emitted to the cache listeners that the entry has been activated. In order to enable passivation just set passivation to true (false by default). When passivation is used, only the first cache loader configured is used and all others are ignored.

5.2.1. Limitations

Due to the unique nature of passivation, it is not supported with some other store configurations.

  • Transactional store - Passivation writes/removes entries from the store outside the scope of the actual Infinispan commit boundaries.

  • Shared store - Shared store requires entries always being in the store for other owners. Thus passivation makes no sense as we can’t remove the entry from the store.

5.2.2. Cache Loader Behavior with Passivation Disabled vs Enabled

When passivation is disabled, whenever an element is modified, added or removed, then that modification is persisted in the backend store via the cache loader. There is no direct relationship between eviction and cache loading. If you don’t use eviction, what’s in the persistent store is basically a copy of what’s in memory. If you do use eviction, what’s in the persistent store is basically a superset of what’s in memory (i.e. it includes entries that have been evicted from memory). When passivation is enabled, and with an unshared store, there is a direct relationship between eviction and the cache loader. Writes to the persistent store via the cache loader only occur as part of the eviction process. Data is deleted from the persistent store when the application reads it back into memory. In this case, what’s in memory and what’s in the persistent store are two subsets of the total information set, with no intersection between the subsets. With a shared store, entries which have been passivated in the past will continue to exist in the store, although they may have a stale value if this has been overwritten in memory.

The following is a simple example, showing what state is in RAM and in the persistent store after each step of a 6 step process:

Operation Passivation Off Passivation On, Shared Off Passivation On, Shared On

Insert keyOne

Memory: keyOne
Disk: keyOne

Memory: keyOne
Disk: (none)

Memory: keyOne
Disk: (none)

Insert keyTwo

Memory: keyOne, keyTwo
Disk: keyOne, keyTwo

Memory: keyOne, keyTwo
Disk: (none)

Memory: keyOne, keyTwo
Disk: (none)

Eviction thread runs, evicts keyOne

Memory: keyTwo
Disk: keyOne, keyTwo

Memory: keyTwo
Disk: keyOne

Memory: keyTwo
Disk: keyOne

Read keyOne

Memory: keyOne, keyTwo
Disk: keyOne, keyTwo

Memory: keyOne, keyTwo
Disk: (none)

Memory: keyOne, keyTwo
Disk: keyOne

Eviction thread runs, evicts keyTwo

Memory: keyOne
Disk: keyOne, keyTwo

Memory: keyOne
Disk: keyTwo

Memory: keyOne
Disk: keyOne, keyTwo

Remove keyTwo

Memory: keyOne
Disk: keyOne

Memory: keyOne
Disk: (none)

Memory: keyOne
Disk: keyOne

5.3. Cache Loaders and transactional caches

When a cache is transactional and a cache loader is present, the cache loader won’t be enlisted in the transaction in which the cache is part. That means that it is possible to have inconsistencies at cache loader level: the transaction to succeed applying the in-memory state but (partially) fail applying the changes to the store. Manual recovery would not work with caches stores.

5.4. Write-Through And Write-Behind Caching

Infinispan can optionally be configured with one or several cache stores allowing it to store data in a persistent location such as shared JDBC database, a local filesystem, etc. Infinispan can handle updates to the cache store in two different ways:

  • Write-Through (Synchronous)

  • Write-Behind (Asynchronous)

5.4.1. Write-Through (Synchronous)

In this mode, which is supported in version 4.0, when clients update a cache entry, i.e. via a Cache.put() invocation, the call will not return until Infinispan has gone to the underlying cache store and has updated it. Normally, this means that updates to the cache store are done within the boundaries of the client thread.

The main advantage of this mode is that the cache store is updated at the same time as the cache, hence the cache store is consistent with the cache contents. On the other hand, using this mode reduces performance because the latency of having to access and update the cache store directly impacts the duration of the cache operation.

Configuring a write-through or synchronous cache store does not require any particular configuration option. By default, unless marked explicitly as write-behind or asynchronous, all cache stores are write-through or synchronous. Please find below a sample configuration file of a write-through unshared local file cache store:

<persistence passivation="false">
   <file-store fetch-state="true"
               purge="false" path="${java.io.tmpdir}"/>

5.4.2. Write-Behind (Asynchronous)

In this mode, updates to the cache are asynchronously written to the cache store. Infinispan puts pending changes into a modification queue so that it can quickly store changes.

The configured number of threads consume the queue and apply the modifications to the underlying cache store. If the configured number of threads cannot consume the modifications fast enough, or if the underlying store becomes unavailable, the modification queue becomes full. In this event, the cache store becomes write-through until the queue can accept new entries.

This mode provides an advantage in that cache operations are not affected by updates to the underlying store. However, because updates happen asynchronously, there is a period of time during which data in the cache store is inconsistent with data in the cache.

The write-behind strategy is suitable for cache stores with low latency and small operational cost; for example, an unshared file-based cache store that is local to the cache itself. In this case, the time during which data is inconsistent between the cache store and the cache is reduced to the lowest possible period.

The following is an example configuration for the write-behind strategy:

<persistence passivation="false">
   <file-store fetch-state="true"
               purge="false" path="${java.io.tmpdir}">
   <!-- start write-behind configuration -->
   <write-behind modification-queue-size="123" thread-pool-size="23" />
   <!-- end write-behind configuration -->

5.4.3. Segmented Stores

You can configure stores so that data resides in segments to which keys map. See Key Ownership for more information about segments and ownership.

Segmented stores increase read performance for bulk operations; for example, streaming over data (Cache.size, Cache.entrySet.stream), pre-loading the cache, and doing state transfer operations.

However, segmented stores can also result in loss of performance for write operations. This performance loss applies particularly to batch write operations that can take place with transactions or write-behind stores. For this reason, you should evaluate the overhead for write operations before you enable segmented stores. The performance gain for bulk read operations might not be acceptable if there is a significant performance loss for write operations.

Loss of data can occur if the number of segments in a cache store are not changed gracefully. For this reason, if you change the numSegments setting in the store configuration, you must migrate the existing store to use the new configuration.

The recommended method to migrate the cache store configuration is to perform a rolling upgrade. The store migrator supports migrating a non-segmented cache store to a segmented cache store only. The store migrator does not currently support migrating from a segmented cache store.

Not all cache stores support segmentation. See the appropriate section for each store to determine if it supports segmentation.

If you plan to convert or write a new store to support segmentation, see the following SPI section that provides more details.

5.5. Filesystem based cache stores

A filesystem-based cache store is typically used when you want to have a cache with a cache store available locally which stores data that has overflowed from memory, having exceeded size and/or time restrictions.

Usage of filesystem-based cache stores on shared filesystems like NFS, Windows shares, etc. should be avoided as these do not implement proper file locking and can cause data corruption. File systems are inherently not transactional, so when attempting to use your cache in a transactional context, failures when writing to the file (which happens during the commit phase) cannot be recovered.

5.5.1. Single File Store

The single file cache store keeps all data in a single file. The way it looks up data is by keeping an in-memory index of keys and the positions of their values in this file. This results in greater performance compared to old file cache store. There is one caveat though. Since the single file based cache store keeps keys in memory, it can lead to increased memory consumption, and hence it’s not recommended for caches with big keys.

In certain use cases, this cache store suffers from fragmentation: if you store larger and larger values, the space is not reused and instead the entry is appended at the end of the file. The space (now empty) is reused only if you write another entry that can fit there. Also, when you remove all entries from the cache, the file won’t shrink, and neither will be de-fragmented.

These are the available configuration options for the single file cache store:

  • path is the file-system location where Infinispan stores data.

    For embedded deployments, Infinispan stores data to the current working directory. To store data in a different location, specify the absolute path to that directory as the value.

    For server deployments, Infinispan stores data in $ISPN_HOME/data. You should not change the default location for Infinispan server. In other words, do not set a value for path with server deployments.

you should create a path declaration and use the relative-to attribute for the file store location. If you set relative-to to a directory location or environment variable, Infinispan server startup fails.

  • max-entries specifies the maximum number of entries to keep in this file store. As mentioned before, in order to speed up lookups, the single file cache store keeps an index of keys and their corresponding position in the file. To avoid this index resulting in memory consumption problems, this cache store can bounded by a maximum number of entries that it stores. If this limit is exceeded, entries are removed permanently using the LRU algorithm both from the in-memory index and the underlying file based cache store. So, setting a maximum limit only makes sense when Infinispan is used as a cache, whose contents can be recomputed or they can be retrieved from the authoritative data store. If this maximum limit is set when the Infinispan is used as an authoritative data store, it could lead to data loss, and hence it’s not recommended for this use case. The default value is -1 which means that the file store size is unlimited.

Segmentation support

The single file cache store supports segmentation and creates a separate instance per segment. Segmentation results in multiple directories under the configured directory, where each directory is a number that represents the segment to which the data maps.

Declarative Configuration

The following examples show declarative configuration for a single file cache store.

Library Mode
   <file-store path="/tmp/myDataStore" max-entries="5000"/>
Server Mode
   <file-store max-entries="5000"/>
Programmatic Configuration

The following example shows programmatic configuration for a single file cache store:

ConfigurationBuilder b = new ConfigurationBuilder();

5.5.2. Soft-Index File Store

The Soft-Index File Store is an experimental local file-based. It is a pure Java implementation that tries to get around Single File Store’s drawbacks by implementing a variant of B+ tree that is cached in-memory using Java’s soft references - here’s where the name Soft-Index File Store comes from. This B+ tree (called Index) is offloaded on filesystem to single file that does not need to be persisted - it is purged and rebuilt when the cache store restarts, its purpose is only offloading.

The data that should be persisted are stored in a set of files that are written in append-only way - that means that if you store this on conventional magnetic disk, it does not have to seek when writing a burst of entries. It is not stored in single file but set of files. When the usage of any of these files drops below 50% (the entries from the file are overwritten to another file), the file starts to be collected, moving the live entries into different file and in the end removing that file from disk.

Most of the structures in Soft Index File Store are bounded, therefore you don’t have to be afraid of OOMEs. For example, you can configure the limits for concurrently open files as well.

Segmentation support

The Soft-Index file store supports segmentation and creates a separate instance per segment. Segmentation results in multiple directories under the configured directory, where each directory is a number that represents the segment to which the data maps.


Here is an example of Soft-Index File Store configuration via XML:

    <soft-index-file-store xmlns="urn:infinispan:config:store:soft-index:8.0">
        <index path="/tmp/sifs/testCache/index" />
        <data path="/tmp/sifs/testCache/data" />

Programmatic configuration would look as follows:

ConfigurationBuilder b = new ConfigurationBuilder();
Current limitations

Size of a node in the Index is limited, by default it is 4096 bytes, though it can be configured. This size also limits the key length (or rather the length of the serialized form): you can’t use keys longer than size of the node - 15 bytes. Moreover, the key length is stored as 'short', limiting it to 32767 bytes. There’s no way how you can use longer keys - SIFS throws an exception when the key is longer after serialization.

When entries are stored with expiration, SIFS cannot detect that some of those entries are expired. Therefore, such old file will not be compacted (method AdvancedStore.purgeExpired() is not implemented). This can lead to excessive file-system space usage.

5.6. JDBC String based Cache Store

A cache store which relies on the provided JDBC driver to load/store values in the underlying database.

Each key in the cache is stored in its own row in the database. In order to store each key in its own row, this store relies on a (pluggable) bijection that maps the each key to a String object. The bijection is defined by the Key2StringMapper interface. Infinispans ships a default implementation (smartly named DefaultTwoWayKey2StringMapper) that knows how to handle primitive types.

By default Infinispan shares are not stored, meaning that all nodes in the cluster will write to the underlying store upon each update. If you wish for an operation to only be written to the underlying database once, you must configure the JDBC store to be shared.

The JDBC string-based cache store does not support segmentation. Support will be available in a future release.

5.6.1. Connection management (pooling)

In order to obtain a connection to the database the JDBC cache store relies on a ConnectionFactory implementation. The connection factory is specified programmatically using one of the connectionPool(), dataSource() or simpleConnection() methods on the JdbcStringBasedStoreConfigurationBuilder class or declaratively using one of the <connectionPool />, <dataSource /> or <simpleConnection /> elements.

Infinispan ships with three ConnectionFactory implementations:

  • PooledConnectionFactoryConfigurationBuilder is a factory based on HikariCP. Additional properties for HikariCP can be provided by a properties file, either via placing a hikari.properties file on the classpath or by specifying the path to the file via PooledConnectionFactoryConfiguration.propertyFile or properties-file in the connection pool’s xml config. N.B. a properties file specified explicitly in the configuration is loaded instead of the hikari.properties file on the class path and Connection pool characteristics which are explicitly set in PooledConnectionFactoryConfiguration always override the values loaded from a properties file.

Refer to the official documentation for details of all configuration properties.

  • ManagedConnectionFactoryConfigurationBuilder is a connection factory that can be used within managed environments, such as application servers. It knows how to look into the JNDI tree at a certain location (configurable) and delegate connection management to the DataSource.

  • SimpleConnectionFactoryConfigurationBuilder is a factory implementation that will create database connection on a per invocation basis. Not recommended in production.

The PooledConnectionFactory is generally recommended for stand-alone deployments (i.e. not running within AS or servlet container). ManagedConnectionFactory can be used when running in a managed environment where a DataSource is present, so that connection pooling is performed within the DataSource.

5.6.2. Sample configurations

Below is a sample configuration for the JdbcStringBasedStore. For detailed description of all the parameters used refer to the JdbcStringBasedStore.

   <string-keyed-jdbc-store xmlns="urn:infinispan:config:store:jdbc:9.4" dialect="H2">
      <connection-pool connection-url="jdbc:h2:mem:infinispan_string_based;DB_CLOSE_DELAY=-1" username="sa" driver="org.h2.Driver"/>
      <string-keyed-table drop-on-exit="true" create-on-start="true" prefix="ISPN_STRING_TABLE">
         <id-column name="ID_COLUMN" type="VARCHAR(255)" />
         <data-column name="DATA_COLUMN" type="BINARY" />
         <timestamp-column name="TIMESTAMP_COLUMN" type="BIGINT" />
ConfigurationBuilder builder = new ConfigurationBuilder();

Finally, below is an example of a JDBC cache store with a managed connection factory, which is chosen implicitly by specifying a datasource JNDI location:

   <string-keyed-jdbc-store xmlns="urn:infinispan:config:store:jdbc:9.4" shared="true" dialect="H2">
      <data-source jndi-url="java:/StringStoreWithManagedConnectionTest/DS" />
      <string-keyed-table drop-on-exit="true" create-on-start="true" prefix="ISPN_STRING_TABLE">
         <id-column name="ID_COLUMN" type="VARCHAR(255)" />
         <data-column name="DATA_COLUMN" type="BINARY" />
         <timestamp-column name="TIMESTAMP_COLUMN" type="BIGINT" />
ConfigurationBuilder builder = new ConfigurationBuilder();

5.7. Remote store

The RemoteStore is a cache loader and writer implementation that stores data in a remote Infinispan cluster. In order to communicate with the remote cluster, the RemoteStore uses the HotRod client/server architecture. HotRod bering the load balancing and fault tolerance of calls and the possibility to fine-tune the connection between the RemoteCacheStore and the actual cluster. Please refer to Hot Rod for more information on the protocol, client and server configuration. For a list of RemoteStore configuration refer to the javadoc . Example:

5.7.1. Segmentation support

The RemoteStore store supports segmentation because it can publish keys and entries by segment, allowing for more efficient bulk operations.

Segmentation is only supported when the remote server supports at least protocol version 2.3 or newer.

Ensure the number of segments and hash are the same between the store configured cache and the remote server otherwise bulk operations will not return correct results.

5.7.2. Sample Usage

   <remote-store xmlns="urn:infinispan:config:store:remote:8.0" cache="mycache" raw-values="true">
      <remote-server host="one" port="12111" />
      <remote-server host="two" />
      <connection-pool max-active="10" exhausted-action="CREATE_NEW" />
      <write-behind />
ConfigurationBuilder b = new ConfigurationBuilder();

In this sample configuration, the remote cache store is configured to use the remote cache named "mycache" on servers "one" and "two". It also configures connection pooling and provides a custom transport executor. Additionally the cache store is asynchronous.

5.8. Cluster cache loader

The ClusterCacheLoader is a cache loader implementation that retrieves data from other cluster members.

It is a cache loader only as it doesn’t persist anything (it is not a Store), therefore features like fetchPersistentState (and like) are not applicable.

A cluster cache loader can be used as a non-blocking (partial) alternative to stateTransfer : keys not already available in the local node are fetched on-demand from other nodes in the cluster. This is a kind of lazy-loading of the cache content.

The cluster cache loader does not support segmentation.

   <cluster-loader remote-timeout="500"/>
ConfigurationBuilder b = new ConfigurationBuilder();

For a list of ClusterCacheLoader configuration refer to the javadoc .

The ClusterCacheLoader does not support preloading(preload=true). It also won’t provide state if fetchPersistentSate=true.

5.9. Command-Line Interface cache loader

The Command-Line Interface (CLI) cache loader is a cache loader implementation that retrieves data from another Infinispan node using the CLI. The node to which the CLI connects to could be a standalone node, or could be a node that it’s part of a cluster. This cache loader is read-only, so it will only be used to retrieve data, and hence, won’t be used when persisting data.

The CLI cache loader is configured with a connection URL pointing to the Infinispan node to which connect to. Here is an example:

Details on the format of the URL and how to make sure a node can receive invocations via the CLI can be found in the Command-Line Interface chapter.

The Command-Line Interface (CLI) cache loader does not support segmentation.

   <cli-loader connection="jmx://" />
ConfigurationBuilder b = new ConfigurationBuilder();

5.10. RocksDB Cache Store

Infinispan supports using RocksDB as a cache store.

5.10.1. Introduction

RocksDB is a fast key-value filesystem-based storage from Facebook. It started as a fork of Google’s LevelDB, but provides superior performance and reliability, especially in highly concurrent scenarios.

5.10.2. Segmentation support

The RocksDB cache store supports segmentation and creates a separate column family per segment, which substantially improves lookup performance and iteration. However, write operations are a little slower when the cache store is segmented.

You should not configure more than a few hundred segments. RocksDB is not designed to have an unlimited number of column families. Too many segments also significantly increases startup time for the cache store.

Sample Usage

The RocksDB cache store requires 2 filesystem directories to be configured - each directory contains a RocksDB database: one location is used to store non-expired data, while the second location is used to store expired keys pending purge.

Configuration cacheConfig = new ConfigurationBuilder().persistence()
EmbeddedCacheManager cacheManager = new DefaultCacheManager(cacheConfig);

Cache<String, User> usersCache = cacheManager.getCache("usersCache");
usersCache.put("raytsang", new User(...));

5.10.3. Configuration

It is also possible to configure the underlying rocks db instance. This can be done via properties in the store configuration. Any property that is prefixed with the name database will configure the rocks db database. Data is now stored in column families, these can be configured independently of the database by setting a property prefixed with the name data.

Note that you do not have to supply properties and this is entirely optional.

Sample Programatic Configuration
Properties props = new Properties();
props.put("database.max_background_compactions", "2");
props.put("data.write_buffer_size", "512MB");

Configuration cacheConfig = new ConfigurationBuilder().persistence()
Parameter Description


Directory to use for RocksDB to store primary cache store data. The directory will be auto-created if it does not exit.


Directory to use for RocksDB to store expiring data pending to be purged permanently. The directory will be auto-created if it does not exit.


Size of the in-memory queue to hold expiring entries before it gets flushed into expired RocksDB store


There are two methods to clear all entries in RocksDB. One method is to iterate through all entries and remove each entry individually. The other method is to delete the database and re-init. For smaller databases, deleting individual entries is faster than the latter method. This configuration sets the max number of entries allowed before using the latter method


Configuration for RocksDB for data compression, see CompressionType enum for options


Configuration for RocksDB - see documentation for performance tuning


Configuration for RocksDB - see documentation for performance tuning

Sample XML Configuration
<local-cache name="vehicleCache">
      <rocksdb-store path="/tmp/rocksdb/data">
         <expiration path="/tmp/rocksdb/expired"/>
         <property name="database.max_background_compactions">2</property>
         <property name="data.write_buffer_size">512MB</property>

Infinispan servers log warning messages at startup when you add configuration properties that are specific to RocksDB.

For example, as in the preceding example, if you set data.write_buffer_size in your configuration, then the following message is written to logs:

WARN  [org.infinispan.configuration.parsing.XmlConfigHelper] ISPN000292:
Unrecognized attribute 'data.write_buffer_size'. Please check your
configuration. Ignoring!

Infinispan applies the RocksDB properties even though the log message indicates they do not take effect. You can ignore these WARN messages.

5.10.4. Additional References

Refer to the test case for code samples in action.

Refer to test configurations for configuration samples.

5.11. LevelDB Cache Store

The LevelDB Cache Store has been deprecated and has been replaced with the RocksDB Cache Store. If you have existing data stored in a LevelDB Cache Store, the RocksDB Cache Store will convert it to the new SST-based format on the first run.

5.12. JPA Cache Store

The implementation depends on JPA 2.0 specification to access entity meta model.

In normal use cases, it’s recommended to leverage Infinispan for JPA second level cache and/or query cache. However, if you’d like to use only Infinispan API and you want Infinispan to persist into a cache store using a common format (e.g., a database with well defined schema), then JPA Cache Store could be right for you.

Things to note
  • When using JPA Cache Store, the key should be the ID of the entity, while the value should be the entity object.

  • Only a single @Id or @EmbeddedId annotated property is allowed.

  • Auto-generated ID is not supported.

  • Lastly, all entries will be stored as immortal entries.

The JPA cache store does not support segmentation.

5.12.1. Sample Usage

For example, given a persistence unit "myPersistenceUnit", and a JPA entity User:

<persistence-unit name="myPersistenceUnit">

User entity class

public class User implements Serializable {
        private String username;
        private String firstName;
        private String lastName;


Then you can configure a cache "usersCache" to use JPA Cache Store, so that when you put data into the cache, the data would be persisted into the database based on JPA configuration.

EmbeddedCacheManager cacheManager = ...;

Configuration cacheConfig = new ConfigurationBuilder().persistence()
cacheManager.defineCache("usersCache", cacheConfig);

Cache<String, User> usersCache = cacheManager.getCache("usersCache");
usersCache.put("raytsang", new User(...));

Normally a single Infinispan cache can store multiple types of key/value pairs, for example:

Cache<String, User> usersCache = cacheManager.getCache("myCache");
usersCache.put("raytsang", new User());
Cache<Integer, Teacher> teachersCache = cacheManager.getCache("myCache");
teachersCache.put(1, new Teacher());

It’s important to note that, when a cache is configured to use a JPA Cache Store, that cache would only be able to store ONE type of data.

Cache<String, User> usersCache = cacheManager.getCache("myJPACache"); // configured for User entity class
usersCache.put("raytsang", new User());
Cache<Integer, Teacher> teachersCache = cacheManager.getCache("myJPACache"); // cannot do this when this cache is configured to use a JPA cache store
teachersCache.put(1, new Teacher());

Use of @EmbeddedId is supported so that you can also use composite keys.

public class Vehicle implements Serializable {
        private VehicleId id;
        private String color;        ...

public class VehicleId implements Serializable
        private String state;
        private String licensePlate;

Lastly, auto-generated IDs (e.g., @GeneratedValue) is not supported. When putting things into the cache with a JPA cache store, the key should be the ID value!

5.12.2. Configuration

Sample Programatic Configuration
Configuration cacheConfig = new ConfigurationBuilder().persistence()
Parameter Description


JPA persistence unit name in JPA configuration (persistence.xml) that contains the JPA entity class


JPA entity class that is expected to be stored in this cache. Only one class is allowed.

Sample XML Configuration
<local-cache name="vehicleCache">
   <persistence passivation="false">
      <jpa-store xmlns="urn:infinispan:config:store:jpa:7.0"
Parameter Description


JPA persistence unit name in JPA configuration (persistence.xml) that contains the JPA entity class


Fully qualified JPA entity class name that is expected to be stored in this cache. Only one class is allowed.

5.12.3. Additional References

Refer to the test case for code samples in action.

Refer to test configurations for configuration samples.

5.13. Custom Cache Stores

If the provided cache stores do not fulfill all of your requirements, it is possible for you to implement your own store. The steps required to create your own store are as follows:

  1. Write your custom store by implementing one of the following interfaces:

    • org.infinispan.persistence.spi.AdvancedCacheWriter

    • org.infinispan.persistence.spi.AdvancedCacheLoader

    • org.infinispan.persistence.spi.CacheLoader

    • org.infinispan.persistence.spi.CacheWriter

    • org.infinispan.persistence.spi.ExternalStore

    • org.infinispan.persistence.spi.AdvancedLoadWriteStore

    • org.infinispan.persistence.spi.TransactionalCacheWriter

    • org.infinispan.persistence.spi.SegmentedAdvancedLoadWriteStore

  2. Annotate your store class with the @Store annotation and specify the properties relevant to your store, e.g. is it possible for the store to be shared in Replicated or Distributed mode: @Store(shared = true).

  3. Create a custom cache store configuration and builder. This requires extending AbstractStoreConfiguration and AbstractStoreConfigurationBuilder. As an optional step, you should add the following annotations to your configuration - @ConfigurationFor, @BuiltBy as well as adding @ConfiguredBy to your store implementation class. These additional annotations will ensure that your custom configuration builder is used to parse your store configuration from xml. If these annotations are not added, then the CustomStoreConfigurationBuilder will be used to parse the common store attributes defined in AbstractStoreConfiguration and any additional elements will be ignored. If a store and its configuration do not declare the @Store and @ConfigurationFor annotations respectively, a warning message will be logged upon cache initialisation.

    If you wish for your store to be segmented, where it will craete a different store instance per segment, instead of extending AbstractStoreConfiguration you should extend AbstractSegmentedStoreConfiguration.

  4. Add your custom store to your cache’s configuration:

    1. Add your custom store to the ConfigurationBuilder, for example:

      Configuration config = new ConfigurationBuilder()
    2. Define your custom store via xml:

      <local-cache name="customStoreExample">
          <store class="org.infinispan.persistence.dummy.DummyInMemoryStore" />

5.13.1. HotRod Deployment

A Custom Cache Store can be packaged into a separate JAR file and deployed in a HotRod server using the following steps:

  1. Follow Custom Cache Stores, steps 1-3>> in the previous section and package your implementations in a JAR file (or use a Custom Cache Store Archetype).

  2. In your Jar create a proper file under META-INF/services/, which contains the fully qualified class name of your store implementation. The name of this service file should reflect the interface that your store implements. For example, if your store implements the AdvancedCacheWriter interface than you need to create the following file:

    • /META-INF/services/org.infinispan.persistence.spi.AdvancedCacheWriter

  3. Deploy the JAR file in the Infinispan Server.

5.14. Store Migrator

Infinispan 9.0 introduced changes to internal marshalling functionality that are not backwardly compatible with previous versions of Infinispan. As a result, Infinispan 9.x and later cannot read cache stores created in earlier versions of Infinispan. Additionally, Infinispan no longer provides some store implementations such as JDBC Mixed and Binary stores.

You can use StoreMigrator.java to migrate cache stores. This migration tool reads data from cache stores in previous versions and rewrites the content for compatibility with the current marshalling implementation.

5.14.1. Migrating Cache Stores

To perform a migration with StoreMigrator,

  1. Put infinispan-tools-9.4.jar and dependencies for your source and target databases, such as JDBC drivers, on your classpath.

  2. Create a .properties file that contains configuration properties for the source and target cache stores.

    You can find an example properties file that contains all applicable configuration options in migrator.properties.

  3. Specify .properties file as an argument for StoreMigrator.

  4. Run mvn exec:java to execute the migrator.

See the following example Maven pom.xml for StoreMigrator:

<?xml version="1.0" encoding="UTF-8"?>
<project xmlns="http://maven.apache.org/POM/4.0.0"
         xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">




                        <argument><!-- PATH TO YOUR MIGRATOR.PROPERTIES FILE --></argument>

Replace ${version.infinispan} with the appropriate version of Infinispan.

5.14.2. Store Migrator Properties

All migrator properties are configured within the context of a source or target store. Each property must start with either source. or target..

All properties in the following sections apply to both source and target stores, except for table.binary.* properties because it is not possible to migrate to a binary table.

Common Properties
Property Description Example value Required






The name of the cache associated with the store




How many segments this store will be created with. If not provided store will not be segmented. (supported as target only - JDBC not yet supported)



It should be noted that the segment_count property should match how many segments your cache will be using. That is that it should match the clustering.hash.numSegments config value. If these do not match, data will not be properly read when running the cache.

JDBC Properties
Property Description Example value Required


The dialect of the underlying database




The marshaller to use for the store. Possible values are:

- LEGACY Infinispan 8.2.x marshaller. Valid for source stores only.

- CURRENT Infinispan 9.x marshaller.

- CUSTOM Custom marshaller.




The class of the marshaller if type=CUSTOM



A comma-separated list of custom AdvancedExternalizer implementations to load [id]:<Externalizer class>



The JDBC connection url




The class of the JDBC driver




Database username



Database password



Database major version



Database minor version



Disable db upsert



Prevent table index being created



Additional prefix for table name



Name of the column




Type of the column




TwoWayKey2StringMapper Class

org.infinispan.persistence.keymappers. DefaultTwoWayKey2StringMapper

LevelDB/RocksDB Properties
Property Description Example value Required


The location of the db directory




The compression type to be used


SingleFileStore Properties
Property Description Example value Required


The directory containing the store’s .dat file



SoftIndexFileStore Properties
Property Description Example value Required


The location of the db directory




The location of the db’s index


Target Only

Following sections describe the SPI and also discuss the SPI implementations that Infinispan ships out of the box.

5.15. SPI

The following class diagram presents the main SPI interfaces of the persistence API:

Figure2 1 persistence API
Figure 1. Persistence SPI

Some notes about the classes:

  • ByteBuffer - abstracts the serialized form of an object

  • MarshalledEntry - abstracts the information held within a persistent store corresponding to a key-value added to the cache. Provides method for reading this information both in serialized (ByteBuffer) and deserialized (Object) format. Normally data read from the store is kept in serialized format and lazily deserialized on demand, within the MarshalledEntry implementation

  • CacheWriter and CacheLoader provide basic methods for reading and writing to a store

  • AdvancedCacheLoader and AdvancedCacheWriter provide operations to manipulate the underlaying storage in bulk: parallel iteration and purging of expired entries, clear and size.

A cache store can be segmented if it does one of the following:

  • Implements the SegmentedAdvancedLoadWriteStore interface. In this case only a single store instance is used per cache.

  • Has a configuration that extends the AbstractSegmentedConfiguration abstract class. Doing this requires you to implement the newConfigurationFrom method where it is expected that a new StoreConfiguration instance is created per invocation. This creates a store instance per segment to which a node can write. Stores might start and stop as data is moved between nodes.

A provider might choose to only implement a subset of these interfaces:

  • Not implementing the AdvancedCacheWriter makes the given writer not usable for purging expired entries or clear

  • If a loader does not implement the AdvancedCacheLoader inteface, then it will not participate in preloading nor in cache iteration (required also for stream operations).

If you’re looking at migrating your existing store to the new API or to write a new store implementation, the SingleFileStore might be a good starting point/example.

5.16. More implementations

Many more cache loader and cache store implementations exist. Visit this website for more details.

6. Clustering

A cache manager can be configured to be either local (standalone) or clustered. When clustered, manager instances use JGroups' discovery protocols to automatically discover neighboring instances on the same local network and form a cluster.

Creating a local-only cache manager is trivial: just use the no-argument DefaultCacheManager constructor, or supply the following XML configuration file.


To start a clustered cache manager, you need to create a clustered configuration.

GlobalConfigurationBuilder gcb = GlobalConfigurationBuilder.defaultClusteredBuilder();
DefaultCacheManager manager = new DefaultCacheManager(gcb.build());

Individual caches can then be configured in different modes:

  • Local: changes and reads are never replicated. This is the only mode available in non-clustered cache managers.

  • Invalidation: changes are not replicated, instead the key is invalidated on all nodes; reads are local.

  • Replicated: changes are replicated to all nodes, reads are always local.

  • Distributed: changes are replicated to a fixed number of nodes, reads request the value from at least one of the owner nodes.

6.1. Which cache mode should I use?

Which cache you should use depends on the qualities/guarantees you need for your data. The following table summarizes the most important ones:

Simple Local Invalidation Replicated Distributed Scattered








Read performance







Write performance



(all nodes, no data)

(all nodes)

(owner nodes)

(single RPC)


Single node

Single node

Single node

Smallest node




Single node

Single node

Single node

All nodes

Owner nodes

Owner nodes


No TX, persistence, indexing






6.2. Local Mode

While Infinispan is particularly interesting in clustered mode, it also offers a very capable local mode. In this mode, it acts as a simple, in-memory data cache similar to a ConcurrentHashMap.

But why would one use a local cache rather than a map? Caches offer a lot of features over and above a simple map, including write-through and write-behind to a persistent store, eviction of entries to prevent running out of memory, and expiration.

Infinispan’s Cache interface extends JDK’s ConcurrentMap — making migration from a map to Infinispan trivial.

Infinispan caches also support transactions, either integrating with an existing transaction manager or running a separate one. Local caches transactions have two choices:

  1. When to lock? Pessimistic locking locks keys on a write operation or when the user calls AdvancedCache.lock(keys) explicitly. Optimistic locking only locks keys during the transaction commit, and instead it throws a WriteSkewCheckException at commit time, if another transaction modified the same keys after the current transaction read them.

  2. Isolation level. We support read-committed and repeatable read.

6.2.1. Simple Cache

Traditional local caches use the same architecture as clustered caches, i.e. they use the interceptor stack. That way a lot of the implementation can be reused. However, if the advanced features are not needed and performance is more important, the interceptor stack can be stripped away and simple cache can be used.

So, which features are stripped away? From the configuration perspective, simple cache does not support:

  • transactions and invocation batching

  • persistence (cache stores and loaders)

  • custom interceptors (there’s no interceptor stack!)

  • indexing

  • compatibility (embedded/server mode)

  • store as binary (which is hardly useful for local caches)

From the API perspective these features throw an exception:

  • adding custom interceptors

  • Distributed Executors Framework

So, what’s left?

  • basic map-like API

  • cache listeners (local ones)

  • expiration

  • eviction

  • security

  • JMX access

  • statistics (though for max performance it is recommended to switch this off using statistics-available=false)

Declarative configuration
   <local-cache name="mySimpleCache" simple-cache="true">
        <!-- expiration, eviction, security... -->
Programmatic configuration
CacheManager cm = getCacheManager();
ConfigurationBuilder builder = new ConfigurationBuilder().simpleCache(true);
cm.defineConfiguration("mySimpleCache", builder.build());
Cache cache = cm.getCache("mySimpleCache");

Simple cache checks against features it does not support, if you configure it to use e.g. transactions, configuration validation will throw an exception.

6.3. Invalidation Mode

In invalidation, the caches on different nodes do not actually share any data. Instead, when a key is written to, the cache only aims to remove data that may be stale from other nodes. This cache mode only makes sense if you have another, permanent store for your data such as a database and are only using Infinispan as an optimization in a read-heavy system, to prevent hitting the database for every read. If a cache is configured for invalidation, every time data is changed in a cache, other caches in the cluster receive a message informing them that their data is now stale and should be removed from memory and from any local store.

clustering invalidation
Figure 2. Invalidation mode

Sometimes the application reads a value from the external store and wants to write it to the local cache, without removing it from the other nodes. To do this, it must call Cache.putForExternalRead(key, value) instead of Cache.put(key, value).

Invalidation mode can be used with a shared cache store. A write operation will both update the shared store, and it would remove the stale values from the other nodes' memory. The benefit of this is twofold: network traffic is minimized as invalidation messages are very small compared to replicating the entire value, and also other caches in the cluster look up modified data in a lazy manner, only when needed.

Never use invalidation mode with a local store. The invalidation message will not remove entries in the local store, and some nodes will keep seeing the stale value.

An invalidation cache can also be configured with a special cache loader, ClusterLoader. When ClusterLoader is enabled, read operations that do not find the key on the local node will request it from all the other nodes first, and store it in memory locally. In certain situation it will store stale values, so only use it if you have a high tolerance for stale values.

Invalidation mode can be synchronous or asynchronous. When synchronous, a write blocks until all nodes in the cluster have evicted the stale value. When asynchronous, the originator broadcasts invalidation messages but doesn’t wait for responses. That means other nodes still see the stale value for a while after the write completed on the originator.

Transactions can be used to batch the invalidation messages. Transactions acquire the key lock on the primary owner. To find more about how primary owners are assigned, please read the Key Ownership section.

  • With pessimistic locking, each write triggers a lock message, which is broadcast to all the nodes. During transaction commit, the originator broadcasts a one-phase prepare message (optionally fire-and-forget) which invalidates all affected keys and releases the locks.

  • With optimistic locking, the originator broadcasts a prepare message, a commit message, and an unlock message (optional). Either the one-phase prepare or the unlock message is fire-and-forget, and the last message always releases the locks.

6.4. Replicated Mode

Entries written to a replicated cache on any node will be replicated to all other nodes in the cluster, and can be retrieved locally from any node. Replicated mode provides a quick and easy way to share state across a cluster, however replication practically only performs well in small clusters (under 10 nodes), due to the number of messages needed for a write scaling linearly with the cluster size. Infinispan can be configured to use UDP multicast, which mitigates this problem to some degree.

Each key has a primary owner, which serializes data container updates in order to provide consistency. To find more about how primary owners are assigned, please read the Key Ownership section.

clustering replicated
Figure 3. Replicated mode

Replicated mode can be synchronous or asynchronous.

  • Synchronous replication blocks the caller (e.g. on a cache.put(key, value)) until the modifications have been replicated successfully to all the nodes in the cluster.

  • Asynchronous replication performs replication in the background, and write operations return immediately. Asynchronous replication is not recommended, because communication errors, or errors that happen on remote nodes are not reported to the caller.

If transactions are enabled, write operations are not replicated through the primary owner.

  • With pessimistic locking, each write triggers a lock message, which is broadcast to all the nodes. During transaction commit, the originator broadcasts a one-phase prepare message and an unlock message (optional). Either the one-phase prepare or the unlock message is fire-and-forget.

  • With optimistic locking, the originator broadcasts a prepare message, a commit message, and an unlock message (optional). Again, either the one-phase prepare or the unlock message is fire-and-forget.

6.5. Distribution Mode

Distribution tries to keep a fixed number of copies of any entry in the cache, configured as numOwners. This allows the cache to scale linearly, storing more data as nodes are added to the cluster.

As nodes join and leave the cluster, there will be times when a key has more or less than numOwners copies. In particular, if numOwners nodes leave in quick succession, some entries will be lost, so we say that a distributed cache tolerates numOwners - 1 node failures.

The number of copies represents a trade-off between performance and durability of data. The more copies you maintain, the lower performance will be, but also the lower the risk of losing data due to server or network failures. Regardless of how many copies are maintained, distribution still scales linearly, and this is key to Infinispan’s scalability.

The owners of a key are split into one primary owner, which coordinates writes to the key, and zero or more backup owners. To find more about how primary and backup owners are assigned, please read the Key Ownership section.

clustering distributed
Figure 4. Distributed mode

A read operation will request the value from the primary owner, but if it doesn’t respond in a reasonable amount of time, we request the value from the backup owners as well. (The infinispan.stagger.delay system property, in milliseconds, controls the delay between requests.) A read operation may require 0 messages if the key is present in the local cache, or up to 2 * numOwners messages if all the owners are slow.

A write operation will also result in at most 2 * numOwners messages: one message from the originator to the primary owner, numOwners - 1 messages from the primary to the backups, and the corresponding ACK messages.

Cache topology changes may cause retries and additional messages, both for reads and for writes.

Just as replicated mode, distributed mode can also be synchronous or asynchronous. And as in replicated mode, asynchronous replication is not recommended because it can lose updates. In addition to losing updates, asynchronous distributed caches can also see a stale value when a thread writes to a key and then immediately reads the same key.

Transactional distributed caches use the same kinds of messages as transactional replicated caches, except lock/prepare/commit/unlock messages are sent only to the affected nodes (all the nodes that own at least one key affected by the transaction) instead of being broadcast to all the nodes in the cluster. As an optimization, if the transaction writes to a single key and the originator is the primary owner of the key, lock messages are not replicated.

6.5.1. Read consistency

Even with synchronous replication, distributed caches are not linearizable. (For transactional caches, we say they do not support serialization/snapshot isolation.) We can have one thread doing a single put:

cache.get(k) -> v1
cache.put(k, v2)
cache.get(k) -> v2

But another thread might see the values in a different order:

cache.get(k) -> v2
cache.get(k) -> v1

The reason is that read can return the value from any owner, depending on how fast the primary owner replies. The write is not atomic across all the owners — in fact, the primary commits the update only after it receives a confirmation from the backup. While the primary is waiting for the confirmation message from the backup, reads from the backup will see the new value, but reads from the primary will see the old one.

6.5.2. Key Ownership

Distributed caches split entries into a fixed number of segments and assign each segment to a list of owner nodes. Replicated caches do the same, with the exception that every node is an owner.

The first node in the list of owners is the primary owner. The other nodes in the list are backup owners. When the cache topology changes, because a node joins or leaves the cluster, the segment ownership table is broadcast to every node. This allows nodes to locate keys without making multicast requests or maintaining metadata for each key.

The numSegments property configures the number of segments available. However, the number of segments cannot change unless the cluster is restarted.

Likewise the key-to-segment mapping cannot change. Keys must always map to the same segments regardless of cluster topology changes. It is important that the key-to-segment mapping evenly distributes the number of segments allocated to each node while minimizing the number of segments that must move when the cluster topology changes.

You can customize the key-to-segment mapping by configuring a KeyPartitioner or by using the Grouping API.

However, Infinispan provides the following implementations:


Uses an algorithm based on consistent hashing. Selected by default when server hinting is disabled.

This implementation always assigns keys to the same nodes in every cache as long as the cluster is symmetric. In other words, all caches run on all nodes. This implementation does have some negative points in that the load distribution is slightly uneven. It also moves more segments than strictly necessary on a join or leave.


Similar to SyncConsistentHashFactory, but adapted for Server Hinting. Selected by default when server hinting is enabled.


Achieves a more even distribution than SyncConsistentHashFactory, but with one disadvantage. The order in which nodes join the cluster determines which nodes own which segments. As a result, keys might be assigned to different nodes in different caches.

Was the default from version 5.2 to version 8.1 with server hinting disabled.


Similar to DefaultConsistentHashFactory, but adapted for Server Hinting.

Was the default from version 5.2 to version 8.1 with server hinting enabled.


Used internally to implement replicated caches. You should never explicitly select this algorithm in a distributed cache.

Capacity Factors

Capacity factors allocate segment-to-node mappings based on resources available to nodes.

To configure capacity factors, you specify any non-negative number and the Infinispan hashing algorithm assigns each node a load weighted by its capacity factor (both as a primary owner and as a backup owner).

For example, nodeA has 2x the memory available than nodeB in the same Infinispan cluster. In this case, setting capacityFactor to a value of 2 configures Infinispan to allocate 2x the number of segments to nodeA.

Setting a capacity factor of 0 is possible but is recommended only in cases where nodes are not joined to the cluster long enough to be useful data owners.

Zero Capacity Node

You might need to configure a whole node where the capacity factor is 0 for every cache, user defined caches and internal caches. When defining a zero capacity node, the node won’t hold any data. This is how you declare a zero capacity node:

   <cache-container zero-capacity-node="true" />
new GlobalConfigurationBuilder().zeroCapacityNode(true);

However, note that this will be true for distributed caches only. If you are using replicated caches, the node will still keep a copy of the value. Use only distributed caches to make the best use of this feature.

Hashing Configuration

This is how you configure hashing declaratively, via XML:

   <distributed-cache name="distributedCache" owners="2" segments="100" capacity-factor="2" />

And this is how you can configure it programmatically, in Java:

Configuration c = new ConfigurationBuilder()

6.5.3. Initial cluster size

Infinispan’s very dynamic nature in handling topology changes (i.e. nodes being added / removed at runtime) means that, normally, a node doesn’t wait for the presence of other nodes before starting. While this is very flexible, it might not be suitable for applications which require a specific number of nodes to join the cluster before caches are started. For this reason, you can specify how many nodes should have joined the cluster before proceeding with cache initialization. To do this, use the initialClusterSize and initialClusterTimeout transport properties. The declarative XML configuration:

   <transport initial-cluster-size="4" initial-cluster-timeout="30000" />

The programmatic Java configuration:

GlobalConfiguration global = new GlobalConfigurationBuilder()

The above configuration will wait for 4 nodes to join the cluster before initialization. If the initial nodes do not appear within the specified timeout, the cache manager will fail to start.

6.5.4. L1 Caching

When L1 is enabled, a node will keep the result of remote reads locally for a short period of time (configurable, 10 minutes by default), and repeated lookups will return the local L1 value instead of asking the owners again.

clustering l1
Figure 5. L1 caching

L1 caching is not free though. Enabling it comes at a cost, and this cost is that every entry update must broadcast an invalidation message to all the nodes. L1 entries can be evicted just like any other entry when the the cache is configured with a maximum size. Enabling L1 will improve performance for repeated reads of non-local keys, but it will slow down writes and it will increase memory consumption to some degree.

Is L1 caching right for you? The correct approach is to benchmark your application with and without L1 enabled and see what works best for your access pattern.

6.5.5. Server Hinting

The following topology hints can be specified:


This is probably the most useful, when multiple JVM instances run on the same node, or even when multiple virtual machines run on the same physical machine.


In larger clusters, nodes located on the same rack are more likely to experience a hardware or network failure at the same time.


Some clusters may have nodes in multiple physical locations for extra resilience. Note that Cross site replication is another alternative for clusters that need to span two or more data centres.

All of the above are optional. When provided, the distribution algorithm will try to spread the ownership of each segment across as many sites, racks, and machines (in this order) as possible.


The hints are configured at transport level:

    site="US-WestCoast" />

6.5.6. Key affinity service

In a distributed cache, a key is allocated to a list of nodes with an opaque algorithm. There is no easy way to reverse the computation and generate a key that maps to a particular node. However, we can generate a sequence of (pseudo-)random keys, see what their primary owner is, and hand them out to the application when it needs a key mapping to a particular node.


Following code snippet depicts how a reference to this service can be obtained and used.

// 1. Obtain a reference to a cache
Cache cache = ...
Address address = cache.getCacheManager().getAddress();

// 2. Create the affinity service
KeyAffinityService keyAffinityService = KeyAffinityServiceFactory.newLocalKeyAffinityService(
      new RndKeyGenerator(),

// 3. Obtain a key for which the local node is the primary owner
Object localKey = keyAffinityService.getKeyForAddress(address);

// 4. Insert the key in the cache
cache.put(localKey, "yourValue");

The service is started at step 2: after this point it uses the supplied Executor to generate and queue keys. At step 3, we obtain a key from the service, and at step 4 we use it.


KeyAffinityService extends Lifecycle, which allows stopping and (re)starting it:

public interface Lifecycle {
   void start();
   void stop();

The service is instantiated through KeyAffinityServiceFactory. All the factory methods have an Executor parameter, that is used for asynchronous key generation (so that it won’t happen in the caller’s thread). It is the user’s responsibility to handle the shutdown of this Executor.

The KeyAffinityService, once started, needs to be explicitly stopped. This stops the background key generation and releases other held resources.

The only situation in which KeyAffinityService stops by itself is when the cache manager with which it was registered is shutdown.

Topology changes

When the cache topology changes (i.e. nodes join or leave the cluster), the ownership of the keys generated by the KeyAffinityService might change. The key affinity service keep tracks of these topology changes and doesn’t return keys that would currently map to a different node, but it won’t do anything about keys generated earlier.

As such, applications should treat KeyAffinityService purely as an optimization, and they should not rely on the location of a generated key for correctness.

In particular, applications should not rely on keys generated by KeyAffinityService for the same address to always be located together. Collocation of keys is only provided by the Grouping API.

6.5.7. The Grouping API

Complementary to Key affinity service and similar to AtomicMap, the grouping API allows you to co-locate a group of entries on the same nodes, but without being able to select the actual nodes.

How does it work?

By default, the segment of a key is computed using the key’s hashCode(). If you use the grouping API, Infinispan will compute the segment of the group and use that as the segment of the key. See the Key Ownership section for more details on how segments are then mapped to nodes.

When the group API is in use, it is important that every node can still compute the owners of every key without contacting other nodes. For this reason, the group cannot be specified manually. The group can either be intrinsic to the entry (generated by the key class) or extrinsic (generated by an external function).

How do I use the grouping API?

First, you must enable groups. If you are configuring Infinispan programmatically, then call:

Configuration c = new ConfigurationBuilder()

Or, if you are using XML:

   <groups enabled="true"/>

If you have control of the key class (you can alter the class definition, it’s not part of an unmodifiable library), then we recommend using an intrinsic group. The intrinsic group is specified by adding the @Group annotation to a method. Let’s take a look at an example:

class User {
   String office;

   public int hashCode() {
      // Defines the hash for the key, normally used to determine location

   // Override the location by specifying a group
   // All keys in the same group end up with the same owners
   public String getOffice() {
      return office;
The group method must return a String

If you don’t have control over the key class, or the determination of the group is an orthogonal concern to the key class, we recommend using an extrinsic group. An extrinsic group is specified by implementing the Grouper interface.

public interface Grouper<T> {
    String computeGroup(T key, String group);

    Class<T> getKeyType();

If multiple Grouper classes are configured for the same key type, all of them will be called, receiving the value computed by the previous one. If the key class also has a @Group annotation, the first Grouper will receive the group computed by the annotated method. This allows you even greater control over the group when using an intrinsic group. Let’s take a look at an example Grouper implementation:

public class KXGrouper implements Grouper<String> {

   // The pattern requires a String key, of length 2, where the first character is
   // "k" and the second character is a digit. We take that digit, and perform
   // modular arithmetic on it to assign it to group "0" or group "1".
   private static Pattern kPattern = Pattern.compile("(^k)(<a>\\d</a>)$");

   public String computeGroup(String key, String group) {
      Matcher matcher = kPattern.matcher(key);
      if (matcher.matches()) {
         String g = Integer.parseInt(matcher.group(2)) % 2 + "";
         return g;
      } else {
         return null;

   public Class<String> getKeyType() {
      return String.class;

Grouper implementations must be registered explicitly in the cache configuration. If you are configuring Infinispan programmatically:

Configuration c = new ConfigurationBuilder()
   .clustering().hash().groups().enabled().addGrouper(new KXGrouper())

Or, if you are using XML:

   <groups enabled="true">
      <grouper class="com.acme.KXGrouper" />
Advanced Interface

AdvancedCache has two group-specific methods:


Retrieves all keys in the cache that belong to a group.


Removes all the keys in the cache that belong to a group.

Both methods iterate over the entire data container and store (if present), so they can be slow when a cache contains lots of small groups.

6.6. Scattered Mode

Scattered mode is very similar to Distribution Mode as it allows linear scaling of the cluster. It allows single node failure by maintaining two copies of the data (as Distribution Mode with numOwners=2). Unlike Distributed, the location of data is not fixed; while we use the same Consistent Hash algorithm to locate the primary owner, the backup copy is stored on the node that wrote the data last time. When the write originates on the primary owner, backup copy is stored on any other node (the exact location of this copy is not important).

This has the advantage of single RPC for any write (Distribution Mode requires one or two RPCs), but reads have to always target the primary owner. That results in faster writes but possibly slower reads, and therefore this mode is more suitable for write-intensive applications.

Storing multiple backup copies also results in slightly higher memory consumption. In order to remove out-of-date backup copies, invalidation messages are broadcast in the cluster, which generates some overhead. This makes scattered mode less performant in very big clusters (this behaviour might be optimized in the future).

When a node crashes, the primary copy may be lost. Therefore, the cluster has to reconcile the backups and find out the last written backup copy. This process results in more network traffic during state transfer.

Since the writer of data is also a backup, even if we specify machine/rack/site ids on the transport level the cluster cannot be resilient to more than one failure on the same machine/rack/site.

Currently it is not possible to use scattered mode in transactional cache. Asynchronous replication is not supported either; use asynchronous Cache API instead. Functional commands are not implemented neither but these are expected to be added soon.

The cache is configured in a similar way as the other cache modes, here is an example of declarative configuration:

   <scattered-cache name="scatteredCache" />

And this is how you can configure it programmatically:

Configuration c = new ConfigurationBuilder()

Scattered mode is not exposed in the server configuration as the server is usually accessed through the Hot Rod protocol. The protocol automatically selects primary owner for the writes and therefore the write (in distributed mode with two owner) requires single RPC inside the cluster, too. Therefore, scattered cache would not bring the performance benefit.

6.7. Asynchronous Options

6.7.1. Asynchronous Communications

All clustered cache modes can be configured to use asynchronous communications with the mode="ASYNC" attribute on the <replicated-cache/>, <distributed-cache>, or <invalidation-cache/> element.

With asynchronous communications, the originator node does not receive any acknowledgement from the other nodes about the status of the operation, so there is no way to check if it succeeded on other nodes.

We do not recommend asynchronous communications in general, as they can cause inconsistencies in the data, and the results are hard to reason about. Nevertheless, sometimes speed is more important than consistency, and the option is available for those cases.

6.7.2. Asynchronous API

The Asynchronous API allows you to use synchronous communications, but without blocking the user thread.

There is one caveat: The asynchronous operations do NOT preserve the program order. If a thread calls cache.putAsync(k, v1); cache.putAsync(k, v2), the final value of k may be either v1 or v2. The advantage over using asynchronous communications is that the final value can’t be v1 on one node and v2 on another.

Prior to version 9.0, the asynchronous API was emulated by borrowing a thread from an internal thread pool and running a blocking operation on that thread.

6.7.3. Return Values

Because the Cache interface extends java.util.Map, write methods like put(key, value) and remove(key) return the previous value by default.

In some cases, the return value may not be correct:

  1. When using AdvancedCache.withFlags() with Flag.IGNORE_RETURN_VALUE, Flag.SKIP_REMOTE_LOOKUP, or Flag.SKIP_CACHE_LOAD.

  2. When the cache is configured with unreliable-return-values="true".

  3. When using asynchronous communications.

  4. When there are multiple concurrent writes to the same key, and the cache topology changes. The topology change will make Infinispan retry the write operations, and a retried operation’s return value is not reliable.

Transactional caches return the correct previous value in cases 3 and 4. However, transactional caches also have a gotcha: in distributed mode, the read-committed isolation level is implemented as repeatable-read. That means this example of "double-checked locking" won’t work:

Cache cache = ...
TransactionManager tm = ...

try {
   Integer v1 = cache.get(k);
   // Increment the value
   Integer v2 = cache.put(k, v1 + 1);
   if (Objects.equals(v1, v2) {
      // success
   } else {
      // retry
} finally {

The correct way to implement this is to use cache.getAdvancedCache().withFlags(Flag.FORCE_WRITE_LOCK).get(k).

In caches with optimistic locking, writes can also return stale previous values. Write skew checks can avoid stale previous values. For more information, see Write Skews.

6.8. Partition handling

An Infinispan cluster is built out of a number of nodes where data is stored. In order not to lose data in the presence of node failures, Infinispan copies the same data — cache entry in Infinispan parlance — over multiple nodes. This level of data redundancy is configured through the numOwners configuration attribute and ensures that as long as fewer than numOwners nodes crash simultaneously, Infinispan has a copy of the data available.

However, there might be catastrophic situations in which more than numOwners nodes disappear from the cluster:

Split brain

Caused e.g. by a router crash, this splits the cluster in two or more partitions, or sub-clusters that operate independently. In these circumstances, multiple clients reading/writing from different partitions see different versions of the same cache entry, which for many application is problematic. Note there are ways to alleviate the possibility for the split brain to happen, such as redundant networks or IP bonding. These only reduce the window of time for the problem to occur, though.

numOwners nodes crash in sequence

When at least numOwners nodes crash in rapid succession and Infinispan does not have the time to properly rebalance its state between crashes, the result is partial data loss.

The partition handling functionality discussed in this section allows the user to configure what operations can be performed on a cache in the event of a split brain occurring. Infinispan provides multiple partition handling strategies, which in terms of Brewer’s CAP theorem determine whether availability or consistency is sacrificed in the presence of partition(s). Below is a list of the provided strategies:

Strategy Description CAP


If the partition does not have all owners for a given segment, both reads and writes are denied for all keys in that segment.



Allows reads for a given key if it exists in this partition, but only allows writes if this partition contains all owners of a segment. This is still a consistent approach because some entries are readable if available in this partition, but from a client application perspective it is not deterministic.



Allow entries on each partition to diverge, with conflict resolution attempted upon the partitions merging.


The requirements of your application should determine which strategy is appropriate. For example, DENY_READ_WRITES is more appropriate for applications that have high consistency requirements; i.e. when the data read from the system must be accurate. Whereas if Infinispan is used as a best-effort cache, partitions maybe perfectly tolerable and the ALLOW_READ_WRITES might be more appropriate as it favours availability over consistency.

The following sections describe how Infinispan handles split brain and successive failures for each of the partition handling strategies. This is followed by a section describing how Infinispan allows for automatic conflict resolution upon partition merges via merge policies. Finally, we provide a section describing how to configure partition handling strategies and merge policies.

6.8.1. Split brain

In a split brain situation, each network partition will install its own JGroups view, removing the nodes from the other partition(s). We don’t have a direct way of determining whether the has been split into two or more partitions, since the partitions are unaware of each other. Instead, we assume the cluster has split when one or more nodes disappear from the JGroups cluster without sending an explicit leave message.

Split Strategies

In this section, we detail how each partition handling strategy behaves in the event of split brain occurring.


Each partition continues to function as an independent cluster, with all partitions remaining in AVAILABLE mode. This means that each partition may only see a part of the data, and each partition could write conflicting updates in the cache. During a partition merge these conflicts are automatically resolved by utilising the ConflictManager and the configured EntryMergePolicy.


When a split is detected each partition does not start a rebalance immediately, but first it checks whether it should enter DEGRADED mode instead:

  • If at least one segment has lost all its owners (meaning at least numOwners nodes left since the last rebalance ended), the partition enters DEGRADED mode.

  • If the partition does not contain a simple majority of the nodes (floor(numNodes/2) + 1) in the latest stable topology, the partition also enters DEGRADED mode.

  • Otherwise the partition keeps functioning normally, and it starts a rebalance.

The stable topology is updated every time a rebalance operation ends and the coordinator determines that another rebalance is not necessary.

These rules ensures that at most one partition stays in AVAILABLE mode, and the other partitions enter DEGRADED mode.

When a partition is in DEGRADED mode, it only allows access to the keys that are wholly owned:

  • Requests (reads and writes) for entries that have all the copies on nodes within this partition are honoured.

  • Requests for entries that are partially or totally owned by nodes that disappeared are rejected with an AvailabilityException.

This guarantees that partitions cannot write different values for the same key (cache is consistent), and also that one partition can not read keys that have been updated in the other partitions (no stale data).

To exemplify, consider the initial cluster M = {A, B, C, D}, configured in distributed mode with numOwners = 2. Further on, consider three keys k1, k2 and k3 (that might exist in the cache or not) such that owners(k1) = {A,B}, owners(k2) = {B,C} and owners(k3) = {C,D}. Then the network splits in two partitions, N1 = {A, B} and N2 = {C, D}, they enter DEGRADED mode and behave like this:

  • on N1, k1 is available for read/write, k2 (partially owned) and k3 (not owned) are not available and accessing them results in an AvailabilityException

  • on N2, k1 and k2 are not available for read/write, k3 is available

A relevant aspect of the partition handling process is the fact that when a split brain happens, the resulting partitions rely on the original segment mapping (the one that existed before the split brain) in order to calculate key ownership. So it doesn’t matter if k1, k2, or k3 already existed cache or not, their availability is the same.

If at a further point in time the network heals and N1 and N2 partitions merge back together into the initial cluster M, then M exits the degraded mode and becomes fully available again. During this merge operation, because M has once again become fully available, the ConflictManager and the configured EntryMergePolicy are used to check for any conflicts that may have occurred in the interim period between the split brain occurring and being detected.

As another example, the cluster could split in two partitions O1 = {A, B, C} and O2 = {D}, partition O1 will stay fully available (rebalancing cache entries on the remaining members). Partition O2, however, will detect a split and enter the degraded mode. Since it doesn’t have any fully owned keys, it will reject any read or write operation with an AvailabilityException.

If afterwards partitions O1 and O2 merge back into M, then the ConflictManager attempts to resolve any conflicts and D once again becomes fully available.


Partitions are handled in the same manner as DENY_READ_WRITES, except that when a partition is in DEGRADED mode read operations on a partially owned key WILL not throw an AvailabilityException.

Current limitations

Two partitions could start up isolated, and as long as they don’t merge they can read and write inconsistent data. In the future, we will allow custom availability strategies (e.g. check that a certain node is part of the cluster, or check that an external machine is accessible) that could handle that situation as well.

6.8.2. Successive nodes stopped

As mentioned in the previous section, Infinispan can’t detect whether a node left the JGroups view because of a process/machine crash, or because of a network failure: whenever a node leaves the JGroups cluster abruptly, it is assumed to be because of a network problem.

If the configured number of copies (numOwners) is greater than 1, the cluster can remain available and will try to make new replicas of the data on the crashed node. However, other nodes might crash during the rebalance process. If more than numOwners nodes crash in a short interval of time, there is a chance that some cache entries have disappeared from the cluster altogether. In this case, with the DENY_READ_WRITES or ALLOW_READS strategy enabled, Infinispan assumes (incorrectly) that there is a split brain and enters DEGRADED mode as described in the split-brain section.

The administrator can also shut down more than numOwners nodes in rapid succession, causing the loss of the data stored only on those nodes. When the administrator shuts down a node gracefully, Infinispan knows that the node can’t come back. However, the cluster doesn’t keep track of how each node left, and the cache still enters DEGRADED mode as if those nodes had crashed.

At this stage there is no way for the cluster to recover its state, except stopping it and repopulating it on restart with the data from an external source. Clusters are expected to be configured with an appropriate numOwners in order to avoid numOwners successive node failures, so this situation should be pretty rare. If the application can handle losing some of the data in the cache, the administrator can force the availability mode back to AVAILABLE via JMX.

6.8.3. Conflict Manager

The conflict manager is a tool that allows users to retrieve all stored replica values for a given key. In addition to allowing users to process a stream of cache entries whose stored replicas have conflicting values. Furthermore, by utilising implementations of the EntryMergePolicy interface it is possible for said conflicts to be resolved automatically.

Detecting Conflicts

Conflicts are detected by retrieving each of the stored values for a given key. The conflict manager retrieves the value stored from each of the key’s write owners defined by the current consistent hash. The .equals method of the stored values is then used to determine whether all values are equal. If all values are equal then no conflicts exist for the key, otherwise a conflict has occurred. Note that null values are returned if no entry exists on a given node, therefore we deem a conflict to have occurred if both a null and non-null value exists for a given key.

Merge Policies

In the event of conflicts arising between one or more replicas of a given CacheEntry, it is necessary for a conflict resolution algorithm to be defined, therefore we provide the EntryMergePolicy interface. This interface consists of a single method, "merge", whose returned CacheEntry is utilised as the "resolved" entry for a given key. When a non-null CacheEntry is returned, this entries value is "put" to all replicas in the cache. However when the merge implementation returns a null value, all replicas associated with the conflicting key are removed from the cache.

The merge method takes two parameters: the "preferredEntry" and "otherEntries". In the context of a partition merge, the preferredEntry is the primary replica of a CacheEntry stored in the partition that contains the most nodes or if partitions are equal the one with the largest topologyId. In the event of overlapping partitions, i.e. a node A is present in the topology of both partitions {A}, {A,B,C}, we pick {A} as the preferred partition as it will have the higher topologId as the other partition’s topology is behind. When a partition merge is not occurring, the "preferredEntry" is simply the primary replica of the CacheEntry. The second parameter, "otherEntries" is simply a list of all other entries associated with the key for which a conflict was detected.

EntryMergePolicy::merge is only called when a conflict has been detected, it is not called if all CacheEntrys are the same.

Currently Infinispan provides the following implementations of EntryMergePolicy:

Policy Description

MergePolicy.NONE (default)

No attempt is made to resolve conflicts. Entries hosted on the minority partition are removed and the nodes in this partition do not hold any data until the rebalance starts. Note, this behaviour is equivalent to prior Infinispan versions which did not support conflict resolution. Note, in this case all changes made to entries hosted on the minority partition are lost, but once the rebalance has finished all entries will be consistent.


Always utilise the "preferredEntry". MergePolicy.NONE is almost equivalent to PREFERRED_ALWAYS, albeit without the performance impact of performing conflict resolution, therefore MergePolicy.NONE should be chosen unless the following scenario is a concern. When utilising the DENY_READ_WRITES or DENY_READ strategy, it is possible for a write operation to only partially complete when the partitions enter DEGRADED mode, resulting in replicas containing inconsistent values. MergePolicy.PREFERRED_ALWAYS will detect said inconsistency and resolve it, whereas with MergePolicy.NONE the CacheEntry replicas will remain inconsistent after the cluster has rebalanced.


Utilise the "preferredEntry" if it is non-null, otherwise utilise the first entry from "otherEntries".


Always remove a key from the cache when a conflict is detected.

Fully qualified class name

The custom implementation for merge will be used Custom merge policy

6.8.4. Usage

During a partition merge the ConflictManager automatically attempts to resolve conflicts utilising the configured EntryMergePolicy, however it is also possible to manually search for/resolve conflicts as required by your application.

The code below shows how to retrieve an EmbeddedCacheManager’s ConflictManager, how to retrieve all versions of a given key and how to check for conflicts across a given cache.

EmbeddedCacheManager manager = new DefaultCacheManager("example-config.xml");
Cache<Integer, String> cache = manager.getCache("testCache");
ConflictManager<Integer, String> crm = ConflictManagerFactory.get(cache.getAdvancedCache());

// Get All Versions of Key
Map<Address, InternalCacheValue<String>> versions = crm.getAllVersions(1);

// Process conflicts stream and perform some operation on the cache
Stream<Map<Address, InternalCacheEntry<Integer, String>>> stream = crm.getConflicts();
stream.forEach(map -> {
   CacheEntry<Object, Object> entry = map.values().iterator().next();
   Object conflictKey = entry.getKey();

// Detect and then resolve conflicts using the configured EntryMergePolicy

// Detect and then resolve conflicts using the passed EntryMergePolicy instance
crm.resolveConflicts((preferredEntry, otherEntries) -> preferredEntry);
Although the ConflictManager::getConflicts stream is processed per entry, the underlying spliterator is in fact lazily-loading cache entries on a per segment basis.

6.8.5. Configuring partition handling

Unless the cache is distributed or replicated, partition handling configuration is ignored. The default partition handling strategy is ALLOW_READ_WRITES and the default EntryMergePolicy is MergePolicies::PREFERRED_ALWAYS.

<distributed-cache name="the-default-cache">
   <partition-handling when-split="ALLOW_READ_WRITES" merge-policy="PREFERRED_NON_NULL"/>

The same can be achieved programmatically:

ConfigurationBuilder dcc = new ConfigurationBuilder();
Implement a custom merge policy

It’s also possible to provide custom implementations of the EntryMergePolicy

<distributed-cache name="the-default-cache">
   <partition-handling when-split="ALLOW_READ_WRITES" merge-policy="org.example.CustomMergePolicy"/>
ConfigurationBuilder dcc = new ConfigurationBuilder();
                    .mergePolicy(new CustomMergePolicy());
public class CustomMergePolicy implements EntryMergePolicy<String, String> {

   public CacheEntry<String, String> merge(CacheEntry<String, String> preferredEntry, List<CacheEntry<String, String>> otherEntries) {
      // decide which entry should be used

      return the_solved_CacheEntry;
Deploy custom merge policies to a Infinispan server instance

To utilise a custom EntryMergePolicy implementation on the server, it’s necessary for the implementation class(es) to be deployed to the server. This is accomplished by utilising the java service-provider convention and packaging the class files in a jar which has a META-INF/services/org.infinispan.conflict.EntryMergePolicy file containing the fully qualified class name of the EntryMergePolicy implementation.

# list all necessary implementations of EntryMergePolicy with the full qualified name

In order for a Custom merge policy to be utilised on the server, you should enable object storage, if your policies semantics require access to the stored Key/Value objects. This is because cache entries in the server may be stored in a marshalled format and the Key/Value objects returned to your policy would be instances of WrappedByteArray. However, if the custom policy only depends on the metadata associated with a cache entry, then object storage is not required and should be avoided (unless needed for other reasons) due to the additional performance cost of marshalling data per request. Finally, object storage is never required if one of the provided merge policies is used.

6.8.6. Monitoring and administration

The availability mode of a cache is exposed in JMX as an attribute in the Cache MBean. The attribute is writable, allowing an administrator to forcefully migrate a cache from DEGRADED mode back to AVAILABLE (at the cost of consistency).

The availability mode is also accessible via the AdvancedCache interface:

AdvancedCache ac = cache.getAdvancedCache();

// Read the availability
boolean available = ac.getAvailability() == AvailabilityMode.AVAILABLE;

// Change the availability
if (!available) {

7. Marshalling

Marshalling is the process of converting Java POJOs into something that can be written in a format that can be transferred over the wire. Unmarshalling is the reverse process whereby data read from a wire format is transformed back into Java POJOs. Infinispan uses marshalling/unmarshalling in order to:

  • Transform data so that it can be send over to other Infinispan nodes in a cluster.

  • Transform data so that it can be stored in underlying cache stores.

  • Store data in Infinispan in a wire format to provide lazy deserialization capabilities.

7.1. The Role Of JBoss Marshalling

Since performance is a very important factor in this process, Infinispan uses JBoss Marshalling framework instead of standard Java Serialization in order to marshall/unmarshall Java POJOs. Amongst other things, this framework enables Infinispan to provide highly efficient ways to marshall internal Infinispan Java POJOs that are constantly used. Apart from providing more efficient ways to marshall Java POJOs, including internal Java classes, JBoss Marshalling uses highly performant java.io.ObjectOutput and java.io.ObjectInput implementations compared to standard java.io.ObjectOutputStream and java.io.ObjectInputStream.

7.2. Support For Non-Serializable Objects

From a users perspective, a very common concern is whether Infinispan supports storing non-Serializable objects. In 4.0, an Infinispan cache instance can only store non-Serializable key or value objects if, and only if:

  • cache is configured to be a local cache and…​

  • cache is not configured with lazy serialization and…​

  • cache is not configured with any write-behind cache store

If either of these options is true, key/value pairs in the cache will need to be marshalled and currently they require to either to extend java.io.Serializable or java.io.Externalizable.

Since Infinispan 5.0, marshalling non-Serializable key/value objects are supported as long as users can to provide meaningful Externalizer implementations for these non-Seralizable objects.

If you’re unable to retrofit Serializable or Externalizable into the classes whose instances are stored in Infinispan, you could alternatively use something like XStream to convert your Non-Serializable objects into a String that can be stored into Infinispan. The one caveat about using XStream is that it slows down the process of storing key/value objects due to the XML transformation that it needs to do.

7.2.1. Store As Binary

Store as binary enables data to be stored in its serialized form. This can be useful to achieve lazy deserialization, which is the mechanism by which Infinispan by which serialization and deserialization of objects is deferred till the point in time in which they are used and needed. This typically means that any deserialization happens using the thread context class loader of the invocation that requires deserialization, and is an effective mechanism to provide classloader isolation. By default lazy deserialization is disabled but if you want to enable it, you can do it like this:

  • Via XML at the Cache level, either under <*-cache /> element:

   <binary />
  • Programmatically:

ConfigurationBuilder builder = ...
Equality Considerations

When using lazy deserialization/storing as binary, keys and values are wrapped as WrappedBytes. It is this wrapper class that transparently takes care of serialization and deserialization on demand, and internally may have a reference to the object itself being wrapped, or the serialized, byte array representation of this object.

This has a particular effect on the behavior of equality. The equals() method of this class will either compare binary representations (byte arrays) or delegate to the wrapped object instance’s equals() method, depending on whether both instances being compared are in serialized or deserialized form at the time of comparison. If one of the instances being compared is in one form and the other in another form, then one instance is either serialized or deserialized.

This will affect the way keys stored in the cache will work, when storeAsBinary is used, since comparisons happen on the key which will be wrapped by a MarshalledValue. Implementers of equals() methods on their keys need to be aware of the behavior of equality comparison, when a key is wrapped as a MarshalledValue, as detailed above.

Store-by-value via defensive copying

The configuration storeAsBinary offers the possibility to enable defensive copying, which allows for store-by-value like behaviour.

Infinispan marshalls objects the moment they’re stored, hence changes made to object references are not stored in the cache, not even for local caches. This provides store-by-value like behaviour. Enabling storeAsBinary can be achieved:

  • Via XML at the Cache level, either under <*-cache /> or <default /> elements:

<store-as-binary keys="true" values="true"/>
  • Programmatically:

ConfigurationBuilder builder = ...

7.3. Advanced Configuration

Internally, Infinispan uses an implementation of this Marshaller interface in order to marshall/unmarshall Java objects so that they’re sent other nodes in the grid, or so that they’re stored in a cache store, or even so to transform them into byte arrays for lazy deserialization.

By default, Infinispan uses the GlobalMarshaller. Optionally, Infinispan users can provide their own marshaller, for example:

  • Via XML at the CacheManager level, under <cache-manager /> element:

<serialization marshaller="com.acme.MyMarshaller"/>
  • Programmatically:

GlobalConfigurationBuilder builder = ...
builder.serialization().marshaller(myMarshaller); // needs an instance of the marshaller

7.3.1. Troubleshooting

Sometimes it might happen that the Infinispan marshalling layer, and in particular JBoss Marshalling, might have issues marshalling/unmarshalling some user object. In Infinispan 4.0, marshalling exceptions will contain further information on the objects that were being marshalled. Example:

java.io.NotSerializableException: java.lang.Object
at org.jboss.marshalling.river.RiverMarshaller.doWriteObject(RiverMarshaller.java:857)
at org.jboss.marshalling.AbstractMarshaller.writeObject(AbstractMarshaller.java:407)
at org.infinispan.marshall.exts.ReplicableCommandExternalizer.writeObject(ReplicableCommandExternalizer.java:54)
at org.infinispan.marshall.jboss.ConstantObjectTable$ExternalizerAdapter.writeObject(ConstantObjectTable.java:267)
at org.jboss.marshalling.river.RiverMarshaller.doWriteObject(RiverMarshaller.java:143)
at org.jboss.marshalling.AbstractMarshaller.writeObject(AbstractMarshaller.java:407)
at org.infinispan.marshall.jboss.JBossMarshaller.objectToObjectStream(JBossMarshaller.java:167)
at org.infinispan.marshall.VersionAwareMarshaller.objectToBuffer(VersionAwareMarshaller.java:92)
at org.infinispan.marshall.VersionAwareMarshaller.objectToByteBuffer(VersionAwareMarshaller.java:170)
at org.infinispan.marshall.DefaultMarshallerTest.testNestedNonSerializable(VersionAwareMarshallerTest.java:415)
Caused by: an exception which occurred:
in object java.lang.Object@b40ec4
in object org.infinispan.commands.write.PutKeyValueCommand@df661da7
... Removed 22 stack frames

The way the "in object" messages are read is the same in which stacktraces are read. The highest "in object" being the most inner one and the lowest "in object" message being the most outer one. So, the above example indicates that a java.lang.Object instance contained in an instance of org.infinispan.commands.write.PutKeyValueCommand could not be serialized because java.lang.Object@b40ec4 is not serializable.

This is not all though! If you enable DEBUG or TRACE logging levels, marshalling exceptions will contain show the toString() representations of objects in the stacktrace. For example:

java.io.NotSerializableException: java.lang.Object
Caused by: an exception which occurred:
in object java.lang.Object@b40ec4
-> toString = java.lang.Object@b40ec4
in object org.infinispan.commands.write.PutKeyValueCommand@df661da7
-> toString = PutKeyValueCommand{key=k, value=java.lang.Object@b40ec4, putIfAbsent=false, lifespanMillis=0, maxIdleTimeMillis=0}

With regards to unmarshalling exceptions, showing such level of information it’s a lot more complicated but where possible. Infinispan will provide class type information. For example:

java.io.IOException: Injected failure!
at org.infinispan.marshall.DefaultMarshallerTest$1.readExternal(VersionAwareMarshallerTest.java:426)
at org.jboss.marshalling.river.RiverUnmarshaller.doReadNewObject(RiverUnmarshaller.java:1172)
at org.jboss.marshalling.river.RiverUnmarshaller.doReadObject(RiverUnmarshaller.java:273)
at org.jboss.marshalling.river.RiverUnmarshaller.doReadObject(RiverUnmarshaller.java:210)
at org.jboss.marshalling.AbstractUnmarshaller.readObject(AbstractUnmarshaller.java:85)
at org.infinispan.marshall.jboss.JBossMarshaller.objectFromObjectStream(JBossMarshaller.java:210)
at org.infinispan.marshall.VersionAwareMarshaller.objectFromByteBuffer(VersionAwareMarshaller.java:104)
at org.infinispan.marshall.VersionAwareMarshaller.objectFromByteBuffer(VersionAwareMarshaller.java:177)
at org.infinispan.marshall.DefaultMarshallerTest.testErrorUnmarshalling(VersionAwareMarshallerTest.java:431)
Caused by: an exception which occurred:
in object of type org.infinispan.marshall.DefaultMarshallerTest$1

In this example, an IOException was thrown when trying to unmarshall a instance of the inner class org.infinispan.marshall.DefaultMarshallerTest$1. In similar fashion to marshalling exceptions, when DEBUG or TRACE logging levels are enabled, classloader information of the class type is provided. For example:

java.io.IOException: Injected failure!
Caused by: an exception which occurred:
in object of type org.infinispan.marshall.DefaultMarshallerTest$1
-> classloader hierarchy:
-> type classloader = sun.misc.Launcher$AppClassLoader@198dfaf
-> parent classloader = sun.misc.Launcher$ExtClassLoader@1858610
... Removed 22 stack frames

Finding the root cause of marshalling/unmarshalling exceptions can sometimes be really daunting but we hope that the above improvements would help get to the bottom of those in a more quicker and efficient manner.

7.4. User Defined Externalizers

One of the key aspects of Infinispan is that it often needs to marshall/unmarshall objects in order to provide some of its functionality. For example, if it needs to store objects in a write-through or write-behind cache store, the stored objects need marshalling. If a cluster of Infinispan nodes is formed, objects shipped around need marshalling. Even if you enable lazy deserialization, objects need to be marshalled so that they can be lazily unmarshalled with the correct classloader.

Using standard JDK serialization is slow and produces payloads that are too big and can affect bandwidth usage. On top of that, JDK serialization does not work well with objects that are supposed to be immutable. In order to avoid these issues, Infinispan uses JBoss Marshalling for marshalling/unmarshalling objects. JBoss Marshalling is fast, produces very space efficient payloads, and on top of that  during unmarshalling, it enables users to have full control over how to construct objects, hence allowing objects to carry on being immutable.

Starting with 5.0, users of Infinispan can now benefit from this marshalling framework as well, and they can provide their own externalizer implementations, but before finding out how to provide externalizers, let’s look at the benefits they bring.

7.4.1. Benefits of Externalizers

The JDK provides a simple way to serialize objects which, in its simplest form, is just a matter of extending java.io.Serializable , but as it’s well known, this is known to be slow and it generates payloads that are far too big. An alternative way to do serialization, still relying on JDK serialization, is for your objects to extend java.io.Externalizable . This allows for users to provide their own ways to marshall/unmarshall classes, but has some serious issues because, on top of relying on slow JDK serialization, it forces the class that you want to serialize to extend this interface, which has two side effects: The first is that you’re forced to modify the source code of the class that you want to marshall/unmarshall which you might not be able to do because you either, don’t own the source, or you don’t even have it. Secondly, since Externalizable implementations do not control object creation, you’re forced to add set methods in order to restore the state, hence potentially forcing your immutable objects to become mutable.

Instead of relying on JDK serialization, Infinispan uses JBoss Marshalling to serialize objects and requires any classes to be serialized to be associated with an Externalizer interface implementation that knows how to transform an object of a particular class into a serialized form and how to read an object of that class from a given input. Infinispan does not force the objects to be serialized to implement Externalizer. In fact, it is recommended that a separate class is used to implement the Externalizer interface because, contrary to JDK serialization, Externalizer implementations control how objects of a particular class are created when trying to read an object from a stream. This means that readObject() implementations are responsible of creating object instances of the target class, hence giving users a lot of flexibility on how to create these instances (whether direct instantiation, via factory or reflection), and more importantly, allows target classes to carry on being immutable. This type of externalizer architecture promotes good OOP designs principles, such as the principle of single responsibility .

It’s quite common, and in general recommended, that Externalizer implementations are stored as inner static public classes within classes that they externalize. The advantages of doing this is that related code stays together, making it easier to maintain. In Infinispan, there are two ways in which Infinispan can be plugged with user defined externalizers:

7.4.2. User Friendly Externalizers

In the simplest possible form, users just need to provide an Externalizer implementation for the type that they want to marshall/unmarshall, and then annotate the marshalled type class with {@link SerializeWith} annotation indicating the externalizer class to use. For example:

import org.infinispan.commons.marshall.Externalizer;
import org.infinispan.commons.marshall.SerializeWith;

public class Person {

   final String name;
   final int age;

   public Person(String name, int age) {
      this.name = name;
      this.age = age;

   public static class PersonExternalizer implements Externalizer<Person> {
      public void writeObject(ObjectOutput output, Person person)
            throws IOException {

      public Person readObject(ObjectInput input)
            throws IOException, ClassNotFoundException {
         return new Person((String) input.readObject(), input.readInt());

At runtime JBoss Marshalling will inspect the object and discover that it is marshallable (thanks to the annotation) and so marshall it using the externalizer class passed. To make externalizer implementations easier to code and more typesafe, make sure you define type <T> as the type of object that’s being marshalled/unmarshalled.

Even though this way of defining externalizers is very user friendly, it has some disadvantages:

  • Due to several constraints of the model, such as support for different versions of the same class or the need to marshall the Externalizer class, the payload sizes generated via this method are not the most efficient.

  • This model requires that the marshalled class be annotated with link:https://docs.jboss.org/infinispan/9.4/apidocs/org/infinispan/commons/marshall/SerializeWith.html but a user might need to provide an Externalizer for a class for which source code is not available, or for any other constraints, it cannot be modified.

  • The use of annotations by this model might be limiting for framework developers or service providers that try to abstract lower level details, such as the marshalling layer, away from the user.

If you’re affected by any of these disadvantages, an alternative method to provide externalizers is available via more advanced externalizers:

7.4.3. Advanced Externalizers

AdvancedExternalizer provides an alternative way to provide externalizers for marshalling/unmarshalling user defined classes that overcome the deficiencies of the more user-friendly externalizer definition model explained in Externalizer. For example:

import org.infinispan.marshall.AdvancedExternalizer;

public class Person {

   final String name;
   final int age;

   public Person(String name, int age) {
      this.name = name;
      this.age = age;

   public static class PersonExternalizer implements AdvancedExternalizer<Person> {
      public void writeObject(ObjectOutput output, Person person)
            throws IOException {

      public Person readObject(ObjectInput input)
            throws IOException, ClassNotFoundException {
         return new Person((String) input.readObject(), input.readInt());

      public Set<Class<? extends Person>> getTypeClasses() {
         return Util.<Class<? extends Person>>asSet(Person.class);

      public Integer getId() {
         return 2345;

The first noticeable difference is that this method does not require user classes to be annotated in anyway, so it can be used with classes for which source code is not available or that cannot be modified. The bound between the externalizer and the classes that are marshalled/unmarshalled is set by providing an implementation for getTypeClasses() which should return the list of classes that this externalizer can marshall:

Linking Externalizers with Marshaller Classes

Once the Externalizer’s readObject() and writeObject() methods have been implemented, it’s time to link them up together with the type classes that they externalize. To do so, the Externalizer implementation must provide a getTypeClasses() implementation. For example:

import org.infinispan.commons.util.Util;
public Set<Class<? extends ReplicableCommand>> getTypeClasses() {
  return Util.asSet(LockControlCommand.class, RehashControlCommand.class,
      StateTransferControlCommand.class, GetKeyValueCommand.class,
      SingleRpcCommand.class, CommitCommand.class,
      PrepareCommand.class, RollbackCommand.class,
      ClearCommand.class, EvictCommand.class,
      InvalidateCommand.class, InvalidateL1Command.class,
      PutKeyValueCommand.class, PutMapCommand.class,
      RemoveCommand.class, ReplaceCommand.class);

In the code above, ReplicableCommandExternalizer indicates that it can externalize several type of commands. In fact, it marshalls all commands that extend ReplicableCommand interface, but currently the framework only supports class equality comparison and so, it’s not possible to indicate that the classes to marshalled are all children of a particular class/interface.

However there might sometimes when the classes to be externalized are private and hence it’s not possible to reference the actual class instance. In this situations, users can attempt to look up the class with the given fully qualified class name and pass that back. For example:

public Set<Class<? extends List>> getTypeClasses() {
  return Util.<Class<? extends List>>asSet(
Externalizer Identifier

Secondly, in order to save the maximum amount of space possible in the payloads generated, advanced externalizers require externalizer implementations to provide a positive identified via getId() implementations or via XML/programmatic configuration that identifies the externalizer when unmarshalling a payload.  In order for this to work however, advanced externalizers require externalizers to be registered on cache manager creation time via XML or programmatic configuration which will be explained in next section. On the contrary, externalizers based on Externalizer and SerializeWith require no pre-registration whatsoever. Internally, Infinispan uses this advanced externalizer mechanism in order to marshall/unmarshall internal classes.

So, getId() should return a positive integer that allows the externalizer to be identified at read time to figure out which Externalizer should read the contents of the incoming buffer, or it can return null. If getId() returns null, it is indicating that the id of this advanced externalizer will be defined via XML/programmatic configuration, which will be explained in next section.

Regardless of the source of the the id, using a positive integer allows for very efficient variable length encoding of numbers, and it’s much more efficient than shipping externalizer implementation class information or class name around. Infinispan users can use any positive integer as long as it does not clash with any other identifier in the system. It’s important to understand that a user defined externalizer can even use the same numbers as the externalizers in the Infinispan Core project because the internal Infinispan Core externalizers are special and they use a different number space to the user defined externalizers. On the contrary, users should avoid using numbers that are within the pre-assigned identifier ranges which can be found at the end of this article. Infinispan checks for id duplicates on startup, and if any are found, startup is halted with an error.

When it comes to maintaining which ids are in use, it’s highly recommended that this is done in a centralized way. For example, getId() implementations could reference a set of statically defined identifiers in a separate class or interface. Such class/interface would give a global view of the identifiers in use and so can make it easier to assign new ids.

Registering Advanced Externalizers

The following example shows the type of configuration required to register an advanced externalizer implementation for Person object shown earlier stored as a static inner class within it:

      <advanced-externalizer class="Person$PersonExternalizer"/>


GlobalConfigurationBuilder builder = ...
   .addAdvancedExternalizer(new Person.PersonExternalizer());

As mentioned earlier, when listing these externalizer implementations, users can optionally provide the identifier of the externalizer via XML or programmatically instead of via getId() implementation. Again, this offers a centralized way to maintain the identifiers but it’s important that the rules are clear: An AdvancedExternalizer implementation, either via XML/programmatic configuration or via annotation, needs to be associated with an identifier. If it isn’t, Infinispan will throw an error and abort startup. If a particular AdvancedExternalizer implementation defines an id both via XML/programmatic configuration and annotation, the value defined via XML/programmatically is the one that will be used. Here’s an example of an externalizer whose id is defined at registration time:

      <advanced-externalizer id="123"


GlobalConfigurationBuilder builder = ...
   .addAdvancedExternalizer(123, new Person.PersonExternalizer());

Finally, a couple of notes about the programmatic configuration. GlobalConfiguration.addExternalizer() takes varargs, so it means that it is possible to register multiple externalizers in just one go, assuming that their ids have already been defined via @Marshalls annotation. For example:

   .addAdvancedExternalizer(new Person.PersonExternalizer(),
                            new Address.AddressExternalizer());
Preassigned Externalizer Id Ranges

This is the list of Externalizer identifiers that are used by Infinispan based modules or frameworks. Infinispan users should avoid using ids within these ranges.

Infinispan Tree Module:

1000 - 1099

Infinispan Server Modules:

1100 - 1199

Hibernate Infinispan Second Level Cache:

1200 - 1299

Infinispan Lucene Directory:

1300 - 1399

Hibernate OGM:

1400 - 1499

Hibernate Search:

1500 - 1599

Infinispan Query Module:

1600 - 1699

Infinispan Remote Query Module:

1700 - 1799

Infinispan Scripting Module:

1800 - 1849

Infinispan Server Event Logger Module:

1850 - 1899

Infinispan Remote Store:

1900 - 1999

Infinispan Counters:

2000 - 2049

Infinispan Multimap:

2050 - 2099

Infinispan Locks:

2100 - 2149

8. Transactions

Infinispan can be configured to use and to participate in JTA compliant transactions. Alternatively, if transaction support is disabled, it is equivalent to using autocommit in JDBC calls, where modifications are potentially replicated after every change (if replication is enabled).

On every cache operation Infinispan does the following:

  1. Retrieves the current Transaction associated with the thread

  2. If not already done, registers XAResource with the transaction manager to be notified when a transaction commits or is rolled back.

In order to do this, the cache has to be provided with a reference to the environment’s TransactionManager. This is usually done by configuring the cache with the class name of an implementation of the TransactionManagerLookup interface. When the cache starts, it will create an instance of this class and invoke its getTransactionManager() method, which returns a reference to the TransactionManager.

Infinispan ships with several transaction manager lookup classes:

Transaction manager lookup implementations
  • EmbeddedTransactionManagerLookup: This provides with a basic transaction manager which should only be used for embedded mode when no other implementation is available. This implementation has some severe limitations to do with concurrent transactions and recovery.

  • JBossStandaloneJTAManagerLookup: If you’re running Infinispan in a standalone environment, or in JBoss AS 7 and earlier, and WildFly 8, 9, and 10, this should be your default choice for transaction manager. It’s a fully fledged transaction manager based on JBoss Transactions which overcomes all the deficiencies of the EmbeddedTransactionManager.

  • WildflyTransactionManagerLookup: If you’re running Infinispan in Wildfly 11 or later, this should be your default choice for transaction manager.

  • GenericTransactionManagerLookup: This is a lookup class that locate transaction managers in the most popular Java EE application servers. If no transaction manager can be found, it defaults on the EmbeddedTransactionManager.

WARN: DummyTransactionManagerLookup has been deprecated in 9.0 and it will be removed in the future. Use EmbeddedTransactionManagerLookup instead.

Once initialized, the TransactionManager can also be obtained from the Cache itself:

//the cache must have a transactionManagerLookupClass defined
Cache cache = cacheManager.getCache();

//equivalent with calling TransactionManagerLookup.getTransactionManager();
TransactionManager tm = cache.getAdvancedCache().getTransactionManager();

8.1. Configuring transactions

Transactions are configured at cache level. Below is the configuration that affects a transaction behaviour and a small description of each configuration attribute.


or programmatically:

ConfigurationBuilder builder = new ConfigurationBuilder();
    .transactionManagerLookup(new GenericTransactionManagerLookup())
  • isolation - configures the isolation level. Check section Isolation Levels for more details. Default is REPEATABLE_READ.

  • write-skew - enables write skew checks (deprecated). Infinispan automatically sets this attribute in Library Mode. Default is false for READ_COMMITTED. Default is true for REPEATABLE_READ. See Write Skews for more details.

  • locking - configures whether the cache uses optimistic or pessimistic locking. Check section Transaction Locking for more details. Default is OPTIMISTIC.

  • auto-commit - if enable, the user does not need to start a transaction manually for a single operation. The transaction is automatically started and committed. Default is true.

  • complete-timeout - the duration in milliseconds to keep information about completed transactions. Default is 60000.

  • mode - configures whether the cache is transactional or not. Default is NONE. The available options are:

    • NONE - non transactional cache

    • FULL_XA - XA transactional cache with recovery enabled. Check section Transaction recovery for more details about recovery.

    • NON_DURABLE_XA - XA transactional cache with recovery disabled.

    • NON_XA - transactional cache with integration via Synchronization instead of XA. Check section Enlisting Synchronizations for details.

    • BATCH- transactional cache using batch to group operations. Check section Batching for details.

  • notifications - enables/disables triggering transactional events in cache listeners. Default is true.

  • protocol - configures the protocol uses. Default is DEFAULT. Values available are:

    • DEFAULT - uses the traditional Two-Phase-Commit protocol. It is described below.

    • TOTAL_ORDER - uses total order ensured by the Transport to commit transactions. Check section Total Order based commit protocol for details.

  • reaper-interval - the time interval in millisecond at which the thread that cleans up transaction completion information kicks in. Defaults is 30000.

  • recovery-cache - configures the cache name to store the recovery information. Check section Transaction recovery for more details about recovery. Default is recoveryInfoCacheName.

  • stop-timeout - the time in millisecond to wait for ongoing transaction when the cache is stopping. Default is 30000.

  • transaction-manager-lookup - configures the fully qualified class name of a class that looks up a reference to a javax.transaction.TransactionManager. Default is org.infinispan.transaction.lookup.GenericTransactionManagerLookup.

  • Versioning scheme - configure the version scheme to use when write skew is enabled with optimistic or total order transactions. Check section Write Skews for more details. Default is NONE.

For more details on how Two-Phase-Commit (2PC) is implemented in Infinispan and how locks are being acquired see the section below. More details about the configuration settings are available in Configuration reference.

8.2. Isolation levels

Infinispan offers two isolation levels - READ_COMMITTED and REPEATABLE_READ.

These isolation levels determine when readers see a concurrent write, and are internally implemented using different subclasses of MVCCEntry, which have different behaviour in how state is committed back to the data container.

Here’s a more detailed example that should help understand the difference between READ_COMMITTED and REPEATABLE_READ in the context of Infinispan. With READ_COMMITTED, if between two consecutive read calls on the same key, the key has been updated by another transaction, the second read may return the new updated value:

Thread1: tx1.begin()
Thread1: cache.get(k) // returns v
Thread2:                                       tx2.begin()
Thread2:                                       cache.get(k) // returns v
Thread2:                                       cache.put(k, v2)
Thread2:                                       tx2.commit()
Thread1: cache.get(k) // returns v2!
Thread1: tx1.commit()

With REPEATABLE_READ, the final get will still return v. So, if you’re going to retrieve the same key multiple times within a transaction, you should use REPEATABLE_READ.

However, as read-locks are not acquired even for REPEATABLE_READ, this phenomena can occur:

cache.get("A") // returns 1
cache.get("B") // returns 1

Thread1: tx1.begin()
Thread1: cache.put("A", 2)
Thread1: cache.put("B", 2)
Thread2:                                       tx2.begin()
Thread2:                                       cache.get("A") // returns 1
Thread1: tx1.commit()
Thread2:                                       cache.get("B") // returns 2
Thread2:                                       tx2.commit()

8.3. Transaction locking

8.3.1. Pessimistic transactional cache

From a lock acquisition perspective, pessimistic transactions obtain locks on keys at the time the key is written.

  1. A lock request is sent to the primary owner (can be an explicit lock request or an operation)

  2. The primary owner tries to acquire the lock:

    1. If it succeed, it sends back a positive reply;

    2. Otherwise, a negative reply is sent and the transaction is rollback.

As an example:

cache.put(k1,v1); //k1 is locked.
cache.remove(k2); //k2 is locked when this returns

When cache.put(k1,v1) returns, k1 is locked and no other transaction running anywhere in the cluster can write to it. Reading k1 is still possible. The lock on k1 is released when the transaction completes (commits or rollbacks).

For conditional operations, the validation is performed in the originator.

8.3.2. Optimistic transactional cache

With optimistic transactions locks are being acquired at transaction prepare time and are only being held up to the point the transaction commits (or rollbacks). This is different from the 5.0 default locking model where local locks are being acquire on writes and cluster locks are being acquired during prepare time.

  1. The prepare is sent to all the owners.

  2. The primary owners try to acquire the locks needed:

    1. If locking succeeds, it performs the write skew check.

    2. If the write skew check succeeds (or is disabled), send a positive reply.

    3. Otherwise, a negative reply is sent and the transaction is rolled back.

As an example:

transactionManager.commit(); //at prepare time, K1 and K2 is locked until committed/rolled back.
For conditional commands, the validation still happens on the originator.

8.3.3. What do I need - pessimistic or optimistic transactions?

From a use case perspective, optimistic transactions should be used when there is not a lot of contention between multiple transactions running at the same time. That is because the optimistic transactions rollback if data has changed between the time it was read and the time it was committed (with write skew check enabled).

On the other hand, pessimistic transactions might be a better fit when there is high contention on the keys and transaction rollbacks are less desirable. Pessimistic transactions are more costly by their nature: each write operation potentially involves a RPC for lock acquisition.

8.4. Write Skews

Write skews occur when two transactions independently and simultaneously read and write to the same key. The result of a write skew is that both transactions successfully commit updates to the same key but with different values.

In Library Mode, Infinispan automatically performs write skew checks to ensure data consistency for REPEATABLE_READ isolation levels in optimistic transactions. This allows Infinispan to detect and roll back one of the transactions.

The write-skew attribute is deprecated for Library Mode. In Remote Client/Server Mode, this attribute is not a valid declaration.

When operating in LOCAL mode, write skew checks rely on Java object references to compare differences, which provides a reliable technique for checking for write skews.

In clustered environments, you should configure data versioning to ensure reliable write skew checks. Infinispan provides an implementation of the EntryVersion interface called SIMPLE versioning, which is backed by a long that is incremented each time the entry is updated.

<versioning scheme="SIMPLE|NONE" />


new ConfigurationBuilder().versioning().scheme(SIMPLE);

8.4.1. Forcing write locks on keys in pessimitic transactions

To avoid write-skews with pessimistic transactions, lock keys at read-time with Flag.FORCE_WRITE_LOCK.

  • In non-transactional caches, Flag.FORCE_WRITE_LOCK does not work. The get() call reads the key value but does not acquire locks remotely.

  • You should use Flag.FORCE_WRITE_LOCK with transactions in which the entity is updated later in the same transaction.

Compare the following code snippets for an example of Flag.FORCE_WRITE_LOCK:

// begin the transaction
if (!cache.getAdvancedCache().lock(key)) {
   // abort the transaction because the key was not locked
} else {
   cache.put(key, value);
   // commit the transaction
// begin the transaction
try {
   // throws an exception if the key is not locked.
   cache.put(key, value);
} catch (CacheException e) {
   // mark the transaction rollback-only
// commit or rollback the transaction

8.5. Dealing with exceptions

If a CacheException (or a subclass of it) is thrown by a cache method within the scope of a JTA transaction, then the transaction is automatically marked for rollback.

8.6. Enlisting Synchronizations

By default Infinispan registers itself as a first class participant in distributed transactions through XAResource. There are situations where Infinispan is not required to be a participant in the transaction, but only to be notified by its lifecycle (prepare, complete): e.g. in the case Infinispan is used as a 2nd level cache in Hibernate.

Infinispan allows transaction enlistment through Synchronization. To enable it just use NON_XA transaction mode.

Synchronizations have the advantage that they allow TransactionManager to optimize 2PC with a 1PC where only one other resource is enlisted with that transaction (last resource commit optimization). E.g. Hibernate second level cache: if Infinispan registers itself with the TransactionManager as a XAResource than at commit time, the TransactionManager sees two XAResource (cache and database) and does not make this optimization. Having to coordinate between two resources it needs to write the tx log to disk. On the other hand, registering Infinispan as a Synchronisation makes the TransactionManager skip writing the log to the disk (performance improvement).

8.7. Batching

Batching allows atomicity and some characteristics of a transaction, but not full-blown JTA or XA capabilities. Batching is often a lot lighter and cheaper than a full-blown transaction.

Generally speaking, one should use batching API whenever the only participant in the transaction is an Infinispan cluster. On the other hand, JTA transactions (involving TransactionManager) should be used whenever the transactions involves multiple systems. E.g. considering the "Hello world!" of transactions: transferring money from one bank account to the other. If both accounts are stored within Infinispan, then batching can be used. If one account is in a database and the other is Infinispan, then distributed transactions are required.
You do not have to have a transaction manager defined to use batching.

8.7.1. API

Once you have configured your cache to use batching, you use it by calling startBatch() and endBatch() on Cache. E.g.,

Cache cache = cacheManager.getCache();
// not using a batch
cache.put("key", "value"); // will replicate immediately

// using a batch
cache.put("k1", "value");
cache.put("k2", "value");
cache.put("k2", "value");
cache.endBatch(true); // This will now replicate the modifications since the batch was started.

// a new batch
cache.put("k1", "value");
cache.put("k2", "value");
cache.put("k3", "value");
cache.endBatch(false); // This will "discard" changes made in the batch

8.7.2. Batching and JTA

Behind the scenes, the batching functionality starts a JTA transaction, and all the invocations in that scope are associated with it. For this it uses a very simple (e.g. no recovery) internal TransactionManager implementation. With batching, you get:

  1. Locks you acquire during an invocation are held until the batch completes

  2. Changes are all replicated around the cluster in a batch as part of the batch completion process. Reduces replication chatter for each update in the batch.

  3. If synchronous replication or invalidation are used, a failure in replication/invalidation will cause the batch to roll back.

  4. All the transaction related configurations apply for batching as well.

8.8. Transaction recovery

Recovery is a feature of XA transactions, which deal with the eventuality of a resource or possibly even the transaction manager failing, and recovering accordingly from such a situation.

8.8.1. When to use recovery

Consider a distributed transaction in which money is transferred from an account stored in an external database to an account stored in Infinispan. When TransactionManager.commit() is invoked, both resources prepare successfully (1st phase). During the commit (2nd) phase, the database successfully applies the changes whilst Infinispan fails before receiving the commit request from the transaction manager. At this point the system is in an inconsistent state: money is taken from the account in the external database but not visible yet in Infinispan (since locks are only released during 2nd phase of a two-phase commit protocol). Recovery deals with this situation to make sure data in both the database and Infinispan ends up in a consistent state.

8.8.2. How does it work

Recovery is coordinated by the transaction manager. The transaction manager works with Infinispan to determine the list of in-doubt transactions that require manual intervention and informs the system administrator (via email, log alerts, etc). This process is transaction manager specific, but generally requires some configuration on the transaction manager.  

Knowing the in-doubt transaction ids, the system administrator can now connect to the Infinispan cluster and replay the commit of transactions or force the rollback. Infinispan provides JMX tooling for this - this is explained extensively in the Transaction recovery and reconciliation section.

8.8.3. Configuring recovery   

Recovery is not enabled by default in Infinispan. If disabled, the TransactionManager won’t be able to work with Infinispan to determine the in-doubt transactions. The Transaction configuration section shows how to enable it.

NOTE: recovery-cache attribute is not mandatory and it is configured per-cache.

For recovery to work, mode must be set to FULL_XA, since full-blown XA transactions are needed.
Enable JMX support

In order to be able to use JMX for managing recovery JMX support must be explicitly enabled. More about enabling JMX in the Management chapter.

8.8.4. Recovery cache

In order to track in-doubt transactions and be able to reply them, Infinispan caches all transaction state for future use. This state is held only for in-doubt transaction, being removed for successfully completed transactions after when the commit/rollback phase completed.

This in-doubt transaction data is held within a local cache: this allows one to configure swapping this info to disk through cache loader in the case it gets too big. This cache can be specified through the recovery-cache configuration attribute. If not specified Infinispan will configure a local cache for you.

It is possible (though not mandated) to share same recovery cache between all the Infinispan caches that have recovery enabled. If the default recovery cache is overridden, then the specified recovery cache must use a TransactionManagerLookup that returns a different transaction manager than the one used by the cache itself.

8.8.5. Integration with the transaction manager

Even though this is transaction manager specific, generally a transaction manager would need a reference to a XAResource implementation in order to invoke XAResource.recover() on it. In order to obtain a reference to an Infinispan XAResource following API can be used:

XAResource xar = cache.getAdvancedCache().getXAResource();

It is a common practice to run the recovery in a different process from the one running the transaction.

8.8.6. Reconciliation

The transaction manager informs the system administrator on in-doubt transaction in a proprietary way. At this stage it is assumed that the system administrator knows transaction’s XID (a byte array).

A normal recovery flow is:

  • STEP 1: The system administrator connects to an Infinispan server through JMX, and lists the in doubt transactions. The image below demonstrates JConsole connecting to an Infinispan node that has an in doubt transaction.

Figure 6. Show in-doubt transactions

The status of each in-doubt transaction is displayed(in this example " PREPARED "). There might be multiple elements in the status field, e.g. "PREPARED" and "COMMITTED" in the case the transaction committed on certain nodes but not on all of them.  

  • STEP 2: The system administrator visually maps the XID received from the transaction manager to an Infinispan internal id, represented as a number. This step is needed because the XID, a byte array, cannot conveniently be passed to the JMX tool (e.g. JConsole) and then re-assembled on Infinispan’s side.

  • STEP 3: The system administrator forces the transaction’s commit/rollback through the corresponding jmx operation, based on the internal id. The image below is obtained by forcing the commit of the transaction based on its internal id.

Figure 7. Force commit
All JMX operations described above can be executed on any node, regardless of where the transaction originated.
Force commit/rollback based on XID

XID-based JMX operations for forcing in-doubt transactions' commit/rollback are available as well: these methods receive byte[] arrays describing the XID instead of the number associated with the transactions (as previously described at step 2). These can be useful e.g. if one wants to set up an automatic completion job for certain in-doubt transactions. This process is plugged into transaction manager’s recovery and has access to the transaction manager’s XID objects.

8.8.7. Want to know more?

The recovery design document describes in more detail the insides of transaction recovery implementation.

8.9. Total Order based commit protocol

The Total Order based protocol is a multi-master scheme (in this context, multi-master scheme means that all nodes can update all the data) as the (optimistic/pessimist) locking mode implemented in Infinispan. This commit protocol relies on the concept of totally ordered delivery of messages which, informally, implies that each node which delivers a set of messages, delivers them in the same order.

This protocol comes with this advantages.

  1. transactions can be committed in one phase, as they are delivered in the same order by the nodes that receive them.

  2. it mitigates distributed deadlocks.

The weaknesses of this approach are the fact that its implementation relies on a single thread per node which delivers the transaction and its modification, and the slightly cost of total ordering the messages in Transport.

Thus, this protocol delivers best performance in scenarios of high contention , in which it can benefit from the single-phase commit and the deliver thread is not the bottleneck.

Currently, the Total Order based protocol is available only in transactional caches for replicated and distributed modes.

8.9.1. Overview

The Total Order based commit protocol only affects how transactions are committed by Infinispan and the isolation level and write skew affects it behaviour.

When write skew is disabled, the transaction can be committed/rolled back in single phase. The data consistency is guaranteed by the Transport that ensures that all owners of a key will deliver the same transactions set by the same order.

On other hand, when write skew is enabled, the protocol adapts and uses one phase commit when it is safe. In XaResource enlistment, we can use one phase if the TransactionManager request a commit in one phase (last resource commit optimization) and the Infinispan cache is configured in replicated mode. This optimization is not safe in distributed mode because each node performs the write skew check validation in different keys subset. When in Synchronization enlistment, the TransactionManager does not provide any information if Infinispan is the only resource enlisted (last resource commit optimization), so it is not possible to commit in a single phase.

Commit in one phase

When the transaction ends, Infinispan sends the transaction (and its modification) in total order. This ensures all the transactions are deliver in the same order in all the involved Infinispan nodes. As a result, when a transaction is delivered, it performs a deterministic write skew check over the same state (if enabled), leading to the same outcome (transaction commit or rollback).

total order 1pc
Figure 8. 1-phase commit

The figure above demonstrates a high level example with 3 nodes. Node1 and Node3 are running one transaction each and lets assume that both transaction writes on the same key. To make it more interesting, lets assume that both nodes tries to commit at the same time, represented by the first colored circle in the figure. The blue circle represents the transaction tx1 and the green the transaction tx2 . Both nodes do a remote invocation in total order (to-send) with the transaction’s modifications. At this moment, all the nodes will agree in the same deliver order, for example, tx1 followed by tx2 . Then, each node delivers tx1 , perform the validation and commits the modifications. The same steps are performed for tx2 but, in this case, the validation will fail and the transaction is rollback in all the involved nodes.

Commit in two phases

In the first phase, it sends the modification in total order and the write skew check is performed. The result of the write skew check is sent back to the originator. As soon as it has the confirmation that all keys are successfully validated, it give a positive response to the TransactionManager. On other hand, if it receives a negative reply, it returns a negative response to the TransactionManager. Finally, the transaction is committed or aborted in the second phase depending of the TransactionManager request.

total order 2pc
Figure 9. 2-phase commit

The figure above shows the scenario described in the first figure but now committing the transactions using two phases. When tx1 is deliver, it performs the validation and it replies to the TransactionManager. Next, lets assume that tx2 is deliver before the TransactionManager request the second phase for tx1. In this case, tx2 will be enqueued and it will be validated only when tx1 is completed. Eventually, the TransactionManager for tx1 will request the second phase (the commit) and all the nodes are free to perform the validation of tx2 .

Transaction Recovery

Transaction recovery is currently not available for Total Order based commit protocol.

State Transfer

For simplicity reasons, the total order based commit protocol uses a blocking version of the current state transfer. The main differences are:

  1. enqueue the transaction deliver while the state transfer is in progress;

  2. the state transfer control messages (CacheTopologyControlCommand) are sent in total order.

This way, it provides a synchronization between the state transfer and the transactions deliver that is the same all the nodes. Although, the transactions caught in the middle of state transfer (i.e. sent before the state transfer start and deliver after it) needs to be re-sent to find a new total order involving the new joiners.

total order joing during st
Figure 10. Node joining during transaction

The figure above describes a node joining. In the scenario, the tx2 is sent in topologyId=1 but when it is received, it is in topologyId=2 . So, the transaction is re-sent involving the new nodes.

8.9.2. Configuration

To use total order in your cache, you need to add the TOA protocol in your jgroups.xml configuration file.

<tom.TOA />
Check the JGroups Manual for more details.
If you are interested in detail how JGroups guarantees total order, check the link::http://jgroups.org/manual/index.html#TOA[TOA manual].

Also, you need to set the protocol=TOTAL_ORDER in the <transaction> element, as shown in Transaction configuration.

8.9.3. When to use it?

Total order shows benefits when used in write intensive and high contented workloads. It avoids contention in the lock keys.

9. Locking and Concurrency

Infinispan makes use of multi-versioned concurrency control (MVCC) - a concurrency scheme popular with relational databases and other data stores. MVCC offers many advantages over coarse-grained Java synchronization and even JDK Locks for access to shared data, including:

  • allowing concurrent readers and writers

  • readers and writers do not block one another

  • write skews can be detected and handled

  • internal locks can be striped

9.1. Locking implementation details

Infinispan’s MVCC implementation makes use of minimal locks and synchronizations, leaning heavily towards lock-free techniques such as compare-and-swap and lock-free data structures wherever possible, which helps optimize for multi-CPU and multi-core environments.

In particular, Infinispan’s MVCC implementation is heavily optimized for readers. Reader threads do not acquire explicit locks for entries, and instead directly read the entry in question.

Writers, on the other hand, need to acquire a write lock. This ensures only one concurrent writer per entry, causing concurrent writers to queue up to change an entry. To allow concurrent reads, writers make a copy of the entry they intend to modify, by wrapping the entry in an MVCCEntry. This copy isolates concurrent readers from seeing partially modified state. Once a write has completed, MVCCEntry.commit() will flush changes to the data container and subsequent readers will see the changes written.

9.1.1. How does it work in clustered caches?

In clustered caches, each key has a node responsible to lock the key. This node is called primary owner.

Non Transactional caches
  1. The write operation is sent to the primary owner of the key.

  2. The primary owner tries to lock the key.

    1. If it succeeds, it forwards the operation to the other owners;

    2. Otherwise, an exception is thrown.

If the operation is conditional and it fails on the primary owner, it is not forwarded to the other owners.
If the operation is executed locally in the primary owner, the first step is skipped.

9.1.2. Transactional caches

The transactional cache supports optimistic and pessimistic locking mode. Check section Transaction Locking for more information about it.

9.1.3. Isolation levels

Isolation level affects what transactions can read when running concurrently with other transaction. Check section Isolation Levels for more details about it.

9.1.4. The LockManager

The LockManager is a component that is responsible for locking an entry for writing. The LockManager makes use of a LockContainer to locate/hold/create locks. LockContainers come in two broad flavours, with support for lock striping and with support for one lock per entry.

9.1.5. Lock striping

Lock striping entails the use of a fixed-size, shared collection of locks for the entire cache, with locks being allocated to entries based on the entry’s key’s hash code. Similar to the way the JDK’s ConcurrentHashMap allocates locks, this allows for a highly scalable, fixed-overhead locking mechanism in exchange for potentially unrelated entries being blocked by the same lock.

The alternative is to disable lock striping - which would mean a new lock is created per entry. This approach may give you greater concurrent throughput, but it will be at the cost of additional memory usage, garbage collection churn, etc.

Default lock striping settings
lock striping is disabled by default, due to potential deadlocks that can happen if locks for different keys end up in the same lock stripe.

The size of the shared lock collection used by lock striping can be tuned using the concurrencyLevel attribute of the `<locking /> configuration element.

Configuration example:

<locking striping="false|true"/>


new ConfigurationBuilder().locking().useLockStriping(false|true);

9.1.6. Concurrency levels

In addition to determining the size of the striped lock container, this concurrency level is also used to tune any JDK ConcurrentHashMap based collections where related, such as internal to DataContainers. Please refer to the JDK ConcurrentHashMap Javadocs for a detailed discussion of concurrency levels, as this parameter is used in exactly the same way in Infinispan.

Configuration example:

<locking concurrency-level="32"/>


new ConfigurationBuilder().locking().concurrencyLevel(32);

9.1.7. Lock timeout

The lock timeout specifies the amount of time, in milliseconds, to wait for a contented lock.

Configuration example:

<locking acquire-timeout="10000"/>


new ConfigurationBuilder().locking().lockAcquisitionTimeout(10000);
new ConfigurationBuilder().locking().lockAcquisitionTimeout(10, TimeUnit.SECONDS);

9.1.8. Consistency

The fact that a single owner is locked (as opposed to all owners being locked) does not break the following consistency guarantee: if key K is hashed to nodes {A, B} and transaction TX1 acquires a lock for K, let’s say on A. If another transaction, TX2, is started on B (or any other node) and TX2 tries to lock K then it will fail with a timeout as the lock is already held by TX1. The reason for this is the that the lock for a key K is always, deterministically, acquired on the same node of the cluster, regardless of where the transaction originates.

9.2. Data Versioning

Infinispan supports two forms of data versioning: simple and external. The simple versioning is used in transactional caches for write skew check. See Write Skews.

The external versioning is used to encapsulate an external source of data versioning within Infinispan, such as when using Infinispan with Hibernate which in turn gets its data version information directly from a database.

In this scheme, a mechanism to pass in the version becomes necessary, and overloaded versions of put() and putForExternalRead() will be provided in AdvancedCache to take in an external data version. This is then stored on the InvocationContext and applied to the entry at commit time.

Write skew checks cannot and will not be performed in the case of external data versioning.

10. Executing code in the Grid

The main benefit of a Cache is the ability to very quickly lookup a value by its key, even across machines. In fact this use alone is probably the reason many users use Infinispan. However Infinispan can provide many more benefits that aren’t immediately apparent. Since Infinispan is usually used in a cluster of machines we also have features available that can help utilize the entire cluster for performing the user’s desired workload.

This section covers only executing code in the grid using an embedded cache, if you are using a remote cache you should check out Executing code in the Remote Grid.

10.1. Cluster Executor

Since you have a group of machines, it makes sense to leverage their combined computing power for executing code on all of them them. The cache manager comes with a nice utility that allows you to execute arbitrary code in the cluster. Note this feature requires no Cache to be used. This Cluster Executor can be retrieved by calling executor() on the EmbeddedCacheManager. This executor is retrievable in both clustered and non clustered configurations.

The ClusterExecutor is specifically designed for executing code where the code is not reliant upon the data in a cache and is used instead as a way to help users to execute code easily in the cluster.

This manager was built specifically using Java 8 and such has functional APIs in mind, thus all methods take a functional inteface as an argument. Also since these arguments will be sent to other nodes they need to be serializable. We even used a nice trick to ensure our lambdas are immediately Serializable. That is by having the arguments implement both Serializable and the real argument type (ie. Runnable or Function). The JRE will pick the most specific class when determining which method to invoke, so in that case your lambdas will always be serializable. It is also possible to use an Externalizer to possibly reduce message size further.

The manager by default will submit a given command to all nodes in the cluster including the node where it was submitted from. You can control on which nodes the task is executed on by using the filterTargets methods as is explained in the section.

10.1.1. Filtering execution nodes

It is possible to limit on which nodes the command will be ran. For example you may want to only run a computation on machines in the same rack. Or you may want to perform an operation once in the local site and again on a different site. A cluster executor can limit what nodes it sends requests to at the scope of same or different machine, rack or site level.

   EmbeddedCacheManager manager = ...;

To use this topology base filtering you must enable topology aware consistent hashing through Server Hinting.

You can also filter using a predicate based on the Address of the node. This can also be optionally combined with topology based filtering in the previous code snippet.

We also allow the target node to be chosen by any means using a Predicate that will filter out which nodes can be considered for execution. Note this can also be combined with Topology filtering at the same time to allow even more fine control of where you code is executed within the cluster.

   EmbeddedCacheManager manager = ...;
   // Just filter
   manager.executor().filterTargets(a -> a.equals(..)).submit(...)
   // Filter only those in the desired topology
   manager.executor().filterTargets(ClusterExecutionPolicy.SAME_SITE, a -> a.equals(..)).submit(...)

10.1.2. Timeout

Cluster Executor allows for a timeout to be set per invocation. This defaults to the distributed sync timeout as configured on the Transport Configuration. This timeout works in both a clustered and non clustered cache manager. The executor may or may not interrupt the threads executing a task when the timeout expires. However when the timeout occurs any Consumer or Future will be completed passing back a TimeoutException. This value can be overridden by ivoking the timeout method and supplying the desired duration.

10.1.3. Single Node Submission

Cluster Executor can also run in single node submission mode instead of submitting the command to all nodes it will instead pick one of the nodes that would have normally received the command and instead submit it it to only one. Each submission will possibly use a different node to execute the task on. This can be very useful to use the ClusterExecutor as a java.util.concurrent.Executor which you may have noticed that ClusterExecutor implements.

   EmbeddedCacheManager manager = ...;

When running in single node submission it may be desirable to also allow the Cluster Executor handle cases where an exception occurred during the processing of a given command by retrying the command again. When this occurs the Cluster Executor will choose a single node again to resubmit the command to up to the desired number of failover attempts. Note the chosen node could be any node that passes the topology or predicate check. Failover is enabled by invoking the overridden singleNodeSubmission method. The given command will be resubmitted again to a single node until either the command completes without exception or the total submission amount is equal to the provided failover count.

10.1.4. Example: PI Approximation

This example shows how you can use the ClusterExecutor to estimate the value of PI.

Pi approximation can greatly benefit from parallel distributed execution via Cluster Executor. Recall that area of the square is Sa = 4r2 and area of the circle is Ca=pi*r2. Substituting r2 from the second equation into the first one it turns out that pi = 4 * Ca/Sa. Now, image that we can shoot very large number of darts into a square; if we take ratio of darts that land inside a circle over a total number of darts shot we will approximate Ca/Sa value. Since we know that pi = 4 * Ca/Sa we can easily derive approximate value of pi. The more darts we shoot the better approximation we get. In the example below we shoot 1 billion darts but instead of "shooting" them serially we parallelize work of dart shooting across the entire Infinispan cluster. Note this will work in a cluster of 1 was well, but will be slower.

public class PiAppx {

   public static void main (String [] arg){
      EmbeddedCacheManager cacheManager = ..
      boolean isCluster = ..

      int numPoints = 1_000_000_000;
      int numServers = isCluster ? cacheManager.getMembers().size() : 1;
      int numberPerWorker = numPoints / numServers;

      ClusterExecutor clusterExecutor = cacheManager.executor();
      long start = System.currentTimeMillis();
      // We receive results concurrently - need to handle that
      AtomicLong countCircle = new AtomicLong();
      CompletableFuture<Void> fut = clusterExecutor.submitConsumer(m -> {
         int insideCircleCount = 0;
         for (int i = 0; i < numberPerWorker; i++) {
            double x = Math.random();
            double y = Math.random();
            if (insideCircle(x, y))
         return insideCircleCount;
      }, (address, count, throwable) -> {
         if (throwable != null) {
            System.out.println("Address: " + address + " encountered an error: " + throwable);
         } else {
      fut.whenComplete((v, t) -> {
         // This is invoked after all nodes have responded with a value or exception
         if (t != null) {
            System.out.println("Exception encountered while waiting:" + t);
         } else {
            double appxPi = 4.0 * countCircle.get() / numPoints;

            System.out.println("Distributed PI appx is " + appxPi +
                  " using " + numServers + " node(s), completed in " + (System.currentTimeMillis() - start) + " ms");

      // May have to sleep here to keep alive if no user threads left

   private static boolean insideCircle(double x, double y) {
      return (Math.pow(x - 0.5, 2) + Math.pow(y - 0.5, 2))
            <= Math.pow(0.5, 2);

10.2. Streams

You may want to process a subset or all data in the cache to produce a result. This may bring thoughts of Map Reduce. Infinispan allows the user to do something very similar but utilizes the standard JRE APIs to do so. Java 8 introduced the concept of a Stream which allows functional-style operations on collections rather than having to procedurally iterate over the data yourself. Stream operations can be implemented in a fashion very similar to MapReduce. Streams, just like MapReduce allow you to perform processing upon the entirety of your cache, possibly a very large data set, but in an efficient way.

Streams are the preferred method when dealing with data that exists in the cache because streams automatically adjust to cluster topology changes.

Also since we can control how the entries are iterated upon we can more efficiently perform the operations in a cache that is distributed if you want it to perform all of the operations across the cluster concurrently.

A stream is retrieved from the entrySet, keySet or values collections returned from the Cache by invoking the stream or parallelStream methods.

10.2.1. Common stream operations

This section highlights various options that are present irrespective of what type of underlying cache you are using.

10.2.2. Key filtering

It is possible to filter the stream so that it only operates upon a given subset of keys. This can be done by invoking the filterKeys method on the CacheStream. This should always be used over a Predicate filter and will be faster if the predicate was holding all keys.

If you are familiar with the AdvancedCache interface you may be wondering why you even use getAll over this keyFilter. There are some small benefits (mostly smaller payloads) to using getAll if you need the entries as is and need them all in memory in the local node. However if you need to do processing on these elements a stream is recommended since you will get both distributed and threaded parallelism for free.

10.2.3. Segment based filtering

This is an advanced feature and should only be used with deep knowledge of Infinispan segment and hashing techniques. These segments based filtering can be useful if you need to segment data into separate invocations. This can be useful when integrating with other tools such as Apache Spark.

This option is only supported for replicated and distributed caches. This allows the user to operate upon a subset of data at a time as determined by the KeyPartitioner. The segments can be filtered by invoking filterKeySegments method on the CacheStream. This is applied after the key filter but before any intermediate operations are performed.

10.2.4. Local/Invalidation

A stream used with a local or invalidation cache can be used just the same way you would use a stream on a regular collection. Infinispan handles all of the translations if necessary behind the scenes and works with all of the more interesting options (ie. storeAsBinary, compatibility mode, and a cache loader). Only data local to the node where the stream operation is performed will be used, for example invalidation only uses local entries.

10.2.5. Example

The code below takes a cache and returns a map with all the cache entries whose values contain the string "JBoss"

Map<Object, String> jbossValues = cache.entrySet().stream()
              .filter(e -> e.getValue().contains("JBoss"))
              .collect(Collectors.toMap(Map.Entry::getKey, Map.Entry::getValue));

10.3. Distribution/Replication/Scattered

This is where streams come into their stride. When a stream operation is performed it will send the various intermediate and terminal operations to each node that has pertinent data. This allows processing the intermediate values on the nodes owning the data, and only sending the final results back to the originating nodes, improving performance.

10.3.1. Rehash Aware

Internally the data is segmented and each node only performs the operations upon the data it owns as a primary owner. This allows for data to be processed evenly, assuming segments are granular enough to provide for equal amounts of data on each node.

When you are utilizing a distributed cache, the data can be reshuffled between nodes when a new node joins or leaves. Distributed Streams handle this reshuffling of data automatically so you don’t have to worry about monitoring when nodes leave or join the cluster. Reshuffled entries may be processed a second time, and we keep track of the processed entries at the key level or at the segment level (depending on the terminal operation) to limit the amount of duplicate processing.

It is possible but highly discouraged to disable rehash awareness on the stream. This should only be considered if your request can handle only seeing a subset of data if a rehash occurs. This can be done by invoking CacheStream.disableRehashAware() The performance gain for most operations when a rehash doesn’t occur is completely negligible. The only exceptions are for iterator and forEach, which will use less memory, since they do not have to keep track of processed keys.

Please rethink disabling rehash awareness unless you really know what you are doing.

10.3.2. Serialization

Since the operations are sent across to other nodes they must be serializable by Infinispan marshalling. This allows the operations to be sent to the other nodes.

The simplest way is to use a CacheStream instance and use a lambda just as you would normally. Infinispan overrides all of the various Stream intermediate and terminal methods to take Serializable versions of the arguments (ie. SerializableFunction, SerializablePredicate…​) You can find these methods at CacheStream. This relies on the spec to pick the most specific method as defined here.

In our previous example we used a Collector to collect all the results into a Map. Unfortunately the Collectors class doesn’t produce Serializable instances. Thus if you need to use these, there are two ways to do so:

One option would be to use the CacheCollectors class which allows for a Supplier<Collector> to be provided. This instance could then use the Collectors to supply a Collector which is not serialized. You can read more details about how the collector peforms in a distributed fashion at distributed execution.

Map<Object, String> jbossValues = cache.entrySet().stream()
              .filter(e -> e.getValue().contains("Jboss"))
              .collect(CacheCollectors.serializableCollector(() -> Collectors.toMap(Map.Entry::getKey, Map.Entry::getValue)));

Alternatively, you can avoid the use of CacheCollectors and instead use the overloaded collect methods that take Supplier<Collector>. These overloaded collect methods are only available via CacheStream interface.

Map<Object, String> jbossValues = cache.entrySet().stream()
              .filter(e -> e.getValue().contains("Jboss"))
              .collect(() -> Collectors.toMap(Map.Entry::getKey, Map.Entry::getValue));

If however you are not able to use the Cache and CacheStream interfaces you cannot utilize Serializable arguments and you must instead cast the lambdas to be Serializable manually by casting the lambda to multiple interfaces. It is not a pretty sight but it gets the job done.

Map<Object, String> jbossValues = map.entrySet().stream()
              .filter((Serializable & Predicate<Map.Entry<Object, String>>) e -> e.getValue().contains("Jboss"))
              .collect(CacheCollectors.serializableCollector(() -> Collectors.toMap(Map.Entry::getKey, Map.Entry::getValue)));

The recommended and most performant way is to use an AdvancedExternalizer as this provides the smallest payload. Unfortunately this means you cannot use lamdbas as advanced externalizers require defining the class before hand.

You can use an advanced externalizer as shown below:

   Map<Object, String> jbossValues = cache.entrySet().stream()
              .filter(new ContainsFilter("Jboss"))
              .collect(() -> Collectors.toMap(Map.Entry::getKey, Map.Entry::getValue));

   class ContainsFilter implements Predicate<Map.Entry<Object, String>> {
      private final String target;

      ContainsFilter(String target) {
         this.target = target;

      public boolean test(Map.Entry<Object, String> e) {
         return e.getValue().contains(target);

   class JbossFilterExternalizer implements AdvancedExternalizer<ContainsFilter> {

      public Set<Class<? extends ContainsFilter>> getTypeClasses() {
         return Util.asSet(ContainsFilter.class);

      public Integer getId() {
         return CUSTOM_ID;

      public void writeObject(ObjectOutput output, ContainsFilter object) throws IOException {

      public ContainsFilter readObject(ObjectInput input) throws IOException, ClassNotFoundException {
         return new ContainsFilter(input.readUTF());

You could also use an advanced externalizer for the collector supplier to reduce the payload size even further.

   Map<Object, String> jbossValues = cache.entrySet().stream()
              .filter(new ContainsFilter("Jboss"))

 class ToMapCollectorSupplier<K, U> implements Supplier<Collector<Map.Entry<K, U>, ?, Map<K, U>>> {
      static final ToMapCollectorSupplier INSTANCE = new ToMapCollectorSupplier();

      private ToMapCollectorSupplier() { }

      public Collector<Map.Entry<K, U>, ?, Map<K, U>> get() {
         return Collectors.toMap(Map.Entry::getKey, Map.Entry::getValue);

   class ToMapCollectorSupplierExternalizer implements AdvancedExternalizer<ToMapCollectorSupplier> {

      public Set<Class<? extends ToMapCollectorSupplier>> getTypeClasses() {
         return Util.asSet(ToMapCollectorSupplier.class);

      public Integer getId() {
         return CUSTOM_ID;

      public void writeObject(ObjectOutput output, ToMapCollectorSupplier object) throws IOException {

      public ToMapCollectorSupplier readObject(ObjectInput input) throws IOException, ClassNotFoundException {
         return ToMapCollectorSupplier.INSTANCE;

10.3.3. Parallel Computation

Distributed streams by default try to parallelize as much as possible. It is possible for the end user to control this and actually they always have to control one of the options. There are 2 ways these streams are parallelized.

Local to each node When a stream is created from the cache collection the end user can choose between invoking stream or parallelStream method. Depending on if the parallel stream was picked will enable multiple threading for each node locally. Note that some operations like a rehash aware iterator and forEach operations will always use a sequential stream locally. This could be enhanced at some point to allow for parallel streams locally.

Users should be careful when using local parallelism as it requires having a large number of entries or operations that are computationally expensive to be faster. Also it should be noted that if a user uses a parallel stream with forEach that the action should not block as this would be executed on the common pool, which is normally reserved for computation operations.

Remote requests When there are multiple nodes it may be desirable to control whether the remote requests are all processed at the same time concurrently or one at a time. By default all terminal operations except the iterator perform concurrent requests. The iterator, method to reduce overall memory pressure on the local node, only performs sequential requests which actually performs slightly better.

If a user wishes to change this default however they can do so by invoking the sequentialDistribution or parallelDistribution methods on the CacheStream.

10.3.4. Task timeout

It is possible to set a timeout value for the operation requests. This timeout is used only for remote requests timing out and it is on a per request basis. The former means the local execution will not timeout and the latter means if you have a failover scenario as described above the subsequent requests each have a new timeout. If no timeout is specified it uses the replication timeout as a default timeout. You can set the timeout in your task by doing the following:

CacheStream<Object, String> stream = cache.entrySet().stream();
stream.timeout(1, TimeUnit.MINUTES);

For more information about this, please check the java doc in timeout javadoc.

10.3.5. Injection

The Stream has a terminal operation called forEach which allows for running some sort of side effect operation on the data. In this case it may be desirable to get a reference to the Cache that is backing this Stream. If your Consumer implements the CacheAware interface the injectCache method be invoked before the accept method from the Consumer interface.

10.3.6. Distributed Stream execution

Distributed streams execution works in a fashion very similiar to map reduce. Except in this case we are sending zero to many intermediate operations (map, filter etc.) and a single terminal operation to the various nodes. The operation basically comes down to the following:

  1. The desired segments are grouped by which node is the primary owner of the given segment

  2. A request is generated to send to each remote node that contains the intermediate and terminal operations including which segments it should process

    1. The terminal operation will be performed locally if necessary

    2. Each remote node will receive this request and run the operations and subsequently send the response back

  3. The local node will then gather the local response and remote responses together performing any kind of reduction required by the operations themselves.

  4. Final reduced response is then returned to the user

In most cases all operations are fully distributed, as in the operations are all fully applied on each remote node and usually only the last operation or something related may be reapplied to reduce the results from multiple nodes. One important note is that intermediate values do not actually have to be serializable, it is the last value sent back that is the part desired (exceptions for various operations will be highlighted below).

Terminal operator distributed result reductions The following paragraphs describe how the distributed reductions work for the various terminal operators. Some of these are special in that an intermediate value may be required to be serializable instead of the final result.

allMatch noneMatch anyMatch

The allMatch operation is ran on each node and then all the results are logically anded together locally to get the appropriate value. The noneMatch and anyMatch operations use a logical or instead. These methods also have early termination support, stopping remote and local operations once the final result is known.


The collect method is interesting in that it can do a few extra steps. The remote node performs everything as normal except it doesn’t perform the final finisher upon the result and instead sends back the fully combined results. The local thread then combines the remote and local result into a value which is then finally finished. The key here to remember is that the final value doesn’t have to be serializable but rather the values produced from the supplier and combiner methods.


The count method just adds the numbers together from each node.

findAny findFirst

The findAny operation returns just the first value they find, whether it was from a remote node or locally. Note this supports early termination in that once a value is found it will not process others. Note the findFirst method is special since it requires a sorted intermediate operation, which is detailed in the exceptions section.

max min

The max and min methods find the respective min or max value on each node then a final reduction is performed locally to ensure only the min or max across all nodes is returned.


The various reduce methods 1 , 2 , 3 will end up serializing the result as much as the accumulator can do. Then it will accumulate the local and remote results together locally, before combining if you have provided that. Note this means a value coming from the combiner doesn’t have to be Serializable.

10.3.7. Key based rehash aware operators

The iterator, spliterator and forEach are unlike the other terminal operators in that the rehash awareness has to keep track of what keys per segment have been processed instead of just segments. This is to guarantee an exactly once (iterator & spliterator) or at least once behavior (forEach) even under cluster membership changes.

The iterator and spliterator operators when invoked on a remote node will return back batches of entries, where the next batch is only sent back after the last has been fully consumed. This batching is done to limit how many entries are in memory at a given time. The user node will hold onto which keys it has processed and when a given segment is completed it will release those keys from memory. This is why sequential processing is preferred for the iterator method, so only a subset of segment keys are held in memory at once, instead of from all nodes.

The forEach() method also returns batches, but it returns a batch of keys after it has finished processing at least a batch worth of keys. This way the originating node can know what keys have been processed already to reduce chances of processing the same entry again. Unfortunately this means it is possible to have an at least once behavior when a node goes down unexpectedly. In this case that node could have been processing a batch and not yet completed one and those entries that were processed but not in a completed batch will be ran again when the rehash failure operation occurs. Note that adding a node will not cause this issue as the rehash failover doesn’t occur until all responses are received.

These operations batch sizes are both controlled by the same value which can be configured by invoking distributedBatchSize method on the CacheStream. This value will default to the chunkSize configured in state transfer. Unfortunately this value is a tradeoff with memory usage vs performance vs at least once and your mileage may vary.

Using iterator with replicated and distributed caches

When a node is the primary or backup owner of all requested segments for a distributed stream, Infinispan performs the iterator or spliterator terminal operations locally, which optimizes performance as remote iterations are more resource intensive.

This optimization applies to both replicated and distributed caches. However, Infinispan performs iterations remotely when using cache stores that are both shared and have write-behind enabled. In this case performing the iterations remotely ensures consistency.

10.3.8. Intermediate operation exceptions

There are some intermediate operations that have special exceptions, these are skip, peek, sorted 1 2. & distinct. All of these methods have some sort of artificial iterator implanted in the stream processing to guarantee correctness, they are documented as below. Note this means these operations may cause possibly severe performance degradation.


An artificial iterator is implanted up to the intermediate skip operation. Then results are brought locally so it can skip the appropriate amount of elements.


WARNING: This operation requires having all entries in memory on the local node. An artificial iterator is implanted up to the intermediate sorted operation. All results are sorted locally. There are possible plans to have a distributed sort which returns batches of elements, but this is not yet implemented.


WARNING: This operation requires having all or nearly all entries in memory on the local node. Distinct is performed on each remote node and then an artificial iterator returns those distinct values. Then finally all of those results have a distinct operation performed upon them.

The rest of the intermediate operations are fully distributed as one would expect.

10.3.9. Examples

Word Count

Word count is a classic, if overused, example of map/reduce paradigm. Assume we have a mapping of key → sentence stored on Infinispan nodes. Key is a String, each sentence is also a String, and we have to count occurrence of all words in all sentences available. The implementation of such a distributed task could be defined as follows:

public class WordCountExample {

    * In this example replace c1 and c2 with
    * real Cache references
    * @param args
   public static void main(String[] args) {
      Cache<String, String> c1 = ...;
      Cache<String, String> c2 = ...;

      c1.put("1", "Hello world here I am");
      c2.put("2", "Infinispan rules the world");
      c1.put("3", "JUDCon is in Boston");
      c2.put("4", "JBoss World is in Boston as well");
      c1.put("12","JBoss Application Server");
      c2.put("15", "Hello world");
      c1.put("14", "Infinispan community");
      c2.put("15", "Hello world");

      c1.put("111", "Infinispan open source");
      c2.put("112", "Boston is close to Toronto");
      c1.put("113", "Toronto is a capital of Ontario");
      c2.put("114", "JUDCon is cool");
      c1.put("211", "JBoss World is awesome");
      c2.put("212", "JBoss rules");
      c1.put("213", "JBoss division of RedHat ");
      c2.put("214", "RedHat community");

      Map<String, Long> wordCountMap = c1.entrySet().parallelStream()
         .map(e -> e.getValue().split("\\s"))
         .collect(() -> Collectors.groupingBy(Function.identity(), Collectors.counting()));

In this case it is pretty simple to do the word count from the previous example.

However what if we want to find the most frequent word in the example? If you take a second to think about this case you will realize you need to have all words counted and available locally first. Thus we actually have a few options.

We could use a finisher on the collector, which is invoked on the user thread after all the results have been collected. Some redundant lines have been removed from the previous example.

public class WordCountExample {
   public static void main(String[] args) {
      // Lines removed

      String mostFrequentWord = c1.entrySet().parallelStream()
         .map(e -> e.getValue().split("\\s"))
         .collect(() -> Collectors.collectingAndThen(
            Collectors.groupingBy(Function.identity(), Collectors.counting()),
               wordCountMap -> {
                  String mostFrequent = null;
                  long maxCount = 0;
                     for (Map.Entry<String, Long> e : wordCountMap.entrySet()) {
                        int count = e.getValue().intValue();
                        if (count > maxCount) {
                           maxCount = count;
                           mostFrequent = e.getKey();
                     return mostFrequent;


Unfortunately the last step is only going to be ran in a single thread, which if we have a lot of words could be quite slow. Maybe there is another way to parallelize this with Streams.

We mentioned before we are in the local node after processing, so we could actually use a stream on the map results. We can therefore use a parallel stream on the results.

public class WordFrequencyExample {
   public static void main(String[] args) {
      // Lines removed

      Map<String, Long> wordCount = c1.entrySet().parallelStream()
              .map(e -> e.getValue().split("\\s"))
              .collect(() -> Collectors.groupingBy(Function.identity(), Collectors.counting()));
      Optional<Map.Entry<String, Long>> mostFrequent = wordCount.entrySet().parallelStream().reduce(
              (e1, e2) -> e1.getValue() > e2.getValue() ? e1 : e2);

This way you can still utilize all of the cores locally when calculating the most frequent element.

Remove specific entries

Distributed streams can also be used as a way to modify data where it lives. For example you may want to remove all entries in your cache that contain a specific word.

public class RemoveBadWords {
   public static void main(String[] args) {
      // Lines removed
      String word = ..

         .filter(e -> e.getValue().contains(word))
         .forEach((c, e) -> c.remove(e.getKey());

If we carefully note what is serialized and what is not, we notice that only the word along with the operations are serialized across to other nods as it is captured by the lambda. However the real saving piece is that the cache operation is performed on the primary owner thus reducing the amount of network traffic required to remove these values from the cache. The cache is not captured by the lambda as we provide a special BiConsumer method override that when invoked on each node passes the cache to the BiConsumer

One thing to keep in mind using the forEach command in this manner is that the underlying stream obtains no locks. The cache remove operation will still obtain locks naturally, but the value could have changed from what the stream saw. That means that the entry could have been changed after the stream read it but the remove actually removed it.

We have specifically added a new variant which is called LockedStream.

Plenty of other examples

The Streams API is a JRE tool and there are lots of examples for using it. Just remember that your operations need to be Serializable in some way.

10.4. Distributed Execution

Distributed Executor has been deprecated as of Infinispan 9.1. You should use either a Cluster Executor or Distributed Stream to perform the operations they were doing before.

Infinispan provides distributed execution through a standard JDK ExecutorService interface. Tasks submitted for execution, instead of being executed in a local JVM, are executed on an entire cluster of Infinispan nodes. Every DistributedExecutorService is bound to one particular cache. Tasks submitted will have access to key/value pairs from that particular cache if and only if the task submitted is an instance of DistributedCallable. Also note that there is nothing preventing users from submitting a familiar Runnable or Callable just like to any other ExecutorService. However, DistributedExecutorService, as it name implies, will likely migrate submitted Callable or Runnable to another JVM in Infinispan cluster, execute it and return a result to task invoker. Due to a potential task migration to other nodes every Callable, Runnable and/or DistributedCallable submitted must be either Serializable or Externalizable. Also the value returned from a callable must be Serializable or Externalizable as well. If the value returned is not serializable a NotSerializableException will be thrown.

Infinispan’s distributed task executors use data from Infinispan cache nodes as input for execution tasks. Most other distributed frameworks do not have that leverage and users have to specify input for distributed tasks from some well known location. Furthermore, users of Infinispan distributed execution framework do not have to configure store for intermediate and final results thus removing another layer of complexity and maintenance.

Our distributed execution framework capitalizes on the fact input data in Infinispan data grid is already load balanced (in case of DIST mode). Since input data is already balanced execution tasks will be automatically balanced as well; users do not have to explicitly assign work tasks to specific Infinispan nodes. However, our framework accommodates users to specify arbitrary subset of cache keys as input for distributed execution tasks.

10.4.1. DistributedCallable API

In case users needs access to Infinispan cache data for an execution of a task we recommend that you encapsulate task in DistributedCallable interface. DistributedCallable is a subtype of the existing Callable from java.util.concurrent package; DistributedCallable can be executed in a remote JVM and receive input from Infinispan cache. Task’s main algorithm could essentially remain unchanged, only the input source is changed. Existing Callable implementations most likely get their input in a form of some Java object/primitive while DistributedCallable gets its input from an Infinispan cache. Therefore, users who have already implemented Callable interface to describe their task units would simply extend DistributedCallable and use keys from Infinispan execution environment as input for the task. Implentation of DistributedCallable can in fact continue to support implementation of an already existing Callable while simultaneously be ready for distribited execution by extending DistributedCallable.

public interface DistributedCallable<K, V, T> extends Callable<T> {

    * Invoked by execution environment after DistributedCallable
    * has been migrated for execution to a specific node.
    * @param cache
    *           cache whose keys are used as input data for this
    *           DistributedCallable task
    * @param inputKeys
    *           keys used as input for this DistributedCallable task
   public void setEnvironment(Cache<K, V> cache, Set<K> inputKeys);


10.4.2. Callable and CDI

Users that do not want or can not implement DistributedCallable yet need a reference to input cache used in DistributedExecutorService have an option of the input cache being injected by CDI mechanism. Upon arrival of user’s Callable to an Infinispan executing node, Infinispan CDI mechanism will provide appropriate cache reference and inject it to executing Callable. All one has to do is to declare a Cache field in Callable and annotate it with org.infinispan.cdi.Input annotation along with mandatory @Inject annotation.

 public class CallableWithInjectedCache implements Callable<Integer>, Serializable {
      private Cache<String, String> cache;

      public Integer call() throws Exception {
        //use injected cache reference
        return 1;

10.4.3. DistributedExecutorService, DistributedTaskBuilder and DistributedTask API

DistributedExecutorService is a simple extension of a familiar ExecutorService from java.util.concurrent package. However, advantages of DistributedExecutorService are not to be overlooked. Existing Callable tasks, instead of being executed in JDK’s ExecutorService, are also eligible for execution on Infinispan cluster. Infinispan execution environment would migrate a task to execution node(s), run the task and return the result(s) to the calling node. Of course, not all Callable tasks would benefit from parallel distributed execution. Excellent candidates are long running and computationally intensive tasks that can run concurrently and/or tasks using input data that can be processed concurrently. For more details about good candidates for parallel execution and parallel algorithms in general refer to Introduction to Parallel Computing .

The second advantage of the DistributedExecutorService is that it allows a quick and simple implementation of tasks that take input from Infinispan cache nodes, execute certain computation and return results to the caller. Users would specify which keys to use as input for specified DistributedCallable and submit that callable for execution on Infinispan cluster. Infinispan runtime would locate the appriate keys, migrate DistributedCallable to target execution node(s) and finally return a list of results for each executed Callable. Of course, users can omit specifying input keys in which case Infinispan would execute DistributedCallable on all keys for a specified cache.

Lets see how we can use DistributedExecutorService If you already have Callable/Runnable tasks defined! Well, simply submit them to an instance of DefaultExecutorService for execution!

ExecutorService des = new DefaultExecutorService(cache);
Future<Boolean> future = des.submit(new SomeCallable());
Boolean r = future.get();

In case you need to specify more task parameters like task timeout, custom failover policy or execution policy use DistributedTaskBuilder and DistributedTask API.

DistributedExecutorService des = new DefaultExecutorService(cache);
DistributedTaskBuilder<Boolean> taskBuilder = des.createDistributedTaskBuilder(new SomeCallable());
DistributedTask<Boolean> distributedTask = taskBuilder.build();
Future<Boolean> future = des.submit(distributedTask);
Boolean r = future.get();

10.4.4. Distributed task failover

Distributed execution framework supports task failover. By default no failover policy is installed and task’s Runnable/Callable/DistributedCallable will simply fail. Failover mechanism is invoked in the following cases:

a) Failover due to a node failure where task is executing

b) Failover due to a task failure (e.g. Callable task throws Exception).

Infinispan provides random node failover policy which will attempt execution of a part of distributed task on another random node, if such node is available. However, users that have a need to implement a more sophisticated failover policy can implement DistributedTaskFailoverPolicy interface. For example, users might want to use consistent hashing (CH) mechanism for failover of uncompleted tasks. CH based failover might for example migrate failed task T to cluster node(s) having a backup of input data that was executed on a failed node F.

 * DistributedTaskFailoverPolicy allows pluggable fail over target selection for a failed remotely
 * executed distributed task.
public interface DistributedTaskFailoverPolicy {

    * As parts of distributively executed task can fail due to the task itself throwing an exception
    * or it can be a system caused failure (e.g node failed or left cluster during task
    * execution etc).
    * @param failoverContext
    *           the FailoverContext of the failed execution
    * @return result the Address of the node selected for fail over execution
   Address failover(FailoverContext context);

    * Maximum number of fail over attempts permitted by this DistributedTaskFailoverPolicy
    * @return max number of fail over attempts
   int maxFailoverAttempts();

Therefore one could for example specify random failover execution policy simply by:

DistributedExecutorService des = new DefaultExecutorService(cache);
DistributedTaskBuilder<Boolean> taskBuilder = des.createDistributedTaskBuilder(new SomeCallable());
DistributedTask<Boolean> distributedTask = taskBuilder.build();
Future<Boolean> future = des.submit(distributedTask);
Boolean r = future.get();

10.4.5. Distributed task execution policy

DistributedTaskExecutionPolicy is an enum that allows tasks to specify its custom task execution policy across Infinispan cluster. DistributedTaskExecutionPolicy effectively scopes execution of tasks to a subset of nodes. For example, someone might want to exclusively execute tasks on a local network site instead of a backup remote network centre as well. Others might, for example, use only a dedicated subset of a certain Infinispan rack nodes for specific task execution. DistributedTaskExecutionPolicy is set per instance of DistributedTask.

DistributedExecutorService des = new DefaultExecutorService(cache);
DistributedTaskBuilder<Boolean> taskBuilder = des.createDistributedTaskBuilder(new SomeCallable());
DistributedTask<Boolean> distributedTask = taskBuilder.build();
Future<Boolean> future = des.submit(distributedTask);
Boolean r = future.get();

10.4.6. Examples

Pi approximation can greatly benefit from parallel distributed execution in DistributedExecutorService. Recall that area of the square is Sa = 4r2 and area of the circle is Ca=pi*r2. Substituting r2 from the second equation into the first one it turns out that pi = 4 * Ca/Sa. Now, image that we can shoot very large number of darts into a square; if we take ratio of darts that land inside a circle over a total number of darts shot we will approximate Ca/Sa value. Since we know that pi = 4 * Ca/Sa we can easily derive approximate value of pi. The more darts we shoot the better approximation we get. In the example below we shoot 10 million darts but instead of "shooting" them serially we parallelize work of dart shooting across entire Infinispan cluster.

  public class PiAppx {

   public static void main (String [] arg){
      List<Cache> caches = ...;
      Cache cache = ...;

      int numPoints = 10000000;
      int numServers = caches.size();
      int numberPerWorker = numPoints / numServers;

      DistributedExecutorService des = new DefaultExecutorService(cache);
      long start = System.currentTimeMillis();
      CircleTest ct = new CircleTest(numberPerWorker);
      List<Future<Integer>> results = des.submitEverywhere(ct);
      int countCircle = 0;
      for (Future<Integer> f : results) {
         countCircle += f.get();
      double appxPi = 4.0 * countCircle / numPoints;

      System.out.println("Distributed PI appx is " + appxPi +
      " completed in " + (System.currentTimeMillis() - start) + " ms");

   private static class CircleTest implements Callable<Integer>, Serializable {

      /** The serialVersionUID */
      private static final long serialVersionUID = 3496135215525904755L;

      private final int loopCount;

      public CircleTest(int loopCount) {
         this.loopCount = loopCount;

      public Integer call() throws Exception {
         int insideCircleCount = 0;
         for (int i = 0; i < loopCount; i++) {
            double x = Math.random();
            double y = Math.random();
            if (insideCircle(x, y))
         return insideCircleCount;

      private boolean insideCircle(double x, double y) {
         return (Math.pow(x - 0.5, 2) + Math.pow(y - 0.5, 2))
         <= Math.pow(0.5, 2);

11. Indexing and Querying

11.1. Overview

Infinispan supports indexing and searching of Java Pojo(s) or objects encoded via Protocol Buffers stored in the grid using powerful search APIs which complement its main Map-like API.

Querying is possible both in library and client/server mode (for Java, C#, Node.js and other clients), and Infinispan can index data using Apache Lucene, offering an efficient full-text capable search engine in order to cover a wide range of data retrieval use cases.

Indexing configuration relies on a schema definition, and for that Infinispan can use annotated Java classes when in library mode, and protobuf schemas for remote clients written in other languages. By standardizing on protobuf, Infinispan allows full interoperability between Java and non-Java clients.

Apart from indexed queries, Infinispan can run queries over non-indexed data (indexless queries) and over partially indexed data (hybrid queries).

In terms of Search APIs, Infinispan has its own query language called Ickle, which is string-based and adds support for full-text querying. The Query DSL can be used for both embedded and remote java clients when full-text is not required; for Java embedded clients Infinispan offers the Hibernate Search Query API which supports running Lucene queries in the grid, apart from advanced search capabilities like Faceted and Spatial search.

Finally, Infinispan has support for Continuous Queries, which works in a reverse manner to the other APIs: instead of creating, executing a query and obtain results, it allows a client to register queries that will be evaluated continuously as data in the cluster changes, generating notifications whenever the changed data matches the queries.

11.2. Embedded Querying

Embedded querying is available when Infinispan is used as a library. No protobuf mapping is required, and both indexing and searching are done on top of Java objects. When in library mode, it is possible to run Lucene queries directly and use all the available Query APIs and it also allows flexible indexing configurations to keep latency to a minimal.

11.2.1. Quick example

We’re going to store Book instances in an Infinispan cache called "books". Book instances will be indexed, so we enable indexing for the cache, letting Infinispan configure the indexing automatically:

Infinispan configuration:

        <transport cluster="infinispan-cluster"/>
        <distributed-cache name="books">
            <indexing index="LOCAL" auto-config="true"/>

Obtaining the cache:

import org.infinispan.Cache;
import org.infinispan.manager.DefaultCacheManager;
import org.infinispan.manager.EmbeddedCacheManager;

EmbeddedCacheManager manager = new DefaultCacheManager("infinispan.xml");
Cache<String, Book> cache = manager.getCache("books");

Each Book will be defined as in the following example; we have to choose which properties are indexed, and for each property we can optionally choose advanced indexing options using the annotations defined in the Hibernate Search project.

import org.hibernate.search.annotations.*;
import java.util.Date;
import java.util.HashSet;
import java.util.Set;

//Values you want to index need to be annotated with @Indexed, then you pick which fields and how they are to be indexed:
public class Book {
   @Field String title;
   @Field String description;
   @Field @DateBridge(resolution=Resolution.YEAR) Date publicationYear;
   @IndexedEmbedded Set<Author> authors = new HashSet<Author>();
public class Author {
   @Field String name;
   @Field String surname;
   // hashCode() and equals() omitted

Now assuming we stored several Book instances in our Infinispan Cache , we can search them for any matching field as in the following example.

Using a Lucene Query:

// get the search manager from the cache:
SearchManager searchManager = org.infinispan.query.Search.getSearchManager(cache);

// create any standard Lucene query, via Lucene's QueryParser or any other means:
org.apache.lucene.search.Query fullTextQuery = //any Apache Lucene Query

// convert the Lucene query to a CacheQuery:
CacheQuery cacheQuery = searchManager.getQuery( fullTextQuery );

// get the results:
List<Object> found = cacheQuery.list();

A Lucene Query is often created by parsing a query in text format such as "title:infinispan AND authors.name:sanne", or by using the query builder provided by Hibernate Search.

// get the search manager from the cache:
SearchManager searchManager = org.infinispan.query.Search.getSearchManager( cache );

// you could make the queries via Lucene APIs, or use some helpers:
QueryBuilder queryBuilder = searchManager.buildQueryBuilderForClass(Book.class).get();

// the queryBuilder has a nice fluent API which guides you through all options.
// this has some knowledge about your object, for example which Analyzers
// need to be applied, but the output is a fairly standard Lucene Query.
org.apache.lucene.search.Query luceneQuery = queryBuilder.phrase()
                  .sentence("a book on highly scalable query engines")

// the query API itself accepts any Lucene Query, and on top of that
// you can restrict the result to selected class types:
CacheQuery query = searchManager.getQuery(luceneQuery, Book.class);

// and there are your results!
List objectList = query.list();

for (Object book : objectList) {

Apart from list() you have the option for streaming results, or use pagination.

For searches that do not require Lucene or full-text capabilities and are mostly about aggregation and exact matches, we can use the Infinispan Query DSL API:

import org.infinispan.query.dsl.QueryFactory;
import org.infinispan.query.dsl.Query;
import org.infinispan.query.Search;

// get the query factory:
QueryFactory queryFactory = Search.getQueryFactory(cache);

Query q = queryFactory.from(Book.class)

List<Book> list = q.list();

Finally, we can use an Ickle query directly, allowing for Lucene syntax in one or more predicates:

import org.infinispan.query.dsl.QueryFactory;
import org.infinispan.query.dsl.Query;

// get the query factory:
QueryFactory queryFactory = Search.getQueryFactory(cache);

Query q = queryFactory.create("from Book b where b.author.name = 'Stephen' and " +
                "b.description : (+'dark' -'tower')");

List<Book> list = q.list();

11.2.2. Indexing

Indexing in Infinispan happens on a per-cache basis and by default a cache is not indexed. Enabling indexing is not mandatory but queries using an index will have a vastly superior performance. On the other hand, enabling indexing can impact negatively the write throughput of a cluster, so make sure to check the query performance guide for some strategies to minimize this impact depending on the cache type and use case.

General format

To enable indexing via XML, you need to add the <indexing> element plus the index (index mode) to your cache configuration, and optionally pass additional properties.

   <cache-container default-cache="default">
      <replicated-cache name="default">
         <indexing index="ALL">
            <property name="property.name">some value</property>


import org.infinispan.configuration.cache.*;

ConfigurationBuilder cacheCfg = ...
      .addProperty("property name", "propery value")
Index names

Each property inside the index element is prefixed with the index name, for the index named org.infinispan.sample.Car the directory_provider is local-heap:

    <indexing index="ALL">
        <property name="org.infinispan.sample.Car.directory_provider">local-heap</property>
      .addProperty("org.infinispan.sample.Car.directory_provider", "local-heap")

Infinispan creates an index for each entity existent in a cache, and it allows to configure those indexes independently. For a class annotated with @Indexed, the index name is the fully qualified class name, unless overridden with the name argument in the annotation.

In the snippet below, the default storage for all entities is infinispan, but Boat instances will be stored on local-heap in an index named boatIndex. Airplane entities will also be stored in local-heap. Any other entity’s index will be configured with the property prefixed by default.

package org.infinispan.sample;

@Indexed(name = "boatIndex")
public class Boat {


public class Airplane {

    <indexing index="ALL">
        <property name="default.directory_provider">infinispan</property>
        <property name="boatIndex.directory_provider">local-heap</property>
        <property name="org.infinispan.sample.Airplane.directory_provider">
Specifying indexed Entities

Infinispan can automatically recognize and manage indexes for different entity types in a cache. Future versions of Infinispan will remove this capability so it’s recommended to declare upfront which types are going to be indexed (list them by their fully qualified class name). This can be done via xml:

   <cache-container default-cache="default">
      <replicated-cache name="default">
         <indexing index="ALL">

or programmatically:


In server mode, the class names listed under the 'indexed-entities' element must use the 'extended' class name format which is composed of a JBoss Modules module identifier, a slot name, and the fully qualified class name, these three components being separated by the ':' character, (eg. "com.acme.my-module-with-entity-classes:my-slot:com.acme.query.test.Car"). The entity classes must be located in the referenced module, which can be either a user supplied module deployed in the 'modules' folder of your server or a plain jar deployed in the 'deployments' folder. The module in question will become an automatic dependency of your Cache, so its eventual redeployment will cause the cache to be restarted.

Only for server, if you fail to follow the requirement of using 'extended' class names and use a plain class name its resolution will fail due to missing class because the wrong ClassLoader is being used (the Infinispan’s internal class path is being used).
Index mode

An Infinispan node typically receives data from two sources: local and remote. Local translates to clients manipulating data using the map API in the same JVM; remote data comes from other Infinispan nodes during replication or rebalancing.

The index mode configuration defines, from a node in the cluster point of view, which data gets indexed.

Possible values:

  • ALL: all data is indexed, local and remote.

  • LOCAL: only local data is indexed.

  • PRIMARY_OWNER: Only entries containing keys that the node is primary owner will be indexed, regardless of local or remote origin.

  • NONE: no data is indexed. Equivalent to not configure indexing at all.

Index Managers

Index managers are central components in Infinispan Querying responsible for the indexing configuration, distribution and internal lifecycle of several query components such as Lucene’s IndexReader and IndexWriter. Each Index Manager is associated with a Directory Provider, which defines the physical storage of the index.

Regarding index distribution, Infinispan can be configured with shared or non-shared indexes.

Shared indexes

A shared index is a single, distributed, cluster-wide index for a certain cache. The main advantage is that the index is visible from every node and can be queried as if the index were local, there is no need to broadcast queries to all members and aggregate the results. The downside is that Lucene does not allow more than a single process writing to the index at the same time, and the coordination of lock acquisitions needs to be done by a proper shared index capable index manager. In any case, having a single write lock cluster-wise can lead to some degree of contention under heavy writing.

Infinispan supports shared indexes leveraging the Infinispan Directory Provider, which stores indexes in a separate set of caches. Two index managers are available to use shared indexes: InfinispanIndexManager and AffinityIndexManager.

Effect of the index mode

Shared indexes should not use the ALL index mode since it’d lead to redundant indexing: since there is a single index cluster wide, the entry would get indexed when inserted via Cache API, and another time when Infinispan replicates it to another node. The ALL mode is usually associates with non-shared indexes in order to create full index replicas on each node.


This index manager uses the Infinispan Directory Provider, and is suitable for creating shared indexes. Index mode should be set to LOCAL in this configuration.


<distributed-cache name="default" >
    <indexing index="LOCAL">
        <property name="default.indexmanager">
        <!-- optional: tailor each index cache -->
        <property name="default.locking_cachename">LuceneIndexesLocking_custom</property>
        <property name="default.data_cachename">LuceneIndexesData_custom</property>
        <property name="default.metadata_cachename">LuceneIndexesMetadata_custom</property>

<!-- Optional -->
<replicated-cache name="LuceneIndexesLocking_custom">
    <indexing index="NONE" />
    <-- extra configuration -->

<!-- Optional -->
<replicated-cache name="LuceneIndexesMetadata_custom">
    <indexing index="NONE" />
    <-- extra configuration -->

<!-- Optional -->
<distributed-cache name="LuceneIndexesData_custom">
    <-- extra configuration -->
    <indexing index="NONE" />

Indexes are stored in a set of clustered caches, called by default LuceneIndexesData, LuceneIndexesMetadata and LuceneIndexesLocking.

The LuceneIndexesLocking cache is used to store Lucene locks, and it is a very small cache: it will contain one entry per entity (index).

The LuceneIndexesMetadata cache is used to store info about the logical files that are part of the index, such as names, chunks and sizes and it is also small in size.

The LuceneIndexesData cache is where most of the index is located: it is much bigger then the other two but should be smaller than the data in the cache itself, thanks to Lucene’s efficient storing techniques.

It’s not necessary to redefine the configuration of those 3 cases, Infinispan will pick sensible defaults. Reasons re-define them would be performance tuning for a specific scenario, or for example to make them persistent by configuring a cache store.

In order to avoid index corruption when two or more nodes of the cluster try to write to the index at the same time, the InfinispanIndexManager internally elects a master in the cluster (which is the JGroups coordinator) and forwards all indexing works to this master.


The AffinityIndexManager is an experimental index manager used for shared indexes that also stores indexes using the Infinispan Directory Provider. Unlike the InfinispanIndexManager, it does not have a single node (master) that handles all the indexing cluster wide, but rather splits the index using multiple shards, each shard being responsible for indexing data associated with one or more Infinispan segments. For an in-depth description of the inner workings, please see the design doc.

The PRIMARY_OWNER index mode is required, together with a special kind of KeyPartitioner.

XML Configuration:

<distributed-cache name="default"
    <indexing index="PRIMARY_OWNER">
        <property name="default.indexmanager">
        <!-- optional: control the number of shards, the default is 4 -->
        <property name="default.sharding_strategy.nbr_of_shards">10</property>


import org.infinispan.distribution.ch.impl.AffinityPartitioner;
import org.infinispan.query.affinity.AffinityIndexManager;

ConfigurationBuilder cacheCfg = ...
cacheCfg.clustering().hash().keyPartitioner(new AffinityPartitioner());
      .addProperty("default.indexmanager", AffinityIndexManager.class.getName())
      .addProperty("default.sharding_strategy.nbr_of_shards", "10")

The AffinityIndexManager by default will have as many shards as Infinispan segments, but this value is configurable as seen in the example above.

The number of shards affects directly the query performance and writing throughput: generally speaking, a high number of shards offers better write throughput but has an adverse effect on query performance.

Non-shared indexes

Non-shared indexes are independent indexes at each node. This setup is particularly advantageous for replicated caches where each node has all the cluster data and thus can hold all the indexes as well, offering optimal query performance with zero network latency when querying. Another advantage is, since the index is local to each node, there is less contention during writes due to the fact that each node is subjected to its own index lock, not a cluster wide one.

Since each node might hold a partial index, it may be necessary to link#query_clustered_query_api[broadcast] queries in order to get correct search results, which can add latency. If the cache is REPL, though, the broadcast is not necessary: each node can hold a full local copy of the index and queries runs at optimal speed taking advantage of a local index.

Infinispan has two index managers suitable for non-shared indexes: directory-based and near-real-time. Storage wise, non-shared indexes can be located in ram, filesystem, or Infinispan local caches.

Effect of the index mode

The directory-based and near-real-time index managers can be associated with different index modes, resulting in different index distributions.

REPL caches combined with the ALL index mode will result in a full copy of the cluster-wide index on each node. This mode allows queries to become effectively local without network latency. This is the recommended mode to index any REPL cache, and that’s the choice picked by the auto-config when the a REPL cache is detected. The ALL mode should not be used with DIST caches.

REPL or DIST caches combined with LOCAL index mode will cause each node to index only data inserted from the same JVM, causing an uneven distribution of the index. In order to obtain correct query results, it’s necessary to use broadcast queries.

REPL or DIST caches combined with PRIMARY_OWNER will also need broadcast queries. Differently from the LOCAL mode, each node’s index will contain indexed entries which key is primarily owned by the node according to the consistent hash, leading to a more evenly distributed indexes among the nodes.

directory-based index manager

This is the default Index Manager used when no index manager is configured. The directory-based index manager is used to manage indexes backed by a local lucene directory. It supports ram, filesystem and non-clustered infinispan storage.

Filesystem storage

This is the default storage, and used when index manager configuration is omitted. The index is stored in the filesystem using a MMapDirectory. It is the recommended storage for local indexes. Although indexes are persistent on disk, they get memory mapped by Lucene and thus offer decent query performance.


<replicated-cache name="myCache">
   <indexing index="ALL">
      <!-- Optional: define base folder for indexes -->
      <property name="default.indexBase">${java.io.tmpdir}/baseDir</property>

Infinispan will create a different folder under default.indexBase for each entity (index) present in the cache.

Ram storage

Index is stored in memory using a Lucene RAMDirectory. Not recommended for large indexes or highly concurrent situations. Indexes stored in Ram are not persistent, so after a cluster shutdown a re-index is needed. Configuration:

<replicated-cache name="myCache">
   <indexing index="ALL">
      <property name="default.directory_provider">local-heap</property>

Infinispan storage

Infinispan storage makes use of the Infinispan Lucene directory that saves the indexes to a set of caches; those caches can be configured like any other Infinispan cache, for example by adding a cache store to have indexes persisted elsewhere apart from memory. In order to use Infinispan storage with a non-shared index, it’s necessary to use LOCAL caches for the indexes:

<replicated-cache name="default">
    <indexing index="ALL">
        <property name="default.locking_cachename">LuceneIndexesLocking_custom</property>
        <property name="default.data_cachename">LuceneIndexesData_custom</property>
        <property name="default.metadata_cachename">LuceneIndexesMetadata_custom</property>

<local-cache name="LuceneIndexesLocking_custom">
    <indexing index="NONE" />

<local-cache name="LuceneIndexesMetadata_custom">
    <indexing index="NONE" />

<local-cache name="LuceneIndexesData_custom">
    <indexing index="NONE" />
near-real-time index manager

Similar to the directory-based index manager but takes advantage of the Near-Real-Time features of Lucene. It has better write performance than the directory-based because it flushes the index to the underlying store less often. The drawback is that unflushed index changes can be lost in case of a non-clean shutdown. Can be used in conjunction with local-heap, filesystem and local infinispan storage. Configuration for each different storage type is the same as the directory-based index manager.

Example with ram:

<replicated-cache name="default">
    <indexing index="ALL">
        <property name="default.indexmanager">near-real-time</property>
        <property name="default.directory_provider">local-heap</property>

Example with filesystem:

<replicated-cache name="default">
    <indexing index="ALL">
        <property name="default.indexmanager">near-real-time</property>
External indexes

Apart from having shared and non-shared indexes managed by Infinispan itself, it is possible to offload indexing to a third party search engine: currently Infinispan supports Elasticsearch as a external index storage.

Elasticsearch IndexManager (experimental)

This index manager forwards all indexes to an external Elasticsearch server. This is an experimental integration and some features may not be available, for example indexNullAs for @IndexedEmbedded annotations is not currently supported.


<indexing index="LOCAL">
    <property name="default.indexmanager">elasticsearch</property>
    <property name="default.elasticsearch.host">link:http://elasticHost:9200</property>
    <!-- other elasticsearch configurations -->

The index mode should be set to LOCAL, since Infinispan considers Elasticsearch as a single shared index. More information about Elasticsearch integration, including the full description of the configuration properties can be found at the Hibernate Search manual.

Automatic configuration

The attribute auto-config provides a simple way of configuring indexing based on the cache type. For replicated and local caches, the indexing is configured to be persisted on disk and not shared with any other processes. Also, it is configured so that minimum delay exists between the moment an object is indexed and the moment it is available for searches (near real time).

      <local-cache name="default">
         <indexing index="LOCAL" auto-config="true">
it is possible to redefine any property added via auto-config, and also add new properties, allowing for advanced tuning.

The auto config adds the following properties for replicated and local caches:

Property name value description



Filesystem based index. More details at Hibernate Search documentation



indexing operation in exclusive mode, allowing Hibernate Search to optimize writes



make use of Lucene near real time feature, meaning indexed objects are promptly available to searches



Reuse index reader across several queries, thus avoiding reopening it

For distributed caches, the auto-config configure indexes in Infinispan itself, internally handled as a master-slave mechanism where indexing operations are sent to a single node which is responsible to write to the index.

The auto config properties for distributed caches are:

Property name value description



Indexes stored in Infinispan. More details at Hibernate Search documentation



indexing operation in exclusive mode, allowing Hibernate Search to optimize writes



Delegates index writing to a single node in the Infinispan cluster



Reuse index reader across several queries, avoiding reopening it


Occasionally you might need to rebuild the Lucene index by reconstructing it from the data stored in the Cache. You need to rebuild the index if you change the definition of what is indexed on your types, or if you change for example some Analyzer parameter, as Analyzers affect how the index is written. Also, you might need to rebuild the index if you had it destroyed by some system administration mistake. To rebuild the index just get a reference to the MassIndexer and start it; beware it might take some time as it needs to reprocess all data in the grid!

// Blocking execution
SearchManager searchManager = Search.getSearchManager(cache);

// Non blocking execution
CompletableFuture<Void> future = searchManager.getMassIndexer().startAsyc();
This is also available as a start JMX operation on the MassIndexer MBean registered under the name org.infinispan:type=Query,manager="{name-of-cache-manager}",cache="{name-of-cache}",component=MassIndexer.
Mapping Entities

Infinispan relies on the rich API of Hibernate Search in order to define fine grained configuration for indexing at entity level. This configuration includes which fields are annotated, which analyzers should be used, how to map nested objects and so on. Detailed documentation is available at the Hibernate Search manual.


Unlike Hibernate Search, using @DocumentId to mark a field as identifier does not apply to Infinispan values; in Infinispan the identifier for all @Indexed objects is the key used to store the value. You can still customize how the key is indexed using a combination of @Transformable , custom types and custom FieldBridge implementations.

@Transformable keys

The key for each value needs to be indexed as well, and the key instance must be transformed in a String. Infinispan includes some default transformation routines to encode common primitives, but to use a custom key you must provide an implementation of org.infinispan.query.Transformer .

Registering a Transformer via annotations

You can annotate your key type with org.infinispan.query.Transformable :

@Transformable(transformer = CustomTransformer.class)
public class CustomKey {

public class CustomTransformer implements Transformer {
   public Object fromString(String s) {
      return new CustomKey(...);

   public String toString(Object customType) {
      CustomKey ck = (CustomKey) customType;
      return ...

Registering a Transformer programmatically

Using this technique, you don’t have to annotate your custom key type:

org.infinispan.query.SearchManager.registerKeyTransformer(Class<?>, Class<? extends Transformer>)
Programmatic mapping

Instead of using annotations to map an entity to the index, it’s also possible to configure it programmatically.

In the following example we map an object Author which is to be stored in the grid and made searchable on two properties but without annotating the class.

import org.apache.lucene.search.Query;
import org.hibernate.search.cfg.Environment;
import org.hibernate.search.cfg.SearchMapping;
import org.hibernate.search.query.dsl.QueryBuilder;
import org.infinispan.Cache;
import org.infinispan.configuration.cache.Configuration;
import org.infinispan.configuration.cache.ConfigurationBuilder;
import org.infinispan.configuration.cache.Index;
import org.infinispan.manager.DefaultCacheManager;
import org.infinispan.query.CacheQuery;
import org.infinispan.query.Search;
import org.infinispan.query.SearchManager;

import java.io.IOException;
import java.lang.annotation.ElementType;
import java.util.Properties;

SearchMapping mapping = new SearchMapping();
       .property("name", ElementType.METHOD).field()
       .property("surname", ElementType.METHOD).field();

Properties properties = new Properties();
properties.put(Environment.MODEL_MAPPING, mapping);
properties.put("hibernate.search.[other options]", "[...]");

Configuration infinispanConfiguration = new ConfigurationBuilder()

DefaultCacheManager cacheManager = new DefaultCacheManager(infinispanConfiguration);

Cache<Long, Author> cache = cacheManager.getCache();
SearchManager sm = Search.getSearchManager(cache);

Author author = new Author(1, "Manik", "Surtani");
cache.put(author.getId(), author);

QueryBuilder qb = sm.buildQueryBuilderForClass(Author.class).get();
Query q = qb.keyword().onField("name").matching("Manik").createQuery();
CacheQuery cq = sm.getQuery(q, Author.class);
assert cq.getResultSize() == 1;

11.2.3. Querying APIs

You can query Infinispan using:

  • Lucene or Hibernate Search Queries. Infinispan exposes the Hibernate Search DSL, which produces Lucene queries. You can run Lucene queries on single nodes or broadcast queries to multiple nodes in an Infinispan cluster.

  • Ickle queries, a custom string-based query language with full-text extensions.

Hibernate Search

Apart from supporting Hibernate Search annotations to configure indexing, it’s also possible to query the cache using other Hibernate Search APIs

Running Lucene queries

To run a Lucene query directly, simply create and wrap it in a CacheQuery:

import org.infinispan.query.Search;
import org.infinispan.query.SearchManager;
import org.apache.lucene.Query;

SearchManager searchManager = Search.getSearchManager(cache);
Query query = searchManager.buildQueryBuilderForClass(Book.class).get()
CacheQuery<Book> cacheQuery = searchManager.getQuery(query);
Using the Hibernate Search DSL

The Hibernate Search DSL can be used to create the Lucene Query, example:

import org.infinispan.query.Search;
import org.infinispan.query.SearchManager;
import org.apache.lucene.search.Query;

Cache<String, Book> cache = ...

SearchManager searchManager = Search.getSearchManager(cache);

Query luceneQuery = searchManager

List<Object> results = searchManager.getQuery(luceneQuery).list();

For a detailed description of the query capabilities of this DSL, see the relevant section of the Hibernate Search manual.

Infinispan support Faceted Searches by using the Hibernate Search FacetManager:

// Cache is indexed
Cache<Integer, Book> cache = ...

// Obtain the Search Manager
SearchManager searchManager = Search.getSearchManager(cache);

// Create the query builder
QueryBuilder queryBuilder = searchManager.buildQueryBuilderForClass(Book.class).get();

// Build any Lucene Query. Here it's using the DSL to do a Lucene term query on a book name
Query luceneQuery = queryBuilder.keyword().wildcard().onField("name").matching("bitcoin").createQuery();

// Wrap into a cache Query
CacheQuery<Book> query = searchManager.getQuery(luceneQuery);

// Define the Facet characteristics
FacetingRequest request = queryBuilder.facet()

// Associated the FacetRequest with the query
FacetManager facetManager = query.getFacetManager().enableFaceting(request);

// Obtain the facets
List<Facet> facetList = facetManager.getFacets("year_facet");

A Faceted search like above will return the number books that match 'bitcoin' released on a yearly basis, for example:

AbstractFacet{facetingName='year_facet', fieldName='year', value='2008', count=1}
AbstractFacet{facetingName='year_facet', fieldName='year', value='2009', count=1}
AbstractFacet{facetingName='year_facet', fieldName='year', value='2010', count=1}
AbstractFacet{facetingName='year_facet', fieldName='year', value='2011', count=1}
AbstractFacet{facetingName='year_facet', fieldName='year', value='2012', count=1}
AbstractFacet{facetingName='year_facet', fieldName='year', value='2016', count=1}
AbstractFacet{facetingName='year_facet', fieldName='year', value='2015', count=2}
AbstractFacet{facetingName='year_facet', fieldName='year', value='2013', count=3}

For more info about Faceted Search, see Hibernate Search Faceting

Spatial Queries

Infinispan also supports Spatial Queries, allowing to combining full-text with restrictions based on distances, geometries or geographic coordinates.

Example, we start by using the @Spatial annotation in our entity that will be searched, together with @Latitude and @Longitude:

public class Restaurant {

      private Double latitude;

      private Double longitude;

      @Field(store = Store.YES)
      String name;

      // Getters, Setters and other members omitted


to run spatial queries, the Hibernate Search DSL can be used:

// Cache is configured as indexed
Cache<String, Restaurant> cache = ...

// Obtain the SearchManager
Searchmanager searchManager = Search.getSearchManager(cache);

// Build the Lucene Spatial Query
Query query = Search.getSearchManager(cache).buildQueryBuilderForClass(Restaurant.class).get()
            .within( 2, Unit.KM )
              .ofLatitude( centerLatitude )
              .andLongitude( centerLongitude )

// Wrap in a cache Query
CacheQuery<Restaurant> cacheQuery = searchManager.getQuery(query);

List<Restaurant> nearBy = cacheQuery.list();

It’s possible to specify a query mode for indexed queries. IndexedQueryMode.BROADCAST allows to broadcast a query to each node of the cluster, retrieve the results and combine them before returning to the caller. It is suitable for use in conjunction with non-shared indexes, since each node’s local index will have only a subset of the data indexed.

IndexedQueryMode.FETCH will execute the query in the caller. If all the indexes for the cluster wide data are available locally, performance will be optimal, otherwise this query mode may involve fetching indexes data from remote nodes.

The IndexedQueryMode is supported for Lucene Queries and Ickle String queries at the moment (no Infinispan Query DSL).


CacheQuery<Person> broadcastQuery = Search.getSearchManager(cache).getQuery(new MatchAllDocsQuery(), IndexedQueryMode.BROADCAST);

List<Person> result = broadcastQuery.list();
Infinispan Query DSL

Infinispan provides its own query DSL, independent of Lucene and Hibernate Search. Decoupling the query API from the underlying query and indexing mechanism makes it possible to introduce new alternative engines in the future, besides Lucene, and still being able to use the same uniform query API. The current implementation of indexing and searching is still based on Hibernate Search and Lucene so all indexing related aspects presented in this chapter still apply.

The new API simplifies the writing of queries by not exposing the user to the low level details of constructing Lucene query objects and also has the advantage of being available to remote Hot Rod clients. But before delving into further details, let’s examine first a simple example of writing a query for the Book entity from the previous example.

Query example using Infinispan's query DSL
import org.infinispan.query.dsl.*;

// get the DSL query factory from the cache, to be used for constructing the Query object:
QueryFactory qf = org.infinispan.query.Search.getQueryFactory(cache);

// create a query for all the books that have a title which contains "engine":
org.infinispan.query.dsl.Query query = qf.from(Book.class)

// get the results:
List<Book> list = query.list();

The API is located in the org.infinispan.query.dsl package. A query is created with the help of the QueryFactory instance which is obtained from the per-cache SearchManager. Each QueryFactory instance is bound to the same Cache instance as the SearchManager, but it is otherwise a stateless and thread-safe object that can be used for creating multiple queries in parallel.

Query creation starts with the invocation of the from(Class entityType) method which returns a QueryBuilder object that is further responsible for creating queries targeted to the specified entity class from the given cache.

A query will always target a single entity type and is evaluated over the contents of a single cache. Running a query over multiple caches or creating queries that target several entity types (joins) is not supported.

The QueryBuilder accumulates search criteria and configuration specified through the invocation of its DSL methods and is ultimately used to build a Query object by the invocation of the QueryBuilder.build() method that completes the construction. Being a stateful object, it cannot be used for constructing multiple queries at the same time (except for nested queries) but can be reused afterwards.

This QueryBuilder is different from the one from Hibernate Search but has a somewhat similar purpose, hence the same name. We are considering renaming it in near future to prevent ambiguity.

Executing the query and fetching the results is as simple as invoking the list() method of the Query object. Once executed the Query object is not reusable. If you need to re-execute it in order to obtain fresh results then a new instance must be obtained by calling QueryBuilder.build().

Filtering operators

Constructing a query is a hierarchical process of composing multiple criteria and is best explained following this hierarchy.

The simplest possible form of a query criteria is a restriction on the values of an entity attribute according to a filtering operator that accepts zero or more arguments. The entity attribute is specified by invoking the having(String attributePath) method of the query builder which returns an intermediate context object (FilterConditionEndContext) that exposes all the available operators. Each of the methods defined by FilterConditionEndContext is an operator that accepts an argument, except for between which has two arguments and isNull which has no arguments. The arguments are statically evaluated at the time the query is constructed, so if you’re looking for a feature similar to SQL’s correlated sub-queries, that is not currently available.

// a single query criterion
QueryBuilder qb = ...
qb.having("title").eq("Hibernate Search in Action");
Table 5. FilterConditionEndContext exposes the following filtering operators:
Filter Arguments Description


Collection values

Checks that the left operand is equal to one of the elements from the Collection of values given as argument.


Object…​ values

Checks that the left operand is equal to one of the (fixed) list of values given as argument.


Object value

Checks that the left argument (which is expected to be an array or a Collection) contains the given element.


Collection values

Checks that the left argument (which is expected to be an array or a Collection) contains all the elements of the given collection, in any order.


Object…​ values

Checks that the left argument (which is expected to be an array or a Collection) contains all of the the given elements, in any order.


Collection values

Checks that the left argument (which is expected to be an array or a Collection) contains any of the elements of the given collection.


Object…​ values

Checks that the left argument (which is expected to be an array or a Collection) contains any of the the given elements.


Checks that the left argument is null.


String pattern

Checks that the left argument (which is expected to be a String) matches a wildcard pattern that follows the JPA rules.


Object value

Checks that the left argument is equal to the given value.


Object value

Alias for eq.


Object value

Checks that the left argument is greater than the given value.


Object value

Checks that the left argument is greater than or equal to the given value.


Object value

Checks that the left argument is less than the given value.


Object value

Checks that the left argument is less than or equal to the given value.


Object from, Object to

Checks that the left argument is between the given range limits.

It’s important to note that query construction requires a multi-step chaining of method invocation that must be done in the proper sequence, must be properly completed exactly once and must not be done twice, or it will result in an error. The following examples are invalid, and depending on each case they lead to criteria being ignored (in benign cases) or an exception being thrown (in more serious ones).

// Incomplete construction. This query does not have any filter on "title" attribute yet,
// although the author may have intended to add one.
QueryBuilder qb1 = ...
Query q1 = qb1.build(); // consequently, this query matches all Book instances regardless of title!

// Duplicated completion. This results in an exception at run-time.
// Maybe the author intended to connect two conditions with a boolean operator,
// but this does NOT actually happen here.
QueryBuilder qb2 = ...
qb2.having("title").like("%Data Grid%");
qb2.having("description").like("%clustering%");   // will throw java.lang.IllegalStateException: Sentence already started. Cannot use 'having(..)' again.
Query q2 = qb2.build();
Filtering based on attributes of embedded entities

The having method also accepts dot separated attribute paths for referring to embedded entity attributes, so the following is a valid query:

// match all books that have an author named "Manik"
Query query = queryFactory.from(Book.class)

Each part of the attribute path must refer to an existing indexed attribute in the corresponding entity or embedded entity class respectively. It’s possible to have multiple levels of embedding.

Boolean conditions

Combining multiple attribute conditions with logical conjunction (and) and disjunction (or) operators in order to create more complex conditions is demonstrated in the following example. The well known operator precedence rule for boolean operators applies here, so the order of DSL method invocations during construction is irrelevant. Here and operator still has higher priority than or even though or was invoked first.

// match all books that have "Data Grid" in their title
// or have an author named "Manik" and their description contains "clustering"
Query query = queryFactory.from(Book.class)
  .having("title").like("%Data Grid%")

Boolean negation is achieved with the not operator, which has highest precedence among logical operators and applies only to the next simple attribute condition.

// match all books that do not have "Data Grid" in their title and are authored by "Manik"
Query query = queryFactory.from(Book.class)
  .not().having("title").like("%Data Grid%")
Nested conditions

Changing the precedence of logical operators is achieved with nested filter conditions. Logical operators can be used to connect two simple attribute conditions as presented before, but can also connect a simple attribute condition with the subsequent complex condition created with the same query factory.

// match all books that have an author named "Manik" and their title contains
// "Data Grid" or their description contains "clustering"
Query query = queryFactory.from(Book.class)
  .and(queryFactory.having("title").like("%Data Grid%")

In some use cases returning the whole domain object is overkill if only a small subset of the attributes are actually used by the application, especially if the domain entity has embedded entities. The query language allows you to specify a subset of attributes (or attribute paths) to return - the projection. If projections are used then the Query.list() will not return the whole domain entity but will return a List of Object[], each slot in the array corresponding to a projected attribute.

// match all books that have "Data Grid" in their title or description
// and return only their title and publication year
Query query = queryFactory.from(Book.class)
  .select("title", "publicationYear")
  .having("title").like("%Data Grid%")
  .or().having("description").like("%Data Grid%"))

Ordering the results based on one or more attributes or attribute paths is done with the QueryBuilder.orderBy( ) method which accepts an attribute path and a sorting direction. If multiple sorting criteria are specified, then the order of invocation of orderBy method will dictate their precedence. But you have to think of the multiple sorting criteria as acting together on the tuple of specified attributes rather than in a sequence of individual sorting operations on each attribute.

// match all books that have "Data Grid" in their title or description
// and return them sorted by the publication year and title
Query query = queryFactory.from(Book.class)
  .orderBy("publicationYear", SortOrder.DESC)
  .orderBy("title", SortOrder.ASC)
  .having("title").like("%Data Grid%")
  .or().having("description").like("%Data Grid%"))

You can limit the number of returned results by setting the maxResults property of QueryBuilder. This can be used in conjunction with setting the startOffset in order to achieve pagination of the result set.

// match all books that have "clustering" in their title
// sorted by publication year and title
// and return 3'rd page of 10 results
Query query = queryFactory.from(Book.class)
  .orderBy("publicationYear", SortOrder.DESC)
  .orderBy("title", SortOrder.ASC)
Even if the results being fetched are limited to maxResults you can still find the total number of matching results by calling Query.getResultSize().
Grouping and Aggregation

Infinispan has the ability to group query results according to a set of grouping fields and construct aggregations of the results from each group by applying an aggregation function to the set of values that fall into each group. Grouping and aggregation can only be applied to projection queries. The supported aggregations are: avg, sum, count, max, min. The set of grouping fields is specified with the groupBy(field) method, which can be invoked multiple times. The order used for defining grouping fields is not relevant. All fields selected in the projection must either be grouping fields or else they must be aggregated using one of the grouping functions described below. A projection field can be aggregated and used for grouping at the same time. A query that selects only grouping fields but no aggregation fields is legal. ⁠

Example: Grouping Books by author and counting them.

Query query = queryFactory.from(Book.class)
    .select(Expression.property("author"), Expression.count("title"))
A projection query in which all selected fields have an aggregation function applied and no fields are used for grouping is allowed. In this case the aggregations will be computed globally as if there was a single global group.

The following aggregation functions may be applied to a field: avg, sum, count, max, min

  • avg() - Computes the average of a set of numbers. Accepted values are primitive numbers and instances of java.lang.Number. The result is represented as java.lang.Double. If there are no non-null values the result is null instead.

  • count() - Counts the number of non-null rows and returns a java.lang.Long. If there are no non-null values the result is 0 instead.

  • max() - Returns the greatest value found. Accepted values must be instances of java.lang.Comparable. If there are no non-null values the result is null instead.

  • min() - Returns the smallest value found. Accepted values must be instances of java.lang.Comparable. If there are no non-null values the result is null instead.

  • sum() - Computes the sum of a set of Numbers. If there are no non-null values the result is null instead. The following table indicates the return type based on the specified field.

Table 6. Table sum return type
Field Type Return Type

Integral (other than BigInteger)


Float or Double






Evaluation of queries with grouping and aggregation

Aggregation queries can include filtering conditions, like usual queries. Filtering can be performed in two stages: before and after the grouping operation. All filter conditions defined before invoking the groupBy method will be applied before the grouping operation is performed, directly to the cache entries (not to the final projection). These filter conditions may reference any fields of the queried entity type, and are meant to restrict the data set that is going to be the input for the grouping stage. All filter conditions defined after invoking the groupBy method will be applied to the projection that results from the projection and grouping operation. These filter conditions can either reference any of the groupBy fields or aggregated fields. Referencing aggregated fields that are not specified in the select clause is allowed; however, referencing non-aggregated and non-grouping fields is forbidden. Filtering in this phase will reduce the amount of groups based on their properties. Sorting may also be specified similar to usual queries. The ordering operation is performed after the grouping operation and can reference any of the groupBy fields or aggregated fields.

Using Named Query Parameters

Instead of building a new Query object for every execution it is possible to include named parameters in the query which can be substituted with actual values before execution. This allows a query to be defined once and be efficiently executed many times. Parameters can only be used on the right-hand side of an operator and are defined when the query is created by supplying an object produced by the org.infinispan.query.dsl.Expression.param(String paramName) method to the operator instead of the usual constant value. Once the parameters have been defined they can be set by invoking either Query.setParameter(parameterName, value) or Query.setParameters(parameterMap) as shown in the examples below. ⁠

import org.infinispan.query.Search;
import org.infinispan.query.dsl.*;

QueryFactory queryFactory = Search.getQueryFactory(cache);
// Defining a query to search for various authors and publication years
Query query = queryFactory.from(Book.class)

// Set actual parameter values
query.setParameter("authorName", "Doe");
query.setParameter("publicationYear", 2010);

// Execute the query
List<Book> found = query.list();

Alternatively, multiple parameters may be set at once by supplying a map of actual parameter values: ⁠

Setting multiple named parameters at once
import java.util.Map;
import java.util.HashMap;


Map<String, Object> parameterMap = new HashMap<>();
parameterMap.put("authorName", "Doe");
parameterMap.put("publicationYear", 2010);

A significant portion of the query parsing, validation and execution planning effort is performed during the first execution of a query with parameters. This effort is not repeated during subsequent executions leading to better performance compared to a similar query using constant values instead of query parameters.
More Query DSL samples

Probably the best way to explore using the Query DSL API is to have a look at our tests suite. QueryDslConditionsTest is a fine example.


Create relational and full-text queries in both Library and Remote Client-Server mode with the Ickle query language.

Ickle is string-based and has the following characteristics:

  • Query Java classes and supports Protocol Buffers.

  • Queries can target a single entity type.

  • Queries can filter on properties of embedded objects, including collections.

  • Supports projections, aggregations, sorting, named parameters.

  • Supports indexed and non-indexed execution.

  • Supports complex boolean expressions.

  • Supports full-text queries.

  • Does not support computations in expressions, such as user.age > sqrt(user.shoeSize+3).

  • Does not support joins.

  • Does not support subqueries.

  • Is supported across various {Infinispan} APIs. Whenever a Query is produced by the QueryBuilder is accepted, including continuous queries or in event filters for listeners.

To use the API, first obtain a QueryFactory to the cache and then call the .create() method, passing in the string to use in the query. For instance:

QueryFactory qf = Search.getQueryFactory(remoteCache);
Query q = qf.create("from sample_bank_account.Transaction where amount > 20");

When using Ickle all fields used with full-text operators must be both Indexed and Analysed.

Ickle Query Language Parser Syntax

The parser syntax for the Ickle query language has some notable rules:

  • Whitespace is not significant.

  • Wildcards are not supported in field names.

  • A field name or path must always be specified, as there is no default field.

  • && and || are accepted instead of AND or OR in both full-text and JPA predicates.

  • ! may be used instead of NOT.

  • A missing boolean operator is interpreted as OR.

  • String terms must be enclosed with either single or double quotes.

  • Fuzziness and boosting are not accepted in arbitrary order; fuzziness always comes first.

  • != is accepted instead of <>.

  • Boosting cannot be applied to >,>=,<, operators. Ranges may be used to achieve the same result.

Fuzzy Queries

To execute a fuzzy query add ~ along with an integer, representing the distance from the term used, after the term. For instance

Query fuzzyQuery = qf.create("from sample_bank_account.Transaction where description : 'cofee'~2");
Range Queries

To execute a range query define the given boundaries within a pair of braces, as seen in the following example:

Query rangeQuery = qf.create("from sample_bank_account.Transaction where amount : [20 to 50]");

Phrase Queries

A group of words may be searched by surrounding them in quotation marks, as seen in the following example:

Query q = qf.create("from sample_bank_account.Transaction where description : 'bus fare'");

Proximity Queries

To execute a proximity query, finding two terms within a specific distance, add a ~ along with the distance after the phrase. For instance, the following example will find the words canceling and fee provided they are not more than 3 words apart:

Query proximityQuery = qf.create("from sample_bank_account.Transaction where description : 'canceling fee'~3 ");
Wildcard Queries

Both single-character and multi-character wildcard searches may be performed:

  • A single-character wildcard search may be used with the ? character.

  • A multi-character wildcard search may be used with the * character.

To search for text or test the following single-character wildcard search would be used:

Query wildcardQuery = qf.create("from sample_bank_account.Transaction where description : 'te?t'");

To search for test, tests, or tester the following multi-character wildcard search would be useD:

Query wildcardQuery = qf.create("from sample_bank_account.Transaction where description : 'test*'");
Regular Expression Queries

Regular expression queries may be performed by specifing a pattern between /. Ickle uses Lucene’s regular expression syntax, so to search for the words moat or boat the following could be used:

Query regExpQuery = qf.create("from sample_library.Book  where title : /[mb]oat/");
Boosting Queries

Terms may be boosted by adding a ^ after the term to increase their relevance in a given query, the higher the boost factor the more relevant the term will be. For instance to search for titles containing beer and wine with a higher relevance on beer, by a factor of 3, the following could be used:

Query boostedQuery = qf.create("from sample_library.Book where title : beer^3 OR wine");
Continuous Query

Continuous Queries allow an application to register a listener which will receive the entries that currently match a query filter, and will be continuously notified of any changes to the queried data set that result from further cache operations. This includes incoming matches, for values that have joined the set, updated matches, for matching values that were modified and continue to match, and outgoing matches, for values that have left the set. By using a Continuous Query the application receives a steady stream of events instead of having to repeatedly execute the same query to discover changes, resulting in a more efficient use of resources. For instance, all of the following use cases could utilize Continuous Queries:

  • Return all persons with an age between 18 and 25 (assuming the Person entity has an age property and is updated by the user application).

  • Return all transactions higher than $2000.

  • Return all times where the lap speed of F1 racers were less than 1:45.00s (assuming the cache contains Lap entries and that laps are entered live during the race).

Continuous Query Execution

A continuous query uses a listener that is notified when:

  • An entry starts matching the specified query, represented by a Join event.

  • A matching entry is updated and continues to match the query, represented by an Update event.

  • An entry stops matching the query, represented by a Leave event.

When a client registers a continuous query listener it immediately begins to receive the results currently matching the query, received as Join events as described above. In addition, it will receive subsequent notifications when other entries begin matching the query, as Join events, or stop matching the query, as Leave events, as a consequence of any cache operations that would normally generate creation, modification, removal, or expiration events. Updated cache entries will generate Update events if the entry matches the query filter before and after the operation. To summarize, the logic used to determine if the listener receives a Join, Update or Leave event is:

  1. If the query on both the old and new values evaluate false, then the event is suppressed.

  2. If the query on the old value evaluates false and on the new value evaluates true, then a Join event is sent.

  3. If the query on both the old and new values evaluate true, then an Update event is sent.

  4. If the query on the old value evaluates true and on the new value evaluates false, then a Leave event is sent.

  5. If the query on the old value evaluates true and the entry is removed or expired, then a Leave event is sent.

Continuous Queries can use the full power of the Query DSL except: grouping, aggregation, and sorting operations.
Running Continuous Queries

To create a continuous query you’ll start by creating a Query object first. This is described in the Query DSL section. Then you’ll need to obtain the ContinuousQuery (org.infinispan.query.api.continuous.ContinuousQuery) object of your cache and register the query and a continuous query listener (org.infinispan.query.api.continuous.ContinuousQueryListener) with it. A ContinuousQuery object associated to a cache can be obtained by calling the static method org.infinispan.client.hotrod.Search.getContinuousQuery(RemoteCache<K, V> cache) if running in remote mode or org.infinispan.query.Search.getContinuousQuery(Cache<K, V> cache) when running in embedded mode. Once the listener has been created it may be registered by using the addContinuousQueryListener method of ContinuousQuery:

continuousQuery.addContinuousQueryListener(query, listener);

The following example demonstrates a simple continuous query use case in embedded mode: ⁠

Registering a Continuous Query
import org.infinispan.query.api.continuous.ContinuousQuery;
import org.infinispan.query.api.continuous.ContinuousQueryListener;
import org.infinispan.query.Search;
import org.infinispan.query.dsl.QueryFactory;
import org.infinispan.query.dsl.Query;

import java.util.Map;
import java.util.concurrent.ConcurrentHashMap;


// We have a cache of Persons
Cache<Integer, Person> cache = ...

// We begin by creating a ContinuousQuery instance on the cache
ContinuousQuery<Integer, Person> continuousQuery = Search.getContinuousQuery(cache);

// Define our query. In this case we will be looking for any Person instances under 21 years of age.
QueryFactory queryFactory = Search.getQueryFactory(cache);
Query query = queryFactory.from(Person.class)

final Map<Integer, Person> matches = new ConcurrentHashMap<Integer, Person>();

// Define the ContinuousQueryListener
ContinuousQueryListener<Integer, Person> listener = new ContinuousQueryListener<Integer, Person>() {
    public void resultJoining(Integer key, Person value) {
        matches.put(key, value);

    public void resultUpdated(Integer key, Person value) {
        // we do not process this event

    public void resultLeaving(Integer key) {

// Add the listener and the query
continuousQuery.addContinuousQueryListener(query, listener);


// Remove the listener to stop receiving notifications

As Person instances having an age less than 21 are added to the cache they will be received by the listener and will be placed into the matches map, and when these entries are removed from the cache or their age is modified to be greater or equal than 21 they will be removed from matches.

Removing Continuous Queries

To stop the query from further execution just remove the listener:

Notes on performance of Continuous Queries

Continuous queries are designed to provide a constant stream of updates to the application, potentially resulting in a very large number of events being generated for particularly broad queries. A new temporary memory allocation is made for each event. This behavior may result in memory pressure, potentially leading to OutOfMemoryErrors (especially in remote mode) if queries are not carefully designed. To prevent such issues it is strongly recommended to ensure that each query captures the minimal information needed both in terms of number of matched entries and size of each match (projections can be used to capture the interesting properties), and that each ContinuousQueryListener is designed to quickly process all received events without blocking and to avoid performing actions that will lead to the generation of new matching events from the cache it listens to.

11.3. Remote Querying

Apart from supporting indexing and searching of Java entities to embedded clients, Infinispan introduced support for remote, language neutral, querying.

This leap required two major changes:

  • Since non-JVM clients cannot benefit from directly using Apache Lucene's Java API, Infinispan defines its own new query language, based on an internal DSL that is easily implementable in all languages for which we currently have an implementation of the Hot Rod client.

  • In order to enable indexing, the entities put in the cache by clients can no longer be opaque binary blobs understood solely by the client. Their structure has to be known to both server and client, so a common way of encoding structured data had to be adopted. Furthermore, allowing multi-language clients to access the data requires a language and platform-neutral encoding. Google’s Protocol Buffers was elected as an encoding format for both over-the-wire and storage due to its efficiency, robustness, good multi-language support and support for schema evolution.

11.3.1. Storing Protobuf encoded entities

Remote clients that want to be able to index and query their stored entities must do so using the Protobuf encoding format. This is key for the search capability to work. But it’s also possible to store Protobuf entities just for gaining the benefit of platform independence and not enable indexing if you do not need it.

Protobuf is all about structured data, so first thing you do to use it is define the structure of your data. This is accomplished by declaring protocol buffer message types in .proto files, like in the following example. Protobuf is a broad subject, we will not detail it here, so please consult the Protobuf Developer Guide for an in-depth explanation. It suffices to say for now that our example defines an entity (message type in protobuf speak) named Book, placed in a package named book_sample. Our entity declares several fields of primitive types and a repeatable field (an array basically) named authors. The Author message instances are embedded in the Book message instance.

package book_sample;

message Book {
  required string title = 1;
  required string description = 2;
  required int32 publicationYear = 3; // no native Date type available in Protobuf

  repeated Author authors = 4;

message Author {
  required string name = 1;
  required string surname = 2;

There are a few important notes we need to make about Protobuf messages:

  • nesting of messages is possible, but the resulting structure is strictly a tree, never a graph

  • there is no concept of type inheritance

  • collections are not supported but arrays can be easily emulated using repeated fields

Using Protobuf with the Java Hot Rod client is a two step process. First, the client must be configured to use a dedicated marshaller, ProtoStreamMarshaller. This marshaller uses the ProtoStream library to assist you in encoding your objects. The second step is instructing ProtoStream library on how to marshall your message types. The following example highlights this process.

import org.infinispan.client.hotrod.configuration.ConfigurationBuilder;
import org.infinispan.client.hotrod.marshall.ProtoStreamMarshaller;
import org.infinispan.protostream.SerializationContext;

ConfigurationBuilder clientBuilder = new ConfigurationBuilder();
    .marshaller(new ProtoStreamMarshaller());

RemoteCacheManager remoteCacheManager = new RemoteCacheManager(clientBuilder.build());

SerializationContext serCtx = ProtoStreamMarshaller.getSerializationContext(remoteCacheManager);

FileDescriptorSource fds = new FileDescriptorSource();
serCtx.registerMarshaller(new BookMarshaller());
serCtx.registerMarshaller(new AuthorMarshaller());

// Book and Author classes omitted for brevity

The interesting part in this sample is obtaining the SerializationContext associated to the RemoteCacheManager and then instructing ProtoStream about the protobuf types we want to marshall. The SerializationContext is provided by the library for this purpose. The SerializationContext.registerProtoFiles method receives the name of one or more classpath resources that is expected to be a protobuf definition containing our type declarations.

A RemoteCacheManager has no SerializationContext associated with it unless it was configured to use a ProtoStreamMarshaller.

The next relevant part is the registration of per entity marshallers for our domain model types. They must be provided by the user for each type or marshalling will fail. Writing marshallers is a simple process. The BookMarshaller example should get you started. The most important thing you need to consider is they need to be stateless and threadsafe as a single instance of them is being used.

import org.infinispan.protostream.MessageMarshaller;

public class BookMarshaller implements MessageMarshaller<Book> {

   public String getTypeName() {
      return "book_sample.Book";

   public Class<? extends Book> getJavaClass() {
      return Book.class;

   public void writeTo(ProtoStreamWriter writer, Book book) throws IOException {
      writer.writeString("title", book.getTitle());
      writer.writeString("description", book.getDescription());
      writer.writeInt("publicationYear", book.getPublicationYear());
      writer.writeCollection("authors", book.getAuthors(), Author.class);

   public Book readFrom(ProtoStreamReader reader) throws IOException {
      String title = reader.readString("title");
      String description = reader.readString("description");
      int publicationYear = reader.readInt("publicationYear");
      Set<Author> authors = reader.readCollection("authors", new HashSet<>(), Author.class);
      return new Book(title, description, publicationYear, authors);

Once you’ve followed these steps to setup your client you can start reading and writing Java objects to the remote cache and the actual data stored in the cache will be protobuf encoded provided that marshallers were registered with the remote client for all involved types (Book and Author in our example). Keeping your objects stored in protobuf format has the benefit of being able to consume them with compatible clients written in different languages.

11.3.2. Indexing of Protobuf encoded entries

After configuring the client as described in the previous section you can start configuring indexing for your caches on the server side. Activating indexing and the various indexing specific configurations is identical to embedded mode and is detailed in the Querying Infinispan chapter.

There is however an extra configuration step involved. While in embedded mode the indexing metadata is obtained via Java reflection by analyzing the presence of various Hibernate Search annotations on the entry’s class, this is obviously not possible if the entry is protobuf encoded. The server needs to obtain the relevant metadata from the same descriptor (.proto file) as the client. The descriptors are stored in a dedicated cache on the server named '___protobuf_metadata'. Both keys and values in this cache are plain strings. Registering a new schema is therefore as simple as performing a put operation on this cache using the schema’s name as key and the schema file itself as the value. Alternatively you can use the CLI (via the cache-container=*:register-proto-schemas() operation), the Management Console or the ProtobufMetadataManager MBean via JMX. Be aware that, when security is enabled, access to the schema cache via the remote protocols requires that the user belongs to the '___schema_manager' role.

Once indexing is enabled for a cache all fields of Protobuf encoded entries will be fully indexed unless you use the @Indexed and @Field protobuf schema pseudo-annotations in order to control precisely what fields need to get indexed. The default behaviour can be very inefficient when dealing with types having many or very larger fields so we encourage you to always specify what fields should be indexed instead of relying on the default indexing behaviour. The indexing behaviour for protobuf message types that are not annotated can also be modified per each schema file by setting the protobuf schema option 'indexed_by_default' to false (its default value is considered true) at the beginning of your schema file.
option indexed_by_default = false;  // This disables indexing of types that are not annotated for indexing

11.3.3. A remote query example

You’ve managed to configure both client and server to talk protobuf and you’ve enabled indexing. Let’s put some data in the cache and try to search for it then!

import org.infinispan.client.hotrod.*;
import org.infinispan.query.dsl.*;

RemoteCacheManager remoteCacheManager = ...;
RemoteCache<Integer, Book> remoteCache = remoteCacheManager.getCache();

Book book1 = new Book();
book1.setTitle("Hibernate in Action");
remoteCache.put(1, book1);

Book book2 = new Book();
book2.setTile("Hibernate Search in Action");
remoteCache.put(2, book2);

QueryFactory qf = Search.getQueryFactory(remoteCache);
Query query = qf.from(Book.class)
            .having("title").like("%Hibernate Search%")

List<Book> list = query.list(); // Voila! We have our book back from the cache!

The key part of creating a query is obtaining the QueryFactory for the remote cache using the org.infinispan.client.hotrod.Search.getQueryFactory() method. Once you have this creating the query is similar to embedded mode which is covered in this section.

11.3.4. Analysis

Analysis is a process that converts input data into one or more terms that you can index and query.

Default Analyzers

Infinispan provides a set of default analyzers as follows:

Definition Description


Splits text fields into tokens, treating whitespace and punctuation as delimiters.


Tokenizes input streams by delimiting at non-letters and then converting all letters to lowercase characters. Whitespace and non-letters are discarded.


Splits text streams on whitespace and returns sequences of non-whitespace characters as tokens.


Treats entire text fields as single tokens.


Stems English words using the Snowball Porter filter.


Generates n-gram tokens that are 3 grams in size by default.


Splits text fields into larger size tokens than the standard analyzer, treating whitespace as a delimiter and converts all letters to lowercase characters.

These analyzer definitions are based on Apache Lucene and are provided "as-is". For more information about tokenizers, filters, and CharFilters, see the appropriate Lucene documentation.

Using Analyzer Definitions

To use analyzer definitions, reference them by name in the .proto schema file.

  1. Include the Analyze.YES attribute to indicate that the property is analyzed.

  2. Specify the analyzer definition with the @Analyzer annotation.

The following example shows referenced analyzer definitions:

/* @Indexed */
message TestEntity {

    /* @Field(store = Store.YES, analyze = Analyze.YES, analyzer = @Analyzer(definition = "keyword")) */
    optional string id = 1;

    /* @Field(store = Store.YES, analyze = Analyze.YES, analyzer = @Analyzer(definition = "simple")) */
    optional string name = 2;
Creating Custom Analyzer Definitions

If you require custom analyzer definitions, do the following:

  1. Create an implementation of the ProgrammaticSearchMappingProvider interface packaged in a JAR file.

  2. Provide a file named org.infinispan.query.spi.ProgrammaticSearchMappingProvider in the META-INF/services/ directory of your JAR. This file should contain the fully qualified class name of your implementation.

  3. Copy the JAR to the standalone/deployments directory of your Infinispan installation.

    Your deployment must be available to the Infinispan server during startup. You cannot add the deployment if the server is already running.

    The following is an example implementation of the ProgrammaticSearchMappingProvider interface:

    import org.apache.lucene.analysis.core.LowerCaseFilterFactory;
    import org.apache.lucene.analysis.core.StopFilterFactory;
    import org.apache.lucene.analysis.standard.StandardFilterFactory;
    import org.apache.lucene.analysis.standard.StandardTokenizerFactory;
    import org.hibernate.search.cfg.SearchMapping;
    import org.infinispan.Cache;
    import org.infinispan.query.spi.ProgrammaticSearchMappingProvider;
    public final class MyAnalyzerProvider implements ProgrammaticSearchMappingProvider {
       public void defineMappings(Cache cache, SearchMapping searchMapping) {
                .analyzerDef("standard-with-stop", StandardTokenizerFactory.class)
  4. Specify the JAR in the cache container configuration, for example:

    <cache-container name="mycache" default-cache="default">
         <module name="deployment.analyzers.jar"/>

11.4. Statistics

Query Statistics can be obtained from the SearchManager, as demonstrated in the following code snippet.

SearchManager searchManager = Search.getSearchManager(cache);
org.hibernate.search.stat.Statistics statistics = searchManager.getStatistics();
This data is also available via JMX through the Hibernate Search StatisticsInfoMBean registered under the name org.infinispan:type=Query,manager="{name-of-cache-manager}",cache="{name-of-cache}",component=Statistics. Please note this MBean is always registered by Infinispan but the statistics are collected only if statistics collection is enabled at cache level.
Hibernate Search has its own configuration properties hibernate.search.jmx_enabled and hibernate.search.generate_statistics for JMX statistics as explained here. Using them with Infinispan Query is forbidden as it will only lead to duplicated MBeans and unpredictable results.

11.5. Performance Tuning

11.5.1. Batch writing in SYNC mode

By default, the Index Managers work in sync mode, meaning when data is written to Infinispan, it will perform the indexing operations synchronously. This synchronicity guarantees indexes are always consistent with the data (and thus visible in searches), but can slowdown write operations since it will also perform a commit to the index. Committing is an extremely expensive operation in Lucene, and for that reason, multiple writes from different nodes can be automatically batched into a single commit to reduce the impact.

So, when doing data loads to Infinispan with index enabled, try to use multiple threads to take advantage of this batching.

If using multiple threads does not result in the required performance, an alternative is to load data with indexing temporarily disabled and run a re-indexing operation afterwards. This can be done writing data with the SKIP_INDEXING flag:


11.5.2. Writing using async mode

If it’s acceptable a small delay between data writes and when that data is visible in queries, an index manager can be configured to work in async mode. The async mode offers much better writing performance, since in this mode commits happen at a configurable interval.


<distributed-cache name="default">
    <indexing index="LOCAL">
        <property name="default.indexmanager">
        <!-- Index data in async mode -->
        <property name="default.worker.execution">async</property>
        <!-- Optional: configure the commit interval, default is 1000ms -->
        <property name="default.index_flush_interval">500</property>

11.5.3. Index reader async strategy

Lucene internally works with snapshots of the index: once an IndexReader is opened, it will only see the index changes up to the point it was opened; further index changes will not be visible until the IndexReader is refreshed. The Index Managers used in Infinispan by default will check the freshness of the index readers before every query and refresh them if necessary.

It is possible to tune this strategy to relax this freshness checking to a pre-configured interval by using the reader.strategy configuration set as async:

<distributed-cache name="default"
    <indexing index="PRIMARY_OWNER">
        <property name="default.indexmanager">
        <property name="default.reader.strategy">async</property>
        <!-- refresh reader every 1s, default is 5s -->
        <property name="default.reader.async_refresh_period_ms">1000</property>

The async reader strategy is particularly useful for Index Managers that rely on shards, such as the AffinityIndexManager.

11.5.4. Lucene Options

It is possible to apply tuning options in Lucene directly. For more details, see the Hibernate Search manual.

12. Clustered Counters

Clustered counters are counters which are distributed and shared among all nodes in the Infinispan cluster. Counters can have different consistency levels: strong and weak.

Although a strong/weak consistent counter has separate interfaces, both support updating its value, return the current value and they provide events when its value is updated. Details are provided below in this document to help you choose which one fits best your uses-case.

12.1. Installation and Configuration

In order to start using the counters, you needs to add the dependency in your Maven pom.xml file:


Replace ${version.infinispan} with the appropriate version of Infinispan.

The counters can be configured Infinispan configuration file or on-demand via the CounterManager interface detailed later in this document. A counters configured in Infinispan configuration file is created at boot time when the EmbeddedCacheManager is starting. Theses counters are started eagerly and they are available in all the cluster’s nodes.

<?xml version="1.0" encoding="UTF-8"?>
    <cache-container ...>
        <!-- if needed to persist counter, global state needs to be configured -->
        <!-- your caches configuration goes here -->
         <counters xmlns="urn:infinispan:config:counters:9.2" num-owners="3" reliability="CONSISTENT">
             <strong-counter name="c1" initial-value="1" storage="PERSISTENT"/>
             <strong-counter name="c2" initial-value="2" storage="VOLATILE">
                 <lower-bound value="0"/>
             <strong-counter name="c3" initial-value="3" storage="PERSISTENT">
                 <upper-bound value="5"/>
             <strong-counter name="c4" initial-value="4" storage="VOLATILE">
                 <lower-bound value="0"/>
                 <upper-bound value="10"/>
             <weak-counter name="c5" initial-value="5" storage="PERSISTENT" concurrency-level="1"/>

or programmatically, in the GlobalConfigurationBuilder:

GlobalConfigurationBuilder globalConfigurationBuilder = ...;
CounterManagerConfigurationBuilder builder = globalConfigurationBuilder.addModule(CounterManagerConfigurationBuilder.class);

On other hand, the counters can be configured on-demand, at any time after the EmbeddedCacheManager is initialized.

CounterManager manager = ...;
manager.defineCounter("c1", CounterConfiguration.builder(CounterType.UNBOUNDED_STRONG).initialValue(1).storage(Storage.PERSISTENT)build());
manager.defineCounter("c2", CounterConfiguration.builder(CounterType.BOUNDED_STRONG).initialValue(2).lowerBound(0).storage(Storage.VOLATILE).build());
manager.defineCounter("c3", CounterConfiguration.builder(CounterType.BOUNDED_STRONG).initialValue(3).upperBound(5).storage(Storage.PERSISTENT).build());
manager.defineCounter("c4", CounterConfiguration.builder(CounterType.BOUNDED_STRONG).initialValue(4).lowerBound(0).upperBound(10).storage(Storage.VOLATILE).build());
manager.defineCounter("c2", CounterConfiguration.builder(CounterType.WEAK).initialValue(5).concurrencyLevel(1).storage(Storage.PERSISTENT).build());
CounterConfiguration is immutable and can be reused.

The method defineCounter() will return true if the counter is successful configured or false otherwise. However, if the configuration is invalid, the method will throw a CounterConfigurationException. To find out if a counter is already defined, use the method isDefined().

CounterManager manager = ...
if (!manager.isDefined("someCounter")) {
    manager.define("someCounter", ...);

Per cluster attributes:

  • num-owners: Sets the number of counter’s copies to keep cluster-wide. A smaller number will make update operations faster but will support a lower number of server crashes. It must be positive and its default value is 2.

  • reliability: Sets the counter’s update behavior in a network partition. Default value is AVAILABLE and valid values are:

    • AVAILABLE: all partitions are able to read and update the counter’s value.

    • CONSISTENT: only the primary partition (majority of nodes) will be able to read and update the counter’s value. The remaining partitions can only read its value.

Per counter attributes:

  • initial-value [common]: Sets the counter’s initial value. Default is 0 (zero).

  • storage [common]: Sets the counter’s behavior when the cluster is shutdown and restarted. Default value is VOLATILE and valid values are:

    • VOLATILE: the counter’s value is only available in memory. The value will be lost when a cluster is shutdown.

    • PERSISTENT: the counter’s value is stored in a private and local persistent store. The value is kept when the cluster is shutdown and restored after a restart.

On-demand and VOLATILE counters will lose its value and configuration after a cluster shutdown. They must be defined again after the restart.
  • lower-bound [strong]: Sets the strong consistent counter’s lower bound. Default value is Long.MIN_VALUE.

  • upper-bound [strong]: Sets the strong consistent counter’s upper bound. Default value is Long.MAX_VALUE.

If neither the lower-bound or upper-bound are configured, the strong counter is set as unbounded.
The initial-value must be between lower-bound and upper-bound inclusive.
  • concurrency-level [weak]: Sets the number of concurrent updates. Its value must be positive and the default value is 16.

12.1.1. List counter names

To list all the counters defined, the method CounterManager.getCounterNames() returns a collection of all counter names created cluster-wide.

12.2. The CounterManager interface.

The CounterManager interface is the entry point to define, retrieve and remove a counter. It automatically listen to the creation of EmbeddedCacheManager and proceeds with the registration of an instance of it per EmbeddedCacheManager. It starts the caches needed to store the counter state and configures the default counters.

Retrieving the CounterManager is as simple as invoke the EmbeddedCounterManagerFactory.asCounterManager(EmbeddedCacheManager) as shown in the example below:

// create or obtain your EmbeddedCacheManager
EmbeddedCacheManager manager = ...;

// retrieve the CounterManager
CounterManager counterManager = EmbeddedCounterManagerFactory.asCounterManager(manager);

For Hot Rod client, the CounterManager is registered in the RemoteCacheManager and it can be retrieved like:

// create or obtain your RemoteCacheManager
RemoteCacheManager manager = ...;

// retrieve the CounterManager
CounterManager counterManager = RemoteCounterManagerFactory.asCounterManager(manager);
Hot Rod messages format can be found in Hot Rod Protocol 2.7

12.2.1. Remove a counter via CounterManager

use with caution.

There is a difference between remove a counter via the Strong/WeakCounter interfaces and the CounterManager. The CounterManager.remove(String) removes the counter value from the cluster and removes all the listeners registered in the counter in the local counter instance. In addition, the counter instance is no longer reusable and it may return an invalid results.

On the other side, the Strong/WeakCounter removal only removes the counter value. The instance can still be reused and the listeners still works.

The counter is re-created if it is accessed after a removal.

12.3. The Counter

A counter can be strong (StrongCounter) or weakly consistent (WeakCounter) and both is identified by a name. They have a specific interface but they share some logic, namely, both of them are asynchronous ( a CompletableFuture is returned by each operation), provide an update event and can be reset to its initial value.

If you don’t want to use the async API, it is possible to return a synchronous counter via sync() method. The API is the same but without the CompletableFuture return value.

The following methods are common to both interfaces:

String getName();
CompletableFuture<Long> getValue();
CompletableFuture<Void> reset();
<T extends CounterListener> Handle<T> addListener(T listener);
CounterConfiguration getConfiguration();
CompletableFuture<Void> remove();
SyncStrongCounter sync(); //SyncWeakCounter for WeakCounter
  • getName() returns the counter name (identifier).

  • getValue() returns the current counter’s value.

  • reset() allows to reset the counter’s value to its initial value.

  • addListener() register a listener to receive update events. More details about it in the Notification and Events section.

  • getConfiguration() returns the configuration used by the counter.

  • remove() removes the counter value from the cluster. The instance can still be used and the listeners are kept.

  • sync() creates a synchronous counter.

The counter is re-created if it is accessed after a removal.

12.3.1. The StrongCounter interface: when the consistency or bounds matters.

The strong counter provides uses a single key stored in Infinispan cache to provide the consistency needed. All the updates are performed under the key lock to updates its values. On other hand, the reads don’t acquire any locks and reads the current value. Also, with this scheme, it allows to bound the counter value and provide atomic operations like compare-and-set/swap.

A StrongCounter can be retrieved from the CounterManager by using the getStrongCounter() method. As an example:

CounterManager counterManager = ...
StrongCounter aCounter = counterManager.getStrongCounter("my-counter);
Since every operation will hit a single key, the StrongCounter has a higher contention rate.

The StrongCounter interface adds the following method:

default CompletableFuture<Long> incrementAndGet() {
   return addAndGet(1L);

default CompletableFuture<Long> decrementAndGet() {
   return addAndGet(-1L);

CompletableFuture<Long> addAndGet(long delta);

CompletableFuture<Boolean> compareAndSet(long expect, long update);

CompletableFuture<Long> compareAndSwap(long expect, long update);
  • incrementAndGet() increments the counter by one and returns the new value.

  • decrementAndGet() decrements the counter by one and returns the new value.

  • addAndGet() adds a delta to the counter’s value and returns the new value.

  • compareAndSet() and compareAndSwap() atomically set the counter’s value if the current value is the expected.

A operation is considered completed when the CompletableFuture is completed.
The difference between compare-and-set and compare-and-swap is that the former returns true if the operation succeeds while the later returns the previous value. The compare-and-swap is successful if the return value is the same as the expected.
Bounded StrongCounter

When bounded, all the update method above will throw a CounterOutOfBoundsException when they reached the lower or upper bound. The exception has the following methods to check which side bound has been reached:

public boolean isUpperBoundReached();
public boolean isLowerBoundReached();
Uses cases

The strong counter fits better in the following uses cases:

  • When counter’s value is needed after each update (example, cluster-wise ids generator or sequences)

  • When a bounded counter is needed (example, rate limiter)

Usage Examples
StrongCounter counter = counterManager.getStrongCounter("unbounded_coutner");

// incrementing the counter
System.out.println("new value is " + counter.incrementAndGet().get());

// decrement the counter's value by 100 using the functional API
counter.addAndGet(-100).thenApply(v -> {
   System.out.println("new value is " + v);
   return null;

// alternative, you can do some work while the counter is updated
CompletableFuture<Long> f = counter.addAndGet(10);
// ... do some work ...
System.out.println("new value is " + f.get());

// and then, check the current value
System.out.println("current value is " + counter.getValue().get());

// finally, reset to initial value
System.out.println("current value is " + counter.getValue().get());

// or set to a new value if zero
System.out.println("compare and set succeeded? " + counter.compareAndSet(0, 1));

And below, there is another example using a bounded counter:

StrongCounter counter = counterManager.getStrongCounter("bounded_counter");

// incrementing the counter
try {
    System.out.println("new value is " + counter.addAndGet(100).get());
} catch (ExecutionException e) {
    Throwable cause = e.getCause();
    if (cause instanceof CounterOutOfBoundsException) {
       if (((CounterOutOfBoundsException) cause).isUpperBoundReached()) {
          System.out.println("ops, upper bound reached.");
       } else if (((CounterOutOfBoundsException) cause).isLowerBoundReached()) {
          System.out.println("ops, lower bound reached.");

// now using the functional API
counter.addAndGet(-100).handle((v, throwable) -> {
   if (throwable != null) {
      Throwable cause = throwable.getCause();
      if (cause instanceof CounterOutOfBoundsException) {
         if (((CounterOutOfBoundsException) cause).isUpperBoundReached()) {
            System.out.println("ops, upper bound reached.");
         } else if (((CounterOutOfBoundsException) cause).isLowerBoundReached()) {
            System.out.println("ops, lower bound reached.");
      return null;
   System.out.println("new value is " + v);
   return null;

Compare-and-set vs Compare-and-swap examples:

StrongCounter counter = counterManager.getStrongCounter("my-counter");
long oldValue, newValue;
do {
   oldValue = counter.getValue().get();
   newValue = someLogic(oldValue);
} while (!counter.compareAndSet(oldValue, newValue).get());

With compare-and-swap, it saves one invocation counter invocation (counter.getValue())

StrongCounter counter = counterManager.getStrongCounter("my-counter");
long oldValue = counter.getValue().get();
long currentValue, newValue;
do {
   currentValue = oldValue;
   newValue = someLogic(oldValue);
} while ((oldValue = counter.compareAndSwap(oldValue, newValue).get()) != currentValue);

12.3.2. The WeakCounter interface: when speed is needed

The WeakCounter stores the counter’s value in multiple keys in Infinispan cache. The number of keys created is configured by the concurrency-level attribute. Each key stores a partial state of the counter’s value and it can be updated concurrently. It main advantage over the StrongCounter is the lower contention in the cache. On other hand, the read of its value is more expensive and bounds are not allowed.

The reset operation should be handled with caution. It is not atomic and it produces intermediates values. These value may be seen by a read operation and by any listener registered.

A WeakCounter can be retrieved from the CounterManager by using the getWeakCounter() method. As an example:

CounterManager counterManager = ...
StrongCounter aCounter = counterManager.getWeakCounter("my-counter);
Weak Counter Interface

The WeakCounter adds the following methods:

default CompletableFuture<Void> increment() {
   return add(1L);

default CompletableFuture<Void> decrement() {
   return add(-1L);

CompletableFuture<Void> add(long delta);

They are similar to the `StrongCounter’s methods but they don’t return the new value.

Uses cases

The weak counter fits best in uses cases where the result of the update operation is not needed or the counter’s value is not required too often. Collecting statistics is a good example of such an use case.


Below, there is an example of the weak counter usage.

WeakCounter counter = counterManager.getWeakCounter("my_counter");

// increment the counter and check its result
System.out.println("current value is " + counter.getValue().get());

CompletableFuture<Void> f = counter.add(-100);
//do some work
f.get(); //wait until finished
System.out.println("current value is " + counter.getValue().get());

//using the functional API
counter.reset().whenComplete((aVoid, throwable) -> System.out.println("Reset done " + (throwable == null ? "successfully" : "unsuccessfully"))).get();
System.out.println("current value is " + counter.getValue().get());

12.4. Notifications and Events

Both strong and weak counter supports a listener to receive its updates events. The listener must implement CounterListener and it can be registerer by the following method:

<T extends CounterListener> Handle<T> addListener(T listener);

The CounterLister has the following interface:

public interface CounterListener {
   void onUpdate(CounterEvent entry);

The Handle object returned has the main goal to remove the CounterListener when it is not longer needed. Also, it allows to have access to the CounterListener instance that is it handling. It has the following interface:

public interface Handle<T extends CounterListener> {
   T getCounterListener();
   void remove();

Finally, the CounterEvent has the previous and current value and state. It has the following interface:

public interface CounterEvent {
   long getOldValue();
   State getOldState();
   long getNewValue();
   State getNewState();
The state is always State.VALID for unbounded strong counter and weak counter. State.LOWER_BOUND_REACHED and State.UPPER_BOUND_REACHED are only valid for bounded strong counters.
The weak counter reset() operation will trigger multiple notification with intermediate values.

13. Clustered Lock

A clustered lock is a lock which is distributed and shared among all nodes in the Infinispan cluster and currently provides a way to execute code that will be synchronized between the nodes in a given cluster.

13.1. Installation

In order to start using the clustered locks, you needs to add the dependency in your Maven pom.xml file:


Replace ${version.infinispan} with the appropriate version of Infinispan.

13.2. ClusteredLock Configuration

Currently there is a single type of ClusteredLock supported : non reentrant, NODE ownership lock.

13.2.1. Ownership

  • NODE When a ClusteredLock is defined, this lock can be used from all the nodes in the Infinispan cluster. When the ownership is NODE type, this means that the owner of the lock is the Infinispan node that acquired the lock at a given time. This means that each time we get a ClusteredLock instance with the ClusteredCacheManager, this instance will be the same instance for each Infinispan node. This lock can be used to synchronize code between Infinispan nodes. The advantage of this lock is that any thread in the node can release the lock at a given time.

  • INSTANCE - not yet supported

When a ClusteredLock is defined, this lock can be used from all the nodes in the Infinispan cluster. When the ownership is INSTANCE type, this means that the owner of the lock is the actual instance we acquired when ClusteredLockManager.get("lockName") is called.

This means that each time we get a ClusteredLock instance with the ClusteredCacheManager, this instance will be a new instance. This lock can be used to synchronize code between Infinispan nodes and inside each Infinispan node. The advantage of this lock is that only the instance that called 'lock' can release the lock.

13.2.2. Reentrancy

When a ClusteredLock is configured reentrant, the owner of the lock can reacquire the lock as many consecutive times as it wants while holding the lock.

Currently, only non reentrant locks are supported. This means that when two consecutive lock calls are sent for the same owner, the first call will acquire the lock if it’s available, and the second call will block.

13.3. ClusteredLockManager Interface

The ClusteredLockManager interface, marked as experimental, is the entry point to define, retrieve and remove a lock. It automatically listen to the creation of EmbeddedCacheManager and proceeds with the registration of an instance of it per EmbeddedCacheManager. It starts the internal caches needed to store the lock state.

Retrieving the ClusteredLockManager is as simple as invoking the EmbeddedClusteredLockManagerFactory.from(EmbeddedCacheManager) as shown in the example below:

// create or obtain your EmbeddedCacheManager
EmbeddedCacheManager manager = ...;

// retrieve the ClusteredLockManager
ClusteredLockManager clusteredLockManager = EmbeddedClusteredLockManagerFactory.from(manager);
public interface ClusteredLockManager {

   boolean defineLock(String name);

   boolean defineLock(String name, ClusteredLockConfiguration configuration);

   ClusteredLock get(String name);

   ClusteredLockConfiguration getConfiguration(String name);

   boolean isDefined(String name);

   CompletableFuture<Boolean> remove(String name);

   CompletableFuture<Boolean> forceRelease(String name);
  • defineLock : Defines a lock with the specified name and the default ClusteredLockConfiguration. It does not overwrite existing configurations.

  • defineLock(String name, ClusteredLockConfiguration configuration) : Defines a lock with the specified name and ClusteredLockConfiguration. It does not overwrite existing configurations.

  • ClusteredLock get(String name) : Get’s a ClusteredLock by it’s name. A call of defineLock must be done at least once in the cluster. See ownership level section to understand the implications of get method call.

Currently, the only ownership level supported is NODE.

  • ClusteredLockConfiguration getConfiguration(String name) :

Returns the configuration of a ClusteredLock, if such exists.

  • boolean isDefined(String name) : Checks if a lock is already defined.

  • CompletableFuture<Boolean> remove(String name) : Removes a ClusteredLock if such exists.

  • CompletableFuture<Boolean> forceRelease(String name) : Releases - or unlocks - a ClusteredLock, if such exists, no matter who is holding it at a given time. Calling this method may cause concurrency issues and has to be used in exceptional situations.

13.4. ClusteredLock Interface

ClusteredLock interface, marked as experimental, is the interface that implements the clustered locks.

public interface ClusteredLock {

   CompletableFuture<Void> lock();

   CompletableFuture<Boolean> tryLock();

   CompletableFuture<Boolean> tryLock(long time, TimeUnit unit);

   CompletableFuture<Void> unlock();

   CompletableFuture<Boolean> isLocked();

   CompletableFuture<Boolean> isLockedByMe();
  • lock : Acquires the lock. If the lock is not available then call blocks until the lock is acquired. Currently, there is no maximum time specified for a lock request to fail, so this could cause thread starvation.

  • tryLock Acquires the lock only if it is free at the time of invocation, and returns true in that case. This method does not block (or wait) for any lock acquisition.

  • tryLock(long time, TimeUnit unit) If the lock is available this method returns immediately with true. If the lock is not available then the call waits until :

    • The lock is acquired

    • The specified waiting time elapses

If the time is less than or equal to zero, the method will not wait at all.

  • unlock

Releases the lock. Only the holder of the lock may release the lock.

  • isLocked Returns true when the lock is locked and false when the lock is released.

  • isLockedByMe Returns true when the lock is owned by the caller and false when the lock is owned by someone else or it’s released.

13.4.1. Usage Examples

  EmbeddedCache cm = ...;
  ClusteredLockManager cclm = EmbeddedClusteredLockManagerFactory.from(cm);

    .thenCompose(result -> {
       if (result) {
        try {
            // manipulate protected state
            } finally {
               return lock.unlock();
       } else {
          // Do something else

13.5. ClusteredLockManager Configuration

You can configure ClusteredLockManager to use different strategies for locks, either declaratively or programmatically, with the following attributes:


Defines the total number of nodes in each cluster that store the states of clustered locks. The default value is -1, which replicates the value to all nodes.


Controls how clustered locks behave when clusters split into partitions or multiple nodes leave a cluster. You can set the following values:

  • AVAILABLE: Nodes in any partition can concurrently operate on locks.

  • CONSISTENT: Only nodes that belong to the majority partition can operate on locks. This is the default value.

The following is an example declarative configuration for ClusteredLockManager:

<?xml version="1.0" encoding="UTF-8"?>
        xsi:schemaLocation="urn:infinispan:config:${infinispan.core.schema.version} http://www.infinispan.org/schemas/infinispan-config-9.4.xsd"
    <cache-container default-cache="default">
        <local-cache name="default">
            <locking concurrency-level="100" acquire-timeout="1000"/>

        <clustered-locks xmlns="urn:infinispan:config:clustered-locks:9.4"
                         num-owners = "3"
            <clustered-lock name="lock1" />
            <clustered-lock name="lock2" />

14. Multimap Cache

MutimapCache is a type of Infinispan Cache that maps keys to values in which each key can contain multiple values.

14.1. Installation and configuration


Replace ${version.infinispan} with the appropriate version of Infinispan.

14.2. MultimapCache API

MultimapCache API exposes several methods to interact with the Multimap Cache. All these methods are non-blocking in most of the cases. See [limitations]

public interface MultimapCache<K, V> {

   CompletableFuture<Void> put(K key, V value);

   CompletableFuture<Collection<V>> get(K key);

   CompletableFuture<Boolean> remove(K key);

   CompletableFuture<Boolean> remove(K key, V value);

   CompletableFuture<Void> remove(Predicate<? super V> p);

   CompletableFuture<Boolean> containsKey(K key);

   CompletableFuture<Boolean> containsValue(V value);

   CompletableFuture<Boolean> containsEntry(K key, V value);

   CompletableFuture<Long> size();

   boolean supportsDuplicates();


14.2.1. CompletableFuture<Void> put(K key, V value)

Puts a key-value pair in the multimap cache.

MultimapCache<String, String> multimapCache = ...;

multimapCache.put("girlNames", "marie")
             .thenCompose(r1 -> multimapCache.put("girlNames", "oihana"))
             .thenCompose(r3 -> multimapCache.get("girlNames"))
             .thenAccept(names -> {
                              System.out.println("Marie is a girl name");

                              System.out.println("Oihana is a girl name");

The output of this code is :

Marie is a girl name
Oihana is a girl name

14.2.2. CompletableFuture<Collection<V>> get(K key)

Asynchronous that returns a view collection of the values associated with key in this multimap cache, if any. Any changes to the retrieved collection won’t change the values in this multimap cache. When this method returns an empty collection, it means the key was not found.

14.2.3. CompletableFuture<Boolean> remove(K key)

Asynchronous that removes the entry associated with the key from the multimap cache, if such exists.

14.2.4. CompletableFuture<Boolean> remove(K key, V value)

Asynchronous that removes a key-value pair from the multimap cache, if such exists.

14.2.5. CompletableFuture<Void> remove(Predicate<? super V> p)

Asynchronous method. Removes every value that match the given predicate.

14.2.6. CompletableFuture<Boolean> containsKey(K key)

Asynchronous that returns true if this multimap contains the key.

14.2.7. CompletableFuture<Boolean> containsValue(V value)

Asynchronous that returns true if this multimap contains the value in at least one key.

14.2.8. CompletableFuture<Boolean> containsEntry(K key, V value)

Asynchronous that returns true if this multimap contains at least one key-value pair with the value.

14.2.9. CompletableFuture<Long> size()

Asynchronous that returns the number of key-value pairs in the multimap cache. It doesn’t return the distinct number of keys.

14.2.10. boolean supportsDuplicates()

Asynchronous that returns true if the multimap cache supports duplicates. This means that the content of the multimap can be 'a' → ['1', '1', '2']. For now this method will always return false, as duplicates are not yet supported. The existence of a given value is determined by 'equals' and `hashcode' method’s contract.

14.3. Creating a Multimap Cache

Currently the MultimapCache is configured as a regular cache. This can be done either by code or XML configuration. See how to configure a regular Cache in the section link to [configure a cache].

14.3.1. Embedded mode

// create or obtain your EmbeddedCacheManager
EmbeddedCacheManager cm = ... ;

// create or obtain a MultimapCacheManager passing the EmbeddedCacheManager
MultimapCacheManager multimapCacheManager = EmbeddedMultimapCacheManagerFactory.from(cm);

// define the configuration for the multimap cache
multimapCacheManager.defineConfiguration(multimapCacheName, c.build());

// get the multimap cache
multimapCache = multimapCacheManager.get(multimapCacheName);

14.4. Limitations

In almost every case the Multimap Cache will behave as a regular Cache, but some limitations exist in the current version.

14.4.1. Support for duplicates

Duplicates are not supported yet. This means that the multimap won’t contain any duplicate key-value pair. Whenever put method is called, if the key-value pair already exist, this key-value par won’t be added. Methods used to check if a key-value pair is already present in the Multimap are the equals and hashcode.

14.4.2. Eviction

For now, the eviction works per key, and not per key-value pair. This means that whenever a key is evicted, all the values associated with the key will be evicted too. Eviction per key-value could be supported in the future.

14.4.3. Transactions

Implicit transactions are supported through the auto-commit and all the methods are non blocking. Explicit transactions work without blocking in most of the cases. Methods that will block are size, containsEntry and remove(Predicate<? super V> p)

15. CDI Support

Infinispan includes integration with Contexts and Dependency Injection (better known as CDI) via Infinispan’s infinispan-cdi-embedded or infinispan-cdi-remote module. CDI is part of Java EE specification and aims for managing beans' lifecycle inside the container. The integration allows to inject Cache interface and bridge Cache and CacheManager events. JCache annotations (JSR-107) are supported by infinispan-jcache and infinispan-jcache-remote artifacts. For more information have a look at Chapter 11 of the JCACHE specification.

15.1. Maven Dependencies

To include CDI support for Infinispan in your project, use one of the following dependencies:

pom.xml for Embedded mode

Replace ${version.infinispan} with the appropriate version of Infinispan.

pom.xml for Remote mode

Replace ${version.infinispan} with the appropriate version of Infinispan.

Which version of Infinispan should I use?
We recommend using the latest final version Infinispan.

15.2. Embedded cache integration

15.2.1. Inject an embedded cache

By default you can inject the default Infinispan cache. Let’s look at the following example:

Default cache injection
import javax.inject.Inject;

public class GreetingService {

    private Cache<String, String> cache;

    public String greet(String user) {
        String cachedValue = cache.get(user);
        if (cachedValue == null) {
            cachedValue = "Hello " + user;
            cache.put(user, cachedValue);
        return cachedValue;

If you want to use a specific cache rather than the default one, you just have to provide your own cache configuration and cache qualifier. See example below:

Qualifier example
import javax.inject.Qualifier;

@Target({ElementType.FIELD, ElementType.PARAMETER, ElementType.METHOD})
public @interface GreetingCache {
Injecting Cache with qualifier
import org.infinispan.configuration.cache.Configuration;
import org.infinispan.configuration.cache.ConfigurationBuilder;
import org.infinispan.cdi.ConfigureCache;
import javax.enterprise.inject.Produces;

public class Config {

    @ConfigureCache("greeting-cache") // This is the cache name.
    @GreetingCache // This is the cache qualifier.
    public Configuration greetingCacheConfiguration() {
        return new ConfigurationBuilder()

    // The same example without providing a custom configuration.
    // In this case the default cache configuration will be used.
    public Configuration greetingCacheConfiguration;

To use this cache in the GreetingService add the @GeetingCache qualifier on your cache injection point.

15.2.2. Override the default embedded cache manager and configuration

You can override the default cache configuration used by the default EmbeddedCacheManager. For that, you just have to create a Configuration producer with default qualifiers as illustrated in the following snippet:

Overriding Configuration
public class Config {

    // By default CDI adds the @Default qualifier if no other qualifier is provided.
    public Configuration defaultEmbeddedCacheConfiguration() {
        return new ConfigurationBuilder()

It’s also possible to override the default EmbeddedCacheManager. The newly created manager must have default qualifiers and Application scope.

Overriding EmbeddedCacheManager
import javax.enterprise.context.ApplicationScoped;

public class Config {

    public EmbeddedCacheManager defaultEmbeddedCacheManager() {
      return new DefaultCacheManager(new ConfigurationBuilder()

15.2.3. Configure the transport for clustered use

To use Infinispan in a clustered mode you have to configure the transport with the GlobalConfiguration. To achieve that override the default cache manager as explained in the previous section. Look at the following snippet:

Overriding default EmbeddedCacheManager
package org.infinispan.configuration.global.GlobalConfigurationBuilder;

public EmbeddedCacheManager defaultClusteredCacheManager() {
    return new DefaultCacheManager(
        new GlobalConfigurationBuilder().transport().defaultTransport().build(),
        new ConfigurationBuilder().memory().size(7).build()

15.3. Remote cache integration

15.3.1. Inject a remote cache

With the CDI integration it’s also possible to use a RemoteCache as illustrated in the following snippet:

Injecting RemoteCache
public class GreetingService {

    private RemoteCache<String, String> cache;

    public String greet(String user) {
        String cachedValue = cache.get(user);
        if (cachedValue == null) {
            cachedValue = "Hello " + user;
            cache.put(user, cachedValue);
        return cachedValue;

If you want to use another cache, for example the greeting-cache, add the @Remote qualifier on the cache injection point which contains the cache name.

Injecting RemoteCache with qualifier
public class GreetingService {

    private RemoteCache<String, String> cache;


Adding the @Remote cache qualifier on each injection point might be error prone. That’s why the remote cache integration provides another way to achieve the same goal. For that you have to create your own qualifier annotated with @Remote:

RemoteCache qualifier
@Target({ElementType.FIELD, ElementType.PARAMETER, ElementType.METHOD})
public @interface RemoteGreetingCache {

To use this cache in the GreetingService add the qualifier @RemoteGreetingCache qualifier on your cache injection.

15.3.2. Override the default remote cache manager

Like the embedded cache integration, the remote cache integration comes with a default remote cache manager producer. This default RemoteCacheManager can be overridden as illustrated in the following snippet:

Overriding default RemoteCacheManager
public class Config {

    public RemoteCacheManager defaultRemoteCacheManager() {
        return new RemoteCacheManager(localhost, 1544);

15.4. Use a custom remote/embedded cache manager for one or more cache

It’s possible to use a custom cache manager for one or more cache. You just need to annotate the cache manager producer with the cache qualifiers. Look at the following example:

public class Config {

   public EmbeddedCacheManager specificEmbeddedCacheManager() {
      return new DefaultCacheManager(new ConfigurationBuilder()

   public RemoteCacheManager specificRemoteCacheManager() {
       return new RemoteCacheManager("localhost", 1544);

With the above code the GreetingCache or the RemoteGreetingCache will be associated with the produced cache manager.

Producer method scope
To work properly the producers must have the scope @ApplicationScoped . Otherwise each injection of cache will be associated to a new instance of cache manager.

15.5. Use JCache caching annotations

There is now a separate module for JSR 107 (JCACHE) integration, including API. See this chapter for details.

When CDI integration and JCache artifacts are present on the classpath, it is possible to use JCache annotations with CDI managed beans. These annotations provide a simple way to handle common use cases. The following caching annotations are defined in this specification:

  • @CacheResult - caches the result of a method call

  • @CachePut - caches a method parameter

  • @CacheRemoveEntry - removes an entry from a cache

  • @CacheRemoveAll - removes all entries from a cache

Annotations target type
These annotations must only be used on methods.

To use these annotations, proper interceptors need to be declared in beans.xml file:

Interceptors for managed environments such as Application Servers
<?xml version="1.0" encoding="UTF-8"?>
<beans xmlns="http://xmlns.jcp.org/xml/ns/javaee"
   xsi:schemaLocation="http://xmlns.jcp.org/xml/ns/javaee http://xmlns.jcp.org/xml/ns/javaee/beans_1_1.xsd"
   version="1.2" bean-discovery-mode="annotated">

Interceptors for unmanaged environments such as standalone applications
<?xml version="1.0" encoding="UTF-8"?>
<beans xmlns="http://xmlns.jcp.org/xml/ns/javaee"
   xsi:schemaLocation="http://xmlns.jcp.org/xml/ns/javaee http://xmlns.jcp.org/xml/ns/javaee/beans_1_1.xsd"
   version="1.2" bean-discovery-mode="annotated">


The following snippet of code illustrates the use of @CacheResult annotation. As you can see it simplifies the caching of the Greetingservice#greet method results.

Using JCache annotations
import javax.cache.interceptor.CacheResult;

public class GreetingService {

    public String greet(String user) {
        return "Hello" + user;

The first version of the GreetingService and the above version have exactly the same behavior. The only difference is the cache used. By default it’s the fully qualified name of the annotated method with its parameter types (e.g. org.infinispan.example.GreetingService.greet(java.lang.String)).

Using other cache than default is rather simple. All you need to do is to specify its name with the cacheName attribute of the cache annotation. For example:

Specifying cache name for JCache
@CacheResult(cacheName = "greeting-cache")

15.6. Use Cache events and CDI

It is possible to receive Cache and Cache Manager level events using CDI Events. You can achieve it using @Observes annotation as shown in the following snippet:

Event listeners based on CDI
import javax.enterprise.event.Observes;
import org.infinispan.notifications.cachemanagerlistener.event.CacheStartedEvent;
import org.infinispan.notifications.cachelistener.event.*;

public class GreetingService {

    // Cache level events
    private void entryRemovedFromCache(@Observes CacheEntryCreatedEvent event) {

    // Cache Manager level events
    private void cacheStarted(@Observes CacheStartedEvent event) {
Check Listeners and Notifications for more information about event types.

16. JCache (JSR-107) provider

Infinispan provides an implementation of JCache 1.0 API ( JSR-107 ). JCache specifies a standard Java API for caching temporary Java objects in memory. Caching java objects can help get around bottlenecks arising from using data that is expensive to retrieve (i.e. DB or web service), or data that is hard to calculate. Caching these type of objects in memory can help speed up application performance by retrieving the data directly from memory instead of doing an expensive roundtrip or recalculation. This document specifies how to use JCache with Infinispan’s implementation of the specification, and explains key aspects of the API.

16.1. Dependencies

In order to start using Infinispan JCache implementation, a single dependency needs to be added to the Maven pom.xml file:


Replace ${version.infinispan} with the appropriate version of Infinispan.

16.2. Create a local cache

Creating a local cache, using default configuration options as defined by the JCache API specification, is as simple as doing the following:

import javax.cache.*;
import javax.cache.configuration.*;

// Retrieve the system wide cache manager
CacheManager cacheManager = Caching.getCachingProvider().getCacheManager();
// Define a named cache with default JCache configuration
Cache<String, String> cache = cacheManager.createCache("namedCache",
      new MutableConfiguration<String, String>());
By default, the JCache API specifies that data should be stored as storeByValue, so that object state mutations outside of operations to the cache, won’t have an impact in the objects stored in the cache. Infinispan has so far implemented this using serialization/marshalling to make copies to store in the cache, and that way adhere to the spec. Hence, if using default JCache configuration with Infinispan, data stored must be marshallable.

Alternatively, JCache can be configured to store data by reference (just like Infinispan or JDK Collections work). To do that, simply call:

Cache<String, String> cache = cacheManager.createCache(