The internal marshalling capabilities of Infinispan have undergone a significant refactoring in 10.0. The marshalling of internal Infinispan objects and user objects are now truly isolated. This means that it’s now possible to configure Marshaller implementations in embedded mode or on the server, without having to handle the marshalling of Infinispan internal classes. Consequently, it’s possible to easily change the marshaller implementation, in a similar manner to how users of the HotRod client are accustomed.
As a consequence of the above changes, the default marshaller used for marshalling user types is no longer based upon JBoss Marshalling. Instead we now utilise the ProtoStream library to store user types in the language agnostic Protocol Buffers format. It is still possible to utilise the old default, JBossUserMarshaller, however it’s necessary to add the
org.infinispan:infinispan-jboss-marshalling artifact to your application’s classpath.
The following interfaces/annotations have been deprecated as a consequence of the marshalling refactoring:
For cluster communication any configured Externalizer's are still utilised to marshall objects, however they are ignored when persisting data to cache stores unless the JBossUserMarshaller is explicitly configured via the global SerializationConfiguration.
It’s highly recommended to migrate from the old Externalizer and JBoss marshalling approach to the new ProtoStream based marshalling, as the interfaces listed above and the JBossUserMarshaller implementation will be removed in future versions.
Unfortunately, the extensive marshalling changes mean that the binary format used by Infinispan stores in
9.4.x is no longer compatible with
10.0.x. Therefore, it’s necessary for any existing stores to be migrated to the new format via the StoreMigrator tool.
|Whilst we regret that 9.4.x stores are no longer binary compatible, these extensive changes should ensure binary compatibility across future major versions.|
Stores now default to being segmented if the property is not configured. Some stores do not support being segmented, which will result in a configuration exception being thrown at startup. The moving forward position is to use segmented stores when possible to increase cache wide performance and reduce memory requirements for various operations including state transfer.
The file based stores (SingleFileStore and SoftIndexFileStore) both support being segmented, but their current implementation requires opening file descriptors based on how many segments there are. This may cause issues in some configurations and users should be aware. Infinispan will print a single WARN message when such a configuration is found.
Caches created through the CacheContainerAdmin API will now be
PERMANENT by default. Use the
VOLATILE flag to obtain the previous behaviour.
Older versions of the Hot Rod protocol treated expiration values greater than the number of milliseconds in 30 days as Unix time. Starting with Hot Rod 3.0 this adjustment no longer happens and expiration is taken literally.
Total Order transaction protocol is going to be removed in a future release. Use the default protocol (2PC).
infinispan.server.hotrod.workerThreads property was introduced as a hack to work around the fact that the configuration did not expose it. The property has been removed and endpoint worker threads must now be exclusively configured using the
AtomicMapLookup, AtomicMap and FineGrainedAtomicMap have been removed. Please see FunctionalMaps or Cache#Merge for similar functionality.
The previously deprecated Delta and DeltaAware interfaces have been removed.
The default cache must now be named explicitly via the GlobalConfigurationBuilder#defaultCacheName() method.
The previously deprecated DistributedExecutor is now removed. References should be updated to use ClusterExecutor.
TreeCache has been unsupported for a long time and was only intended as a quick stopgap for JBossCache users. The module has now been removed completely.
Previously the JDBC PooledConnectionFactory provided c3p0 and HikariCP based connection pools. From 10.0 we only provide a PooledConnectionFactory based upon the Agroal project. This means that it is no longer possible to utilise
hikari.properties files to configure the pool, instead an agroal compatiblet properties file can be provided.
Several configuration elements and attributes that were deprecated since 9.0 have been removed:
<eviction>- replaced with
<versioning>- automatically enabled
<data-container>- no longer customizable
deadlock-detection-spin- always disabled
write-skew- enabled automatically
The xsite state transfer chunk size (
<backup><state-transfer chunk-size="X"/></backup>) can no longer be
>= 0, same as the regular state transfer chunk size. Previously a value <= 0 would transfer the entire cache in a single batch, which is almost always a bad idea.
The default marshaller is no longer GenericJbossMarshaller. We now utilise the ProtoStream library as the default. If Java Serialization is required by clients, we strongly recommend utilising the link:JavaSerializationMarshaller instead. However if the GenericJbossMarshaller must be used, it’s necessary to add the
org.infinispan:infinispan-jboss-marshalling artifact to your client’s classpath and for the
GenericJbossMarshaller to be configured as the marshaller.
The getBulk method is an expensive method as it requires holding all keys in memory at once and requires a possibly very single result to populate it. The new retrieveEntries, entrySet, keySet and values methods handle this in a much more efficient way. Therefore the getBulk methods have been removed in favor of them.
File-based cache stores (SingleFileStore, SoftIndexFileStore, RocksDBStore) filesystem layout has been normalized so that they will use the
GlobalStateConfigurationpersistent location as a default location. Additionally, all stores will now use the cache name as part of the data file/directory naming allowing multiple stores to avoid conflicts and ambiguity.
The CLI loader (
infinispan-persistence-cli) has been removed.
The LevelDB store (
infinispan-cachestore-leveldb) has been removed. Use the RocksDB store instead, as it is fully backwards compatible.
singletonstore configuration option and the wrapper class
SingletonCacheWriterhave been removed.
shared=trueis enough, as only the primary owner of each key will write to a shared store.
The Hot Rod client and server only support protocol versions 2.0 and higher. Support for Hot Rod versions 1.0 to 1.3 has been dropped.
SKIP_LISTENER_NOTIFICATION notification flag has been added in the hotrod client. This flag only works when the client and the server version is 9.4.15 or higher. Spring Session integration uses this flag when a session id has changed. If you are using Spring Session with Infinispan 9.4, consider upgrading the client and the server.
performAsync header was removed from the REST server. Clients that want to perform async operations with the REST server should manage the request and response on their side to avoid blocking.
REST operations that don’t return resources and are used with
DELETE methods now return status
204 (No content) instead of
With the introduction of inline XML JGroups stacks in the configuration, two default stacks are always enabled:
tcp. If you are declaring your own stacks with the same names, an exception reporting the conflict will be thrown. Simply rename your own configurations to avoid the conflict.
Because of changes in AWS’s access policy regarding signatures, S3_PING will not work in newer regions and will stop working in older regions too. For this reason, you should migrate to using NATIVE_S3_PING instead.
Listeners in the past that were sync, always ran in the thread that caused the event. We now allow a Listener method to be non-blocking in that it will still fire in the original thread, under the assumption that it will return immediately. Please read the Listener Javadoc for information and examples on this.
Distributed Streams has parts rewritten to utilize non blocking reactive streams based operations. As such null values are not supported as values from operations as per the reactive streams spec. Please utilize other means to denote a null value.
The infinispan-cloud module has been removed and the
azure default configurations have been included in
infinispan-core and can be referenced as default named JGroups configurations.
Almost as soon as GUARANTEED_DELIVERY was added, UNICAST3 and NAKACK2.resend_last_seqno removed the need for it. It was always documented as experimental, so we removed it without deprecation and we also removed the RSVP protocol from the default JGroups stacks.
The possible statuses of the cache health are now HEALTHY, HEALTHY_REBALANCING and DEGRADED to better reflect the fact that
rebalancing doesn’t mean a cluster is unhealthy.
When using multi-tenancy in the Wildfly based server, it’s necessary to specify the
content-path for each of the REST connectors, to match the
prefix element under
Off Heap memory containers now will dynamically resize based on number of entries in the container. Due to this the address count configuration value is now deprecated for APIs and has been removed from the xml parser.
Compatibility mode has been deprecated and will be removed in the next Infinispan version.
To use a cache from multiple endpoints, it is recommended to store data in binary format and to configure the MediaType for keys and values.
If storing data as unmarshalled objects is still desired, the equivalent of compatibility mode is to configure keys and values to store object content:
<encoding> <key media-type="application/x-java-object"/> <value media-type="application/x-java-object"/> </encoding>
For better interoperability between endpoints, the Memcached server no longer stores keys as
java.lang.String, but as UTF-8
If using memcached, it’s recommended to run a rolling upgrade from 9.3 to store data in the new format, or reload the data in the cache.
Distributed scripts with text-based data type no longer return
null when the result from each server is null. The response is now a JSON array with each individual result, e.g.
The WebSocket endpoint has been unmaintained for several years. It has been removed.
Since the Hot Rod client was overhauled in 9.2, the way the connection pool configuration is handled has changed. Infinispan 9.4 introduces a new naming scheme for the connection pool properties which deprecates the old commons-pool names. For a complete reference of the available configuration options for the properties file please refer to remote client configuration javadoc.
A store cannot be configured as both shared and having passivation enabled. Doing so can cause data inconsistencies as there is no way to synchronize data between all the various nodes. As such this configuration will now cause a startup exception. Please update your configuration as appropriate.
The AdvancedCacheLoader SPI has been enhanced to provide an alternative method to process and instead allows reactive streams based publishKeys and publishEntries methods which provide benefits in performance, threading and ease of use. Note this change will only affect you if you wish take advantage of it in any custom CacheLoaders you may have implemented.
In 9.3 the default MergePolicy is now MergePolicy.NONE, opposed to MergePolicy.PREFERRED_ALWAYS.
We have reverted the stat changes introduced in 9.1, so average values for read, write and removals are once again returned as milliseconds.
Several new event log messages have been added, and one message has been removed (ISPN100013).
The max idle entry expiration information is sent between owners in the cluster. However when an entry expires via max idle on a given node, this was not replicated (only removing it locally). Max idle has been enhanced to now expire an entry across the entire cluster, instead of per node. This includes ensuring that max idle expiration is applied across all owners (meaning if another node has accessed the entry within the given time it will prevent that entry from expiring on other nodes that didn’t have an access).
Max idle in a transactional clustered cache does not remove expired entries on access (although it will not be returned). These entries are only removed via the expiration reaper.
Iteration in a clustered cache will still show entries that are expired via maxIdle to ensure good performance, but could be removed at any point due to expiration reaper.
The Infinispan Wildfly modules are now located in the
system/add-ons/ispn dir as per the Wildfly module conventions.
Deserialization of content sent by clients to the server are no longer allowed by default. This applies to JSON, XML, and marshalled byte that, depending on the cache configuration, will cause the server to convert it to Java Objects either to store it or to perform any operation that cannot be done on a byte directly.
The deserialization needs to be enabled using system properties, ether by class name or regular expressions:
// Comma separated list of fully qualified class names -Dinfinispan.deserialization.whitelist.classes=java.time.Instant,com.myclass.Entity // Regex expression -Dinfinispan.deserialization.whitelist.regexps=.*
The latest version of Kubernetes Ping uses unified environmental variables for both Kubernetes and OpenShift. Some of them were shortened for example
OPENSHIFT_KUBE_PING_NAMESPACE was changed to
KUBERNETES_NAMESPACE. Please refer to Kubernetes Ping documentation.
Average values for read, write and removals are now returned in Nanoseconds, opposed to Milliseconds.
Infinispan now contains a new implementation of both
FineGrainedAtomicMap, but the semantics has been preserved. The new implementation does not use
DeltaAware interface but the Functional API instead.
There are no changes needed for
AtomicMap, but it now supports non-transactional use case as well.
FineGrainedAtomicMap now uses the Grouping API and therefore you need to enable groups in configuration. Also it holds entries as regular cache entries, plus one cache entry for cached key set (the map itself). Therefore the cache size or iteration/streaming results may differ. Note that fine grained atomic maps are still supported on transactional caches only.
RemoteCache now implements all of the collection backed methods from
Map interface. Previously
keySet was implemented, however it was a deep copy. This has now changed and it is a backing set. That is that the set retrieves the updated values on each invocation or updates to the backing remote cache for writes. The entrySet and values methods are also now supported as backing variants as well.
If you wish to have a copy like was provided before it is recommended to copy the contents into a in memory local set such as
Set<K> keysCopy = remoteCache.keySet().stream().collect(Collectors.toSet());
CopyableDeltaAware have been deprecated. Method
AdvancedCache.applyDelta() has been deprecated and the implementation does not allow custom set of locked keys.
ApplyDeltaCommand and its uses in interceptor stack are deprecated.
Any partial updates to an entry should be replaced using the Functional API.
The configuration property
directory_provider now accepts a new value
local-heap. This value replaces the now deprecated
ram, and as its predecessor will cause the index to be stored in a
The configuration value
ram is still accepted and will have the same effect, but failing to replace
local-heap will cause a warning to be logged. We suggest to perform this replacement, as the
ram value will no longer be recognised by Infinispan in a future version.
This change was made as the team believes the
local-heap name better expresses the storage model, especially as this storage method will not allow real-time replication of the index across multiple nodes. This index storage option is mostly useful for single node integration testing of the query functionality.
TableManipulation::batchSize and JpaStoreConfiguration::batchSize have been deprecated and replaced by the higher level AbstractStoreConfiguration::maxBatchSize.
In Infinispan 9.1 partition handling has been improved to allow for automatic conflict resolution on partition merges. Consequently, PartitionHandlingConfiguration::enabled has been deprecated in favour of PartitionHandlingConfiguration::whenSplit. Configuring whenSplit to the DENY_READ_WRITES strategy is equivalent to setting enabled to true, whilst specifying ALLOW_READ_WRITES is equivalent to disabling partition handling (default).
Furthermore, during a partition merge with ALLOW_READ_WRITES, the default EntryMergePolicy is MergePolicies.PREFERRED_ALWAYS which provides a deterministic way of tie-breaking CacheEntry conflicts. If you require the old behaviour, simply set the merge-policy to null.
The default configuration for transactional caches changed from
OPTIMISTIC locking to
OPTIMISTIC locking with
Also, using the
REPEATABLE_READ isolation level and
OPTIMISTIC locking without
write-skew enabled is no longer allowed. To help with the upgrade,
write-skew will be automatically enabled in this case.
The following configuration has been deprecated:
write-skew: as said, it is automatically enabled.
<versioning>and its attributes. It is automatically enabled and configured when needed.
Both were deprecated since version 5.1.
eagerLocking(true) can be replaced with
eagerLockingSingleNode() does not need a replacement because it was a no-op.
Asynchronous mode is no longer supported in transactional caches and it will automatically use the synchronous cache mode. In addition, the second phase of a transaction commit is done synchronously. The following methods (and related) are deprecated:
The following classes have been deprecated and they will be removed in the future:
DummyBaseTransactionManager: replaced by
DummyXid: replaced by
DummyTransaction: replaced by
DummyTransactionManager: replaced by
RecoveryDummyTransactionManagerLookup: replaced by
DummyUserTransaction: replaced by
mode attribute in the XML declaration of clustered caches is no longer mandatory. It defaults to SYNC.
Up to Infinispan 8.x, the default cache always implicitly existed, even if not declared in the XML configuration. Additionally, the default cache configuration affected all other cache configurations, acting as some kind of base template. Since 9.0, the default cache only exists if it has been explicitly configured. Additionally, even if it has been specified, it will never act as base template for other caches.
Internally Infinispan 9.x has introduced many improvements to its marshalling codebase in order to improve performance and allow for greater flexibility. Consequently, data marshalled and persisted by Infinispan 8.x is no longer compatible with Infinispan 9.0. To aid you in migrating your existing stores to 9.0, we have provided a Store Migrator, however at present this only allows the migration of JDBC stores.
In Infinispan 8.x, cloud related configuration were added to
infinispan-core module. Since 9.0 they were moved to
The entry retriever feature has been removed. Please update to use the new Streams feature detailed in the User Guide. The
org.infinispan.filter.CacheFilters class can be used to convert
Converter instances into proper Stream operations that are able to be marshalled.
Map reduce has been removed in favor of the new Streams feature which should provide more features and performance. There are no bridge classes to convert to the new streams and all references must be rewritten.
SerializableSupplier has moved from the
org.infinispan.stream package to the
CloseableSupplier has moved from the
org.infinispan.util package to the
SerializableCallable have all been moved from the
org.infinispan.util package to the
SegmentCompletionListener has moved from the interface
org.infinispan.CacheStream to the new
All Infinispan, Spring and Logger dependencies are now in the
provided scope. One can decide whether to use small jars or uber jars but they need to be added to the classpath of the application. It also gives one freedom in choosing Spring (or Spring Boot) version.
Here is an example:
<dependencies> <dependency> <groupId>org.infinispan</groupId> <artifactId>infinispan-embedded</artifactId> </dependency> <dependency> <groupId>org.infinispan</groupId> <artifactId>infinispan-spring5-embedded</artifactId> </dependency> <dependency> <groupId>org.springframework</groupId> <artifactId>spring-context</artifactId> </dependency> <dependency> <groupId>org.springframework.session</groupId> <artifactId>spring-session</artifactId> </dependency> </dependencies>
Additionally there is no Logger implementation specified (since this may vary depending on use case).
The total order protocol now uses the
remote-command-executor. The attribute
<container> tag is removed.
HikariCP offers superior performance to c3p0 and is now the default implementation. Additional properties for HikariCP can be provided by placing a
hikari.properties file on the classpath or by specifying the path to the file via
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.
Support for c3p0 has been deprecated and will be removed in a future release. Users can force c3p0 to be utilised as before by providing the system property
The LevelDB cache store was replaced with a RocksDB. RocksDB is a fork of LevelDB which provides superior performance in high concurrency scenarios. The new cache store can parse old LevelDB configurations but will always use the RocksDB implementation.
The JDBC Mixed and Binary stores have been removed due to the poor performance associated with storing entries in buckets. Storing entries in buckets is non-optimal as each read/write to the store requires an existing bucket for a given hash to be retrieved, deserialised, updated, serialised and then re-inserted back into the db. If you were previously using one of the removed stores, we have provided a migrator tool to assist in migrating data from an existing binary table to a JDBC string based store.
A new annotation,
@Store, has been added for persistence stores. This allows a store’s properties to be explicitly defined and validated against the provided store configuration. Existing stores should be updated to use this annotation and the store’s configuration class should also declare the
@ConfigurationFor annotation. If neither of these annotations are present on the store or configuration class, then a your store will continue to function as before, albeit with a warning that additional store validation cannot be completed.
The no-anonymous policy is now automatically enabled for Hot Rod authentication unless explicitly specified.
Classes regarding EntrySizeCalculator have now been moved down to the org.infinispan.util package.
Store as Binary configuration is now deprecated and will be removed in a future release. This is replaced by the new memory configuration.
The keySet, entrySet and values methods on DataContainer have been deprecated. These behavior of these methods are very inconsistent and will be removed later. It is recommended to update references to use iterator or iteratorIncludingExpired methods intead.
CDI module (GroupId:ArtifactId
org.infinispan:infinispan-cdi) has been split into
org.infinispan:infinispan-cdi-remote. Please make sure that you use proper artifact.
Async marshalling has been entirely dropped since it was never reliable enough. The "async-marshalling" attribute has been removed from the 8.0 XML schema and will be ignored when parsing 7.x configuration files. The programmatic configuration methods related to asyncMarshalling/syncMarshalling are now deprecated and have no effect aside from producing a WARN message in the logs.
Because of the inability to configure write skew in the server, the isolation level attribute was ignored and defaulted to READ_COMMITTED. Now, when enabling REPEATABLE_READ together with optimistic locking, write skew is enabled by default in local and synchronous configurations.
In order to avoid conflict and confusion with the similar subsystems in WildFly, we have renamed the following subsystems in server: * infinispan → datagrid-infinispan * jgroups → datagrid-jgroups * endpoint → datagrid-infinispan-endpoint
To be more inline with JCache and java.util.collections interfaces we have changed the first argument type for the CacheLoader.load & CacheLoader.contains methods to be Object from type K.
To be more inline with JCache and java.util.collections interfaces we have changed the first argument type for the CacheWriter.delete method to be Object from type K.
Over time Infinispan added 2 interfaces with identical names and almost identical methods. The org.infinispan.notifications.KeyFilter and org.infinispan.persistence.spi.AdvancedCacheLoader$KeyFilter interfaces.
Both of these interfaces are used for the sole purpose of filtering an entry by it’s given key. Infinispan 7.0 has also introduced the KeyValueFilter which is similar to both but also can filter on the entries value and/or metadata.
As such all of these classes have been moved into a new package org.infinispan.filter and all of their related helper classes.
The new org.infinispan.filter.KeyFilter interface has replaced both of the previous interfaces and all previous references use the new interface.
In order to use all of the latest features, make sure you change the namespace declaration at the top of your XML configuration files as follows:
<infinispan xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="urn:infinispan:config:6.0 http://www.infinispan.org/schemas/infinispan-config-6.0.xsd" xmlns="urn:infinispan:config:6.0"> ... </infinispan>
ObjectOutput startObjectOutput(OutputStream os, boolean isReentrant)from class
CacheEntry getCacheEntry(Object key, EnumSet<Flag> explicitFlags, ClassLoader explicitClassLoader)from class
org.infinispan.AdvancedCache. Please use instead:
AdvanceCache.withFlags(Flag… flags).with(ClassLoader classLoader).getCacheEntry(K key).
AtomicMap<K, V> getAtomicMap(Cache<MK, ?> cache, MK key, FlagContainer flagContainer)from class
org.infinispan.atomic.AtomicMapLookup. Please use instead
AtomicMapLookup.getAtomicMap(cache.getAdvancedCache().withFlags(Flag… flags), MK key).
org.infinispan.config(and all methods involving the old configuration classes). All methods removed has an overloaded method which receives the new configuration classes as parameters. Please refer to [_configuration] for more information about the new configuration classes.
|This only affects the programmatic configuration.|
boolean isLocal(Object key)from class
org.infinispan.distribution.DistributionManager. Please use instead
void setStatisticsEnabled(boolean enabled)from class
org.infinispan.interceptors.TxInterceptorPlease use instead the
boolean delete(boolean synchronous)from class
org.infinispan.io.GridFile. Please use instead
long getLocallyInterruptedTransactions()from class
In order to use all of the latest features, make sure you change the namespace declaration at the top of your XML configuration files as follows:
<infinispan xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="urn:infinispan:config:5.2 http://www.infinispan.org/schemas/infinispan-config-5.2.xsd" xmlns="urn:infinispan:config:5.3"> ... </infinispan>
In order to use all of the latest features, make sure you change the namespace declaration at the top of your XML configuration files as follows:
<infinispan xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="urn:infinispan:config:5.2 http://www.infinispan.org/schemas/infinispan-config-5.2.xsd" xmlns="urn:infinispan:config:5.2"> ... </infinispan>
In practical terms, if you were using the default values, this should not cause any backward compatibility issues but an increase in performance of about 5-7%. However in order to use the old configuration defaults, you need to configure the following:
<transaction useSynchronization="false"> <recovery enabled="false"/> </transaction>
or the programmatic configuration equivalent:
ConfigurationBuilder builder = new ConfigurationBuilder(); builder.transaction().useSynchronization(false).recovery().enabled(false)
Cache Loader and Store configuration has changed greatly in Infinispan 5.2.
The cache and cache manager hierarchies have changed slightly in 5.1 with the introduction of
BasicCacheContainer , which are parent classes of existing
CacheContainer classes respectively. What’s important is that Hot Rod clients must now code against
BasicCacheContainer rather than
CacheContainer. So previous code that was written like this will no longer compile.
import org.infinispan.Cache; import org.infinispan.manager.CacheContainer; import org.infinispan.client.hotrod.RemoteCacheManager; ... CacheContainer cacheContainer = new RemoteCacheManager(); Cache cache = cacheContainer.getCache();
Instead, if Hot Rod clients want to continue using interfaces higher up the hierarchy from the remote cache/container classes, they’ll have to write:
import org.infinispan.BasicCache; import org.infinispan.manager.BasicCacheContainer; import org.infinispan.client.hotrod.RemoteCacheManager; ... BasicCacheContainer cacheContainer = new RemoteCacheManager(); BasicCache cache = cacheContainer.getCache();
However, previous code that interacted against the
RemoteCacheManager will work as it used to:
import org.infinispan.client.hotrod.RemoteCache; import org.infinispan.client.hotrod.RemoteCacheManager; ... RemoteCacheManager cacheContainer = new RemoteCacheManager(); RemoteCache cache = cacheContainer.getCache();
The eviction XML element no longer defines the
wakeUpIntervalattribute. This is now configured via the
<expiration wakeUpInterval="60000"... />
maxEntries is used as guide for the entire cache, but eviction happens on a per cache segment, so when the segment is full, the segment is evicted. That’s why
maxEntries is a theoretical limit but in practical terms, it’ll be a bit less than that. This is done for performance reasons.
A cache marked as
TRANSACTIONALcannot be accessed outside of a transaction, and a
NON_TRANSACTIONALcache cannot be accessed within a transaction. In 5.0, a transactional cache would support non-transactional calls as well. This change was done to be in-line with expectations set out in JSR-107 as well as to provide more consistent behavior.
In 5.0, commit and rollback phases were asynchronous by default. Starting with 5.1, these are now synchronous by default, to provide the guarantees required by a single lock-owner model.
One of the big changes we made in 5.1 was to use the same push-based state transfer we introduced in 5.0 both for rehashing in distributed mode and for state retrieval in replicated mode. We even borrow the consistent hash concept in replicated mode to transfer state from all previous cache members at once in order to speed up transfer.
As a consequence we’ve unified the state transfer configuration as well, there is now a
stateTransfer element containing a simplified state transfer configuration. The corresponding attributes in the
hash elements have been deprecated, as have been some attributes that are no longer used.
If you use XML to configure Infinispan, you shouldn’t notice any change, except a much faster startup, courtesy of the StAX based parser. However, if you use programmatic configuration, read on for the important differences.
Configuration is now packaged in
org.infinispan.configuration, and you must use a fluent, builder style:
Configuration c1 = new ConfigurationBuilder() // Adjust any configuration defaults you want .clustering() .l1() .disable() .mode(DIST_SYNC) .hash() .numOwners(5) .build();
The old javabean style configuration is now deprecated and will be removed in a later version.
Configuration properties which can be safely changed at runtime are mutable, and all others are immutable.
To copy a configuration, use the
read()method on the builder, for example:
Configuration c2 = new ConfigurationBuilder() // Read in C1 to provide defaults .read(c1) .clustering() .l1() .enable() // This cache is DIST_SYNC, will have 5 owners, with L1 cache enabled .build();
This completely replaces the old system of defining a set of overrides on bean properties. Note that this means the behaviour of Infinispan configuration is somewhat different when used programmatically. Whilst before, you could define a default configuration, and any overrides would be applied on top of your defaults when defined, now you must explicitly read in your defaults to the builder. This allows for much greater flexibility in your code (you can have a as many "default" configurations as you want), and makes your code more explicit and type safe (finding references works).
The schema is unchanged from before. Infinispan 4.0 configurations are currently not being parsed. To upgrade, just change the schema definition from:
<infinispan xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="urn:infinispan:config:4.1 http://www.infinispan.org/schemas/infinispan-config-4.1.xsd" xmlns="urn:infinispan:config:4.1">
<infinispan xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="urn:infinispan:config:5.1 http://www.infinispan.org/schemas/infinispan-config-5.1.xsd" xmlns="urn:infinispan:config:5.1">
The schema documentation has changed format, as it is now produced using the standard tool
xsddoc. This should be a significant improvement, as better navigation is offered. Some elements and attributes are missing docs right now, we are working on adding this. As an added benefit, your IDE should now show documentation when an xsd referenced (as above)
We are in the process of adding in support for this configuration style for modules (such as cache stores). In the meantime, please use the old configuration or XML if you require support for cache store module configuration.
ClassLoader API has changed. In the past, the following would work:
cache.withFlags(f1, f2); cache.withClassLoader(cl); cache.put(k, v);
In 5.1.0, these
withX() methods return a new instance and not the cache itself, so thread locals are avoided and the code above will not work. If used in a fluent manner however, things still work:
cache.withFlags(f1, f2).withClassLoader(cl).put(k, v);
The above pattern has always been the intention of this API anyway.