Class OrderedGracefulLeaveHandler

java.lang.Object

org.infinispan.topology.OrderedGracefulLeaveHandler

public class OrderedGracefulLeaveHandler
extends Object

Serializes graceful leave operations on the coordinator to prevent data loss during concurrent scale-down.

When multiple nodes leave simultaneously (e.g., Kubernetes scaling down a StatefulSet), each leave triggers a rebalance that redistributes data to surviving nodes. Without serializing the operations, all leaves execute before any rebalance completes, causing segments to lose all owners.

This handler maintains a per-cache chain of CompletableFutures. Each leave waits for the previous leave's rebalance to finish before proceeding:

  Leave B --(doLeave + wait rebalance)--> Leave C --(doLeave + wait rebalance)--> ...

Only used when the leaving node requests a graceful stop with a positive timeout (via CacheLeaveCommand). Abrupt leaves (view change, kill) bypass this handler entirely.

Since:

16.2

See Also:

• ClusterTopologyManagerImpl.handleLeave(String, Address, long, TimeUnit)
• StateTransferTracker

Field Summary

Fields

Modifier and Type	Field	Description
protected BlockingManager	blockingManager
protected TimeService	timeService
protected org.infinispan.statetransfer.StateTransferTracker	tracker

Constructor Summary

Constructors

Constructor	Description
OrderedGracefulLeaveHandler()

Method Summary

Modifier and Type	Method	Description
boolean	drainAll(long timeout, TimeUnit unit)	Waits for all in-flight graceful leave chains to complete.
CompletionStage<Void>	enqueue(String cacheName, Address leaver, long timeout, TimeUnit unit, ClusterCacheStatus status)	Enqueues a graceful leave for the given cache, serialized behind any pending leaves.
void	remove(String cacheName)	Removes the leave chain for a cache.

Methods inherited from class Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Details

blockingManager

protected BlockingManager blockingManager

timeService

protected TimeService timeService

tracker

protected org.infinispan.statetransfer.StateTransferTracker tracker

Constructor Details

OrderedGracefulLeaveHandler

public OrderedGracefulLeaveHandler()

Method Details

enqueue

public CompletionStage<Void> enqueue(String cacheName,
 Address leaver,
 long timeout,
 TimeUnit unit,
 ClusterCacheStatus status)

Enqueues a graceful leave for the given cache, serialized behind any pending leaves.

Parameters:

cacheName - the cache the node is leaving

leaver - the address of the leaving node

timeout - maximum time to wait for the rebalance triggered by this leave

unit - time unit for the timeout

status - the coordinator-side cache status, used to execute the leave and read topology state

Returns:

a stage that completes when this leave's rebalance finishes (or the timeout expires)

remove

public void remove(String cacheName)

Removes the leave chain for a cache. Called when the last member leaves and the cache status is removed.

Parameters:

cacheName - the cache to clean up

drainAll

public boolean drainAll(long timeout,
 TimeUnit unit)
                 throws InterruptedException

Waits for all in-flight graceful leave chains to complete.

The method ensures that all currently pending leave requests are consumed before returning. Concurrent leave requests included while draining are also included in the consuming list. The method waits until all the requests and associated state transfer complete, or a timeout elapses.

Parameters:

timeout - maximum time to wait for all chains to complete

unit - time unit for the timeout

Returns:

true if all chains completed before the deadline; false if the timeout expired

Throws:

InterruptedException - if the calling thread is interrupted while waiting