Hazelcast羣集成員因「IsStillRunningService」對象數量龐大而導致內存不足

Question

我們有一個系統在3.5版本中使用Hazelcast IExecutor Service和IMap。我們最近遇到了Hazelcast集羣成員一個接一個地出現內存不足，最後所有節點都與OOM崩潰。
在進行因果分析時，我們發現有數千個以下的日誌條目和日誌文件大小呈指數級增長。另外存在日誌的存儲空間也已經空間不足。

WARNING: [10.7.90.189]:30103 [FB] [3.5] Asking if operation execution has been started: com.hazelcast.spi.impl.operationservice.impl.IsStillRunningService$InvokeIsStillRunningOperationRunnable@48b3ac3b 
Mar 30, 2016 11:09:29 AM com.hazelcast.spi.impl.operationservice.impl.Invocation 
WARNING: [10.7.90.189]:30103 [FB] [3.5] While asking 'is-executing': Invocation{ serviceName='hz:core:partitionService', op=com.hazelcast.spi.impl.operationservice.impl.operations.IsStillExecutingOperation{serviceName='hz:core:partition 
Service', partitionId=-1, callId=59834, invocationTime=1459349279980, waitTimeout=-1, callTimeout=5000}, partitionId=-1, replicaIndex=0, tryCount=0, tryPauseMillis=0, invokeCount=1, callTimeout=5000, target=Address[1.2.3.4]:30102, b 
ackupsExpected=0, backupsCompleted=0} 
com.hazelcast.core.OperationTimeoutException: No response for 10000 ms. Aborting invocation! Invocation{ serviceName='hz:core:partitionService', op=com.hazelcast.spi.impl.operationservice.impl.operations.IsStillExecutingOperation{servic 
eName='hz:core:partitionService', partitionId=-1, callId=268177, invocationTime=1459349295209, waitTimeout=-1, callTimeout=5000}, partitionId=-1, replicaIndex=0, tryCount=0, tryPauseMillis=0, invokeCount=1, callTimeout=5000, target=Addr 
ess[10.7.90.190]:30102, backupsExpected=0, backupsCompleted=0} No response has been received! backups-expected:0 backups-completed: 0 
     at com.hazelcast.spi.impl.operationservice.impl.Invocation.newOperationTimeoutException(Invocation.java:491) 
     at com.hazelcast.spi.impl.operationservice.impl.IsStillRunningService$IsOperationStillRunningCallback.setOperationTimeout(IsStillRunningService.java:224) 
     at com.hazelcast.spi.impl.operationservice.impl.IsStillRunningService$IsOperationStillRunningCallback.onFailure(IsStillRunningService.java:219) 
     at com.hazelcast.spi.impl.operationservice.impl.InvocationFuture$1.run(InvocationFuture.java:137) 
     at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142) 
     at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617) 
     at java.lang.Thread.run(Thread.java:745) 
     at com.hazelcast.util.executor.HazelcastManagedThread.executeRun(HazelcastManagedThread.java:76) 
     at com.hazelcast.util.executor.HazelcastManagedThread.run(HazelcastManagedThread.java:92) 

我的理解是，集羣成員將不斷的心跳，以確保所有成員都活着，我相信默認爲10秒。現在的問題是，如果任何成員無反應或休息狀態，其餘成員將繼續進行正在執行的呼叫。在查看堆轉儲後，知道堆中有73％的堆滿了「IsStillRunningService」對象。

問題：

1. 如何去了解究竟出在哪裏？
2. 用完存儲空間只是一個共同發生或可能有任何相關性？我們懷疑這可能導致其他人，因爲它在一週內發生過兩次。

Hazelcast XML配置：

<hazelcast xsi:schemaLocation="http://www.hazelcast.com/schema/config http://www.hazelcast.com/schema/config/hazelcast-config-3.5.xsd" 
      xmlns="http://www.hazelcast.com/schema/config" 
      xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"> 

    <map name="myMap"> 
     <backup-count>0</backup-count> 
     <time-to-live-seconds>43200</time-to-live-seconds> 
     <eviction-policy>LRU</eviction-policy> 
     <max-size policy="USED_HEAP_PERCENTAGE">75</max-size> 
     <eviction-percentage>10</eviction-percentage> 
     <in-memory-format>OBJECT</in-memory-format> 
    </map> 

    <executor-service name="calculation"> 
      <pool-size>10</pool-size> 
     <queue-capacity>400</queue-capacity> 
    </executor-service> 

    <executor-service name="loader"> 
      <pool-size>5</pool-size> 
     <queue-capacity>400</queue-capacity> 
    </executor-service> 

    <properties> 
     <property name="hazelcast.icmp.timeout">5000</property> 
     <property name="hazelcast.initial.wait.seconds">10</property> 
     <property name="hazelcast.connection.monitor.interval">5000</property> 
    </properties> 

    <network> 
     <port auto-increment="true" port-count="100">30101</port> 
     <join> 
      <multicast enabled="false"> 
       <multicast-group>224.2.2.3</multicast-group> 
       <multicast-port>54327</multicast-port> 
      </multicast> 
      <tcp-ip enabled="true"> 
       <interface>1.2.3.4</interface> 
       <interface>1.2.3.5</interface> 
       <interface>1.2.3.6</interface> 
      </tcp-ip> 
      <aws enabled="false"/> 
     </join> 
     <interfaces enabled="false"> 
      <interface>127.0.0.1</interface> 
     </interfaces> 
    </network> 
</hazelcast> 

StackTrace
LinkedBlockingQueue which holds IsStillRunningService Objects

Answer 1

可以升級到3.6。添加了修復程序，以防止使用is-still-running進入OOME。在3.7中，整個機制將被刪除並由一個問題較少的方法取代。

https://github.com/hazelcast/hazelcast/pull/7719