HBase运维-节点故障Server REPORT rejected;

hbase性能测试，加载了一个晚上的数据，早上来时发现一个节点挂掉了，其他一切正常。

查看日志，发下如下问题

12/01/04 09:45:39 FATAL regionserver.HRegionServer: ABORTING region server serverName=hadoop5.site,60020,1325663355680, load=(requests=983, regions=252, usedHeap=3085, maxHeap=4983): Unhandled exception: org.apache.hadoop.hbase.YouAreDeadException: Server REPORT rejected; currently processing hadoop5.site,60020,1325663355680 as dead server
org.apache.hadoop.hbase.YouAreDeadException: org.apache.hadoop.hbase.YouAreDeadException: Server REPORT rejected; currently processing hadoop5.site,60020,1325663355680 as dead server
    at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method)
    at sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:39)
    at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:27)
    at java.lang.reflect.Constructor.newInstance(Constructor.java:513)
    at org.apache.hadoop.ipc.RemoteException.instantiateException(RemoteException.java:95)
    at org.apache.hadoop.ipc.RemoteException.unwrapRemoteException(RemoteException.java:79)
    at org.apache.hadoop.hbase.regionserver.HRegionServer.tryRegionServerReport(HRegionServer.java:735)
    at org.apache.hadoop.hbase.regionserver.HRegionServer.run(HRegionServer.java:596)
    at java.lang.Thread.run(Thread.java:619)
Caused by: org.apache.hadoop.ipc.RemoteException: org.apache.hadoop.hbase.YouAreDeadException: Server REPORT rejected; currently processing hadoop5.site,60020,1325663355680 as dead server
    at org.apache.hadoop.hbase.master.ServerManager.checkIsDead(ServerManager.java:204)
    at org.apache.hadoop.hbase.master.ServerManager.regionServerReport(ServerManager.java:262)
    at org.apache.hadoop.hbase.master.HMaster.regionServerReport(HMaster.java:669)
    at sun.reflect.GeneratedMethodAccessor2.invoke(Unknown Source)
    at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:25)
    at java.lang.reflect.Method.invoke(Method.java:597)
    at org.apache.hadoop.hbase.ipc.HBaseRPC$Server.call(HBaseRPC.java:570)
    at org.apache.hadoop.hbase.ipc.HBaseServer$Handler.run(HBaseServer.java:1039)

    at org.apache.hadoop.hbase.ipc.HBaseClient.call(HBaseClient.java:771)
    at org.apache.hadoop.hbase.ipc.HBaseRPC$Invoker.invoke(HBaseRPC.java:257)
    at $Proxy6.regionServerReport(Unknown Source)
    at org.apache.hadoop.hbase.regionserver.HRegionServer.tryRegionServerReport(HRegionServer.java:729)
    ... 2 more

 

再往上找，可以看到

2012-01-04T09:42:27.829-0500: 24795.829: [GC 24795.829: [ParNew: 151317K->10586K(153344K), 0.5282750 secs] 4970251K->4832124K(5102976K) icms_dc=0 , 0.5284260 secs] [Times: user=3.29 sys=0.01, real=0.53 secs]
2012-01-04T09:42:28.721-0500: 24796.721: [GC 24796.721: [ParNew (promotion failed): 146906K->140702K(153344K), 0.5622020 secs]24797.283: [CMS: 4824062K->3150755K(4949632K), 189.5658760 secs] 4968444K->3150755K(5102976K), [CMS Perm : 20156K->20153K(33704K)] icms_dc=0 , 190.1283170 secs] [Times: user=7.43 sys=0.96, real=190.14 secs]
2012-01-04T09:45:38.852-0500: 24986.852: [GC [1 CMS-initial-mark: 3150755K(4949632K)] 3152726K(5102976K), 0.0015480 secs] [Times: user=0.00 sys=0.00, real=0.00 secs]
2012-01-04T09:45:38.853-0500: 24986.854: [CMS-concurrent-mark-start]

12/01/04 09:45:38 INFO zookeeper.ClientCnxn: Client session timed out, have not heard from server in 237682ms for sessionid 0x34a7b17bf80004, closing socket connection and attempting reconnect
12/01/04 09:45:38 INFO zookeeper.ClientCnxn: Client session timed out, have not heard from server in 237682ms for sessionid 0x34a7b17bf80003, closing socket connection and attempting reconnect

12/01/04 09:45:38 WARN ipc.HBaseServer: IPC Server Responder, call getClosestRowBefore([B@166cb249, [B@3a2ce21f, [B@58b17f0f) from 192.9.200.164:34106: output error
12/01/04 09:45:38 WARN ipc.HBaseServer: IPC Server Responder, call getClosestRowBefore([B@6816a498, [B@26902c8b, [B@435c6d74) from 192.9.200.238:38457: output error
12/01/04 09:45:38 WARN ipc.HBaseServer: IPC Server Responder, call getClosestRowBefore([B@23b4b286, [B@2c348dba, [B@2e44c502) from 192.9.200.164:34106: output error
12/01/04 09:45:38 WARN ipc.HBaseServer: PRI IPC Server handler 6 on 60020 caught: java.nio.channels.ClosedChannelException
    at sun.nio.ch.SocketChannelImpl.ensureWriteOpen(SocketChannelImpl.java:126)
    at sun.nio.ch.SocketChannelImpl.write(SocketChannelImpl.java:324)
    at org.apache.hadoop.hbase.ipc.HBaseServer.channelWrite(HBaseServer.java:1341)
    at org.apache.hadoop.hbase.ipc.HBaseServer$Responder.processResponse(HBaseServer.java:727)
    at org.apache.hadoop.hbase.ipc.HBaseServer$Responder.doRespond(HBaseServer.java:792)
    at org.apache.hadoop.hbase.ipc.HBaseServer$Handler.run(HBaseServer.java:1083)

 

系统在挂掉前进行了一次长达190s+的gc，导致长时间未与ZooKeeper通信，系统就认为这个节点挂掉了。

 

再分析这次fullgc的原因ParNew (promotion failed)

 

这个问题的产生是由于救助空间不够，从而向年老代转移对象，年老代没有足够的空间来容纳这些对象，导致一次full gc的产生。 解决这个问题的办法有两种完全相反的倾向：增大救助空间、增大年老代或者去掉救助空间 。 增大救助空间就是调整-XX:SurvivorRatio参数，这个参数是Eden区和Survivor区的大小比值，默认是32，也就是说Eden区是 Survivor区的32倍大小，要注意Survivo是有两个区的，因此Surivivor其实占整个young genertation的1/34。调小这个参数将增大survivor区，让对象尽量在survitor区呆长一点，减少进入年老代的对象。去掉救助空 间的想法是让大部分不能马上回收的数据尽快进入年老代，加快年老代的回收频率，减少年老代暴涨的可能性，这个是通过将-XX:SurvivorRatio 设置成比较大的值（比如65536)来做到。

 

还有一个系统的原因，那是因为这台机器比别的节点多部署了一个约占2G内存的应用，导致这台机器挂掉，但是其他机器没有出现问题