重启的namenode遭受块报告风暴

Restarted namenode suffer from block report storm

当我们基于hadoop v2.4.1的standby namenode故障重启时，发现namenode退出safemode后忙得无法及时响应。我们扔了好几堆，它们看起来都是这样的，

Thread 212 (IPC Server handler 148 on 8020):
  State: WAITING
  Blocked count: 66
  Waited count: 598
  Waiting on java.util.concurrent.locks.ReentrantLock$FairSync@60ea5634
  Stack:
    sun.misc.Unsafe.park(Native Method)
    java.util.concurrent.locks.LockSupport.park(LockSupport.java:186)
    java.util.concurrent.locks.AbstractQueuedSynchronizer.parkAndCheckInterrupt(AbstractQueuedSynchronizer.java:834)
    java.util.concurrent.locks.AbstractQueuedSynchronizer.acquireQueued(AbstractQueuedSynchronizer.java:867)
    java.util.concurrent.locks.AbstractQueuedSynchronizer.acquire(AbstractQueuedSynchronizer.java:1197)
    java.util.concurrent.locks.ReentrantLock$FairSync.lock(ReentrantLock.java:229)
    java.util.concurrent.locks.ReentrantLock.lock(ReentrantLock.java:290)
    org.apache.hadoop.hdfs.server.namenode.FSNamesystem.writeLock(FSNamesystem.java:1378)
    org.apache.hadoop.hdfs.server.blockmanagement.BlockManager.processReport(BlockManager.java:1676)
    org.apache.hadoop.hdfs.server.namenode.NameNodeRpcServer.blockReport(NameNodeRpcServer.java:1019)
    org.apache.hadoop.hdfs.protocolPB.DatanodeProtocolServerSideTranslatorPB.blockReport(DatanodeProtocolServerSideTranslatorPB.java:152)
    org.apache.hadoop.hdfs.protocol.proto.DatanodeProtocolProtos$DatanodeProtocolService.callBlockingMethod(DatanodeProtocolProtos.java:28061)
    org.apache.hadoop.ipc.ProtobufRpcEngine$Server$ProtoBufRpcInvoker.call(ProtobufRpcEngine.java:585)
    org.apache.hadoop.ipc.RPC$Server.call(RPC.java:928)
    org.apache.hadoop.ipc.Server$Handler.run(Server.java:2013)
    org.apache.hadoop.ipc.Server$Handler.run(Server.java:2009)
    java.security.AccessController.doPrivileged(Native Method)
    javax.security.auth.Subject.doAs(Subject.java:415)
    org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1556)
    org.apache.hadoop.ipc.Server$Handler.run(Server.java:2007)

几乎所有服务器处理程序都在等待 FSNameSystem#writeLock 处理 incremental/full 报告！

配置文件：

dfs.blockreport.initialDelay: 120.
dfs.blockreport.intervalMsec: 6h.
server handlers number: 200.
datanodes number:400.
The namenode takes 0.5~1s to process a block report.

datanodes的日志看到很多offerService IOExceptions和重试。

NN日志显示一个datanode的存储被处理了两次以上，有的甚至达到10次。

blockLog.info("BLOCK* processReport: from storage " + storage.getStorageID() + " node " + nodeID + ", blocks: " + newReport.getNumberOfBlocks() + ", processing time: " + (endTime - startTime) + " msecs");

有没有人遇到过同样的问题，有什么想法吗？

最后我们根据服务器的特点，减少了服务器的IPC handler数量，解决了问题。希望对遇到同样问题的小伙伴有所帮助！

重启的namenode遭受块报告风暴

Restarted namenode suffer from block report storm

java

multithreading

hadoop

rpc