zookeeper频繁异常问题分析

Reference: https://blog.csdn.net/xjping0794/article/details/77784171

1.1            操作系统信息
1.1.1          CPU信息

从上述可以看出在问题点CPU使用率并不高，故排除CPU问题。

1.1.2          内存信息

CM监控界面并无内存信息，无法提供图片。

但问题期间，监控机器内存，发现还有剩余，故排除内存问题。

1.2            ZKFAILOVER日志信息

2017-09-0109:45:57,390 INFO org.apache.zookeeper.ClientCnxn: Client session timed out,have not heard from server in 1668ms for sessionid 0x0, closing socketconnection and attempting reconnect

2017-09-01 09:45:58,224 ERRORorg.apache.hadoop.ha.ActiveStandbyElector: Connection timed out: couldn'tconnect to ZooKeeper in 5000 milliseconds

2017-09-0109:45:58,450 INFO org.apache.zookeeper.ZooKeeper: Session: 0x0 closed

2017-09-0109:45:58,450 INFO org.apache.zookeeper.ClientCnxn: EventThread shut down

2017-09-0109:45:58,451 WARN org.apache.hadoop.ha.ActiveStandbyElector:org.apache.zookeeper.KeeperException$ConnectionLossException: KeeperErrorCode =ConnectionLoss

2017-09-0109:46:03,453 INFO org.apache.zookeeper.ZooKeeper: Initiating client connection,connectString=test-ssps-s-02:2181,test-ssps-s-03:2181,test-ssps-s-04:2181sessionTimeout=5000watcher=org.apache.hadoop.ha.ActiveStandbyElector$WatcherWithClientRef@7f0d36c

2017-09-0109:46:03,455 INFO org.apache.zookeeper.ClientCnxn: Opening socket connection toserver test-ssps-s-04/10.117.210.216:2181. Will not attempt to authenticateusing SASL (unknown error)

2017-09-0109:46:03,463 INFO org.apache.zookeeper.ClientCnxn: Socket connectionestablished to test-ssps-s-04/10.117.210.216:2181, initiating session

2017-09-0109:46:05,131 INFO org.apache.zookeeper.ClientCnxn: Client session timed out,have not heard from server in 1668ms for sessionid 0x0, closing socketconnection and attempting reconnect

2017-09-0109:46:05,885 INFO org.apache.zookeeper.ClientCnxn: Opening socket connection toserver test-ssps-s-03/10.51.20.155:2181. Will not attempt to authenticate usingSASL (unknown error)

2017-09-0109:46:06,626 INFO org.apache.zookeeper.ClientCnxn: Socket connectionestablished to test-ssps-s-03/10.51.20.155:2181, initiating session

2017-09-0109:46:08,293 INFO org.apache.zookeeper.ClientCnxn: Client session timed out,have not heard from server in 1667ms for sessionid 0x0, closing socketconnection and attempting reconnect

2017-09-01 09:46:08,454 ERROR org.apache.hadoop.ha.ActiveStandbyElector:Connection timed out: couldn't connect to ZooKeeper in 5000 milliseconds

2017-09-0109:46:09,157 INFO org.apache.zookeeper.ZooKeeper: Session: 0x0 closed

2017-09-01 09:46:09,157 INFO org.apache.zookeeper.ClientCnxn:EventThread shut down

2017-09-0109:46:09,158 WARN org.apache.hadoop.ha.ActiveStandbyElector:org.apache.zookeeper.KeeperException$ConnectionLossException: KeeperErrorCode =ConnectionLoss

标红日志，显示无法与zk建立连接，导致failover终止。

1.3            YARN RESOURCEMANAGER日志信息
具体日志信息未保留，但均是由于ZK异常，导致服务挂掉。

1.4            ZOOKEEPER服务日志
摘取leader和follower 服务日志分析。

1.4.1           leader日志
分析异常时间点9点40到9点50日志，发现以下报错信息：

2017-09-0109:45:49,754 INFO org.apache.zookeeper.server.NIOServerCnxnFactory: Acceptedsocket connection from /10.51.20.155:40713

2017-09-0109:45:49,754 INFO org.apache.zookeeper.server.ZooKeeperServer: Clientattempting to establish new session at /10.51.20.155:40713

2017-09-0109:45:51,090 INFO org.apache.zookeeper.server.PrepRequestProcessor: Gotuser-level KeeperException when processing sessionid:0x15e2a2922a213b5type:setData cxid:0x21 zxid:0x6d00050f33 txntype:-1 reqpath:n/a ErrorPath:/yarn-leader-election/yarnRM/ActiveBreadCrumb Error:KeeperErrorCode =BadVersion for /yarn-leader-election/yarnRM/ActiveBreadCrumb

2017-09-0109:45:52,000 INFO org.apache.zookeeper.server.ZooKeeperServer: Expiring session0x35e2a2922671668, timeout of 5000ms exceeded

2017-09-0109:45:52,001 INFO org.apache.zookeeper.server.PrepRequestProcessor: Processedsession termination for sessionid: 0x35e2a2922671668

2017-09-0109:45:52,482 INFO org.apache.zookeeper.server.NIOServerCnxnFactory: Acceptedsocket connection from /10.117.68.10:33589

2017-09-0109:45:52,484 INFO org.apache.zookeeper.server.ZooKeeperServer: Clientattempting to renew session 0x15e2a2922a21870 at /10.117.68.10:33589

2017-09-0109:45:52,484 INFO org.apache.zookeeper.server.ZooKeeperServer: Establishedsession 0x15e2a2922a21870 with negotiated timeout 5000 for client/10.117.68.10:33589

2017-09-0109:45:52,884 WARN org.apache.zookeeper.server.NIOServerCnxn: caught end ofstream exception

2017-09-0109:45:52,884 INFO org.apache.zookeeper.server.NIOServerCnxn: Closed socketconnection for client /10.51.20.155:40712 which had sessionid 0x15de73379b60000

2017-09-0109:45:53,244 INFO org.apache.zookeeper.server.NIOServerCnxnFactory: Acceptedsocket connection from /10.51.20.155:40730

2017-09-0109:46:36,520 INFO org.apache.zookeeper.server.NIOServerCnxnFactory: Acceptedsocket connection from /10.51.20.155:40749

2017-09-01 09:46:29,773 WARN org.apache.zookeeper.server.persistence.FileTxnLog:fsync-ing the write ahead log in SyncThread:3 took53812ms which will adversely effect operation latency. See theZooKeeper troubleshooting guide

2017-09-01 09:46:17,239 INFOorg.apache.zookeeper.server.quorum.Leader: Shutting down

2017-09-0109:45:58,000 INFO org.apache.zookeeper.server.ZooKeeperServer: Expiring session0x15e2a2922a21870, timeout of 5000ms exceeded

2017-09-01 09:45:53,439 ERRORorg.apache.zookeeper.server.NIOServerCnxn: Unexpected Exception:

java.nio.channels.CancelledKeyException atsun.nio.ch.SelectionKeyImpl.ensureValid(SelectionKeyImpl.java:73) atsun.nio.ch.SelectionKeyImpl.interestOps(SelectionKeyImpl.java:77) atorg.apache.zookeeper.server.NIOServerCnxn.sendBuffer(NIOServerCnxn.java:418) atorg.apache.zookeeper.server.NIOServerCnxn.sendResponse(NIOServerCnxn.java:1509)atorg.apache.zookeeper.server.FinalRequestProcessor.processRequest(FinalRequestProcessor.java:171)atorg.apache.zookeeper.server.quorum.CommitProcessor.run(CommitProcessor.java:73)

2017-09-0109:46:36,648 INFO org.apache.zookeeper.server.ZooKeeperServer: Expiring session0x15e2a2922a21871, timeout of 10000ms exceeded

2017-09-0109:46:36,648 INFO org.apache.zookeeper.server.ZooKeeperServer: Expiring session0x15e2a2922a213b5, timeout of 10000ms exceeded

2017-09-0109:46:36,648 INFO org.apache.zookeeper.server.ZooKeeperServer: Expiring session0x15de73379b60000, timeout of 6000ms exceeded

2017-09-0109:46:36,648 INFO org.apache.zookeeper.server.ZooKeeperServer: Expiring session0x25de7337bb00000, timeout of 6000ms exceeded

2017-09-0109:46:36,648 INFO org.apache.zookeeper.server.ZooKeeperServer: Expiring session0x35e2a2922671890, timeout of 10000ms exceeded

2017-09-0109:46:36,648 INFO org.apache.zookeeper.server.ZooKeeperServer: Expiring session0x15e2a2922a213b6, timeout of 10000ms exceeded

2017-09-01 09:46:36,647 INFOorg.apache.zookeeper.server.quorum.Leader: Shutdown called

从标红日志信息可以看出，zookeerper 服务在同步日志过程中耗时太长，花了53812ms（正常应该在3秒内），同步日志会导致ZK无法响应外部请求，进而引发session过期，进而引发zk 服务端shut down。

另外，CancelledKeyException错误，是由于session失效后，socket已关闭，但服务端仍往该session发送回复信号，引发该错误，该错误并不致命，影响不大。是zookeeper版本（3.4.5）bug所致，已在ZK新版本中优化掉。

其他点日志报同样问题，尤其10点33分，zk刷新日志竟然花了将近2分钟，导致服务异常退出。

1.4.2          Follower日志

分析异常时间点13点00到13点10日志，发现以下报错信息：

2017-09-0113:08:55,526 INFO org.apache.zookeeper.server.ZooKeeperServer: Establishedsession 0x15e3b4962bc0000 with negotiated timeout 6000 for client/10.117.210.216:39709

2017-09-0113:08:57,259 INFO org.apache.zookeeper.server.NIOServerCnxnFactory: Acceptedsocket connection from /10.51.20.155:53325

2017-09-0113:08:57,260 INFO org.apache.zookeeper.server.ZooKeeperServer: Clientattempting to renew session 0x25e3b531e84006e at /10.51.20.155:53325

2017-09-0113:08:57,261 INFO org.apache.zookeeper.server.ZooKeeperServer: Established session0x25e3b531e84006e with negotiated timeout 6000 for client /10.51.20.155:53325

2017-09-0113:08:57,261 INFO org.apache.zookeeper.server.PrepRequestProcessor: Gotuser-level KeeperException when processing sessionid:0x25e3b531e84006etype:delete cxid:0x11e3d zxid:0x7000006e8f txntype:-1 reqpath:n/a ErrorPath:/admin/preferred_replica_election Error:KeeperErrorCode = NoNode for/admin/preferred_replica_election

2017-09-0113:08:59,525 WARN org.apache.zookeeper.server.NIOServerCnxn: caught end ofstream exception

2017-09-0113:08:59,526 INFO org.apache.zookeeper.server.NIOServerCnxn: Closed socketconnection for client /10.117.210.216:39709 which had sessionid0x15e3b4962bc0000

2017-09-01 13:09:00,302 WARNorg.apache.zookeeper.server.persistence.FileTxnLog: fsync-ing the write aheadlog in SyncThread:3 took 35356ms which will adversely effect operation latency.See the ZooKeeper troubleshooting guide

2017-09-0113:09:01,263 WARN org.apache.zookeeper.server.NIOServerCnxn: caught end ofstream exception

2017-09-0113:09:01,264 INFO org.apache.zookeeper.server.NIOServerCnxn: Closed socketconnection for client /10.51.20.155:53325 which had sessionid 0x25e3b531e84006e

原因同上。

1.5            磁盘

在问题点，发现IO使用很高。

9点46分，问题出现后，基于当时的集群IO值分析，发现IO达到200M/s，同样，10点33分IO也达到了将近200M/s，磁盘IO可能是一个瓶颈。

1.6            网络

网络正常。

2  总结及建议

从监控及分析结果来看，均属ZK服务端在fsync-ing the write ahead log日志时超长引起。从问题产生时间段的IO来分析，发现磁盘IO较正常点高了很多，达到200M/s，证明问题发生时，有大作业在进行磁盘IO，与开发人员杨仪军确认，其今天确实较以往跑了大数据的作业。颈，需进一步通过专业工具测试磁盘IO能力。

2.1.1          官网建议
关于ZK日志存放，官网给出如下建议:

Having a dedicated log devicehas a large impact on throughput and stable latencies. It is highly recommenedto dedicate a log device and set dataLogDir to point to a directory on thatdevice, and then make sure to point dataDir to a directory not residing on thatdevice.

故为避免此类问题，dataLogDir存放目录应该与dataDir分开，可单独采用一套存储设备来存放ZK日志。

2.1.2          磁盘IO

从今天观察情况来看，磁盘IO严重不行，200M/s即达到瓶颈，需进一步通过专业工具测试磁盘IO能力。

2.1.3          关闭forceSync参数

在ZOO.CFG中增加：

forceSync=no

默认是开启的，为避免同步延迟问题，ZK接收到数据后会立刻去讲当前状态信息同步到磁盘日志文件中，同步完成后才会应答。将此项关闭后，客户端连接可以得到快速响应。Zk涮日志源码如下图：

关闭forceSync选项后，会存在潜在风险，虽然依旧会刷磁盘（log.flush()首先被执行），但因为操作系统为提高写磁盘效率，会先写缓存，当机器异常后，可能导致一些zk状态信息没有同步到磁盘，从而带来ZK前后信息不一样问题。

2.1.4          解决CancelledKeyException
该问题，已在ZooKeeper 3.4.8版本中得到修复。

如需解决该版本问题，可打补丁https://issues.apache.org/jira/browse/ZOOKEEPER-1237

然后重新编译ZK并使用。

---------------------
作者：jimmyxyalj
来源：CSDN
原文：https://blog.csdn.net/xjping0794/article/details/77784171
版权声明：本文为博主原创文章，转载请附上博文链接！