ElasticSearch 的一次非正常master脱离的调查 （转 和我碰到的情况一模一样）

转自 http://simonlei.iteye.com/blog/1669992

一共有4个节点的cluster，其中es4 是master，某个时间突然es1脱离了整个cluster，调查过程如下：

[es@bigdata1 logs]$ date; ssh bd4 date

2012年 09月 03日 星期一 09:41:26 CST

2012年 09月 03日 星期一 09:41:00 CST

es4比 es1 慢 26 秒，以下日志时间修改为es1的时间

在es4的日志中：

 写道

[2012-09-01 04:43:52,174][INFO ][cluster.service ] [es4] removed {[es1][VMFIvWAhRQuYoCcPIeedIQ][inet[/192.168.2.11:9300]],}, reason: zen-disco-node_failed([es1][VMFIvWAhRQuYoCcPIeedIQ][inet[/192.168.2.11:9300]]), reason failed to ping, tried [3] times, each with maximum [30s] timeout

这个时候，es4已经重试了3次，每次30s，也就是说90s以前，es1就已经出问题了，也就是42.22秒的时候。这个时间段es1发生了什么？

 写道

[2012-09-01 04:41:07,849][INFO ][monitor.jvm ] [es1] [gc][PS Scavenge][145683][1813] duration [5.9s], collections [1]/[6.5s], total [5.9s]/[46.1m], memory [21.8gb]->[15.7gb]/[23.3gb], all_pools {[Code Cache] [12.3mb]->[12.3mb]/[48mb]}{[PS Eden Space] [6.1gb]->[2.5mb]/[6.5gb]}{[PS Survivor Space] [289.5mb]->[214.6mb]/[820.4mb]}{[PS Old Gen] [15.3gb]->[15.5gb]/[16.2gb]}{[PS Perm Gen] [33.1mb]->[33.1mb]/[82mb]}
[2012-09-01 04:41:31,398][INFO ][monitor.jvm ] [es1] [gc][PS Scavenge][145701][1814] duration [5.8s], collections [1]/[6.5s], total [5.8s]/[46.2m], memory [22gb]->[15.8gb]/[23.3gb], all_pools {[Code Cache] [12.3mb]->[12.3mb]/[48mb]}{[PS Eden Space] [6.2gb]->[932.2kb]/[6.5gb]}{[PS Survivor Space] [214.6mb]->[259.8mb]/[815.8mb]}{[PS Old Gen] [15.5gb]->[15.6gb]/[16.2gb]}{[PS Perm Gen] [33.1mb]->[33.1mb]/[82mb]}
[2012-09-01 04:43:59,621][INFO ][monitor.jvm ] [es1] [gc][PS Scavenge][145745][1817] duration [8.9s], collections [1]/[1.6m], total [8.9s]/[46.5m], memory [22.8gb]->[2.1gb]/[23.3gb], all_pools {[Code Cache] [12.3mb]->[12.3mb]/[48mb]}{[PS Eden Space] [6.5gb]->[53.4kb]/[6.7gb]}{[PS Survivor Space] [318.4mb]->[0b]/[737.2mb]}{[PS Old Gen] [15.9gb]->[2.1gb]/[16.2gb]}{[PS Perm Gen] [33.1mb]->[33.1mb]/[82mb]}
[2012-09-01 04:43:59,621][WARN ][monitor.jvm ] [es1] [gc][PS MarkSweep][145745][14] duration [1.4m], collections [1]/[1.6m], total [1.4m]/[4.9m], memory [22.8gb]->[2.1gb]/[23.3gb], all_pools {[Code Cache] [12.3mb]->[12.3mb]/[48mb]}{[PS Eden Space] [6.5gb]->[53.4kb]/[6.7gb]}{[PS Survivor Space] [318.4mb]->[0b]/[737.2mb]}{[PS Old Gen] [15.9gb]->[2.1gb]/[16.2gb]}{[PS Perm Gen] [33.1mb]->[33.1mb]/[82mb]}
[2012-09-01 04:44:00,795][WARN ][transport.netty ] [es1] Exception caught on netty layer [[id: 0x3b3b2835, /192.168.2.14:55393 => /192.168.2.11:9300]]
java.io.IOException: 断开的管道

这期间，发生了 5.9s、5.8s、8.9s、1.4m的gc动作，特别是最后一个gc，长达1.4分钟，接近90s了。感觉应该正是这个gc导致es1无响应，从而从cluster当中被踢出去了。

很有意思的是，es1发现master es4不在了，它重新选举es3做为master，但是紧接着es3也失效了，日志如下：

 写道

[2012-09-01 04:44:03,756][INFO ][discovery.zen ] [es1] master_left [[es4][uAc5D7jkT_esrUTEBMmATQ][inet[/192.168.2.14:9300]]], reason [do not exists on master, act as master failure]
[2012-09-01 04:44:04,631][INFO ][cluster.service ] [es1] master {new [es3][F4mB4HYkTfu9umD8CSkcnw][inet[/192.168.2.13:9300]], previous [es4][uAc5D7jkT_esrUTEBMmATQ][inet[/192.168.2.14:9300]]}, removed {[es4][uAc5D7jkT_esrUTEBMmATQ][inet[/192.168.2.14:9300]],}, reason: zen-disco-master_failed ([es4][uAc5D7jkT_esrUTEBMmATQ][inet[/192.168.2.14:9300]])
[2012-09-01 04:44:06,782][INFO ][discovery.zen ] [es1] master_left [[es3][F4mB4HYkTfu9umD8CSkcnw][inet[/192.168.2.13:9300]]], reason [no longer master]
[2012-09-01 04:44:06,782][INFO ][cluster.service ] [es1] master {new [es1][VMFIvWAhRQuYoCcPIeedIQ][inet[/192.168.2.11:9300]], previous [es3][F4mB4HYkTfu9umD8CSkcnw][inet[/192.168.2.13:9300]]}, removed {[es3][F4mB4HYkTfu9umD8CSkcnw][inet[/192.168.2.13:9300]],}, reason: zen-disco-master_failed ([es3][F4mB4HYkTfu9umD8CSkcnw][inet[/192.168.2.13:9300]])

我们看一下es3当时发生了什么事情：

[es@bigdata1 logs]$ date; ssh bd3 date

2012年 09月 03日 星期一 09:51:14 CST

2012年 09月 03日 星期一 09:51:11 CST

它们只差3秒，es3的日志如下：

 写道

[2012-09-01 04:43:45,985][INFO ][cluster.service ] [es3] removed {[es1][VMFIvWAhRQuYoCcPIeedIQ][inet[/192.168.2.11:9300]],}, reason: zen-disco-receive(from master [[es4][uAc5D7jkT_esrUTEBMmATQ][inet[/192.168.2.14:9300]]])
[2012-09-01 04:44:12,631][DEBUG][action.admin.indices.status] [es3] [xabindex3][2], node[hLoNg9d0T1GYz1tqUSJwUA], [R], s[INITIALIZING]: Failed to execute [org.elasticsearch.action.admin.indices.status.IndicesStatusRequest@50d8a381]
org.elasticsearch.transport.RemoteTransportException: [es2][inet[/192.168.2.12:9300]][indices/status/s]

奇怪，04:44:06左右，es3啥都没发生，感觉就是es3没理睬es1，因此es1只好又抛弃es3，把自己组建成独立的master。

现象找到了，该怎么解决了。两个思路：

1. 把gc的时间尽量再压缩，哪怕多进行几次gc，每次时间不要太长

2. 修改zen的配置，把fault detection的timeout时间和retry times 都增加。

第一步比较麻烦，还是第二步比较简单，先把retries修改为6试试看。