• MongoDB实例重启失败探究(大事务Redo导致)


     1.实例重启背景

    收到监控组同学反馈,连接某一个MongoDB实例的应用耗时异常,并且出现了超时。查看数据库监控平台,发现此实例服务器的IO异常飙升,而查看副本集状态(rs.status()),主从是坏掉的,从节点不可达。

    登入从节点,查看mongodb服务状态,是stop的。

    查看服务器的log,发现出现了OOM,Mongodb被关闭了。需要手动重启。

    Jan 17 12:02:48 qqorderdb02 kernel: Out of memory: Kill process 83717 (mongod) score 919 or sacrifice child
    Jan 17 12:02:48 qqorderdb02 kernel: Killed process 83717 (mongod), UID 1001, total-vm:21256876kB, anon-rss:15529572kB, file-rss:0kB, shmem-rss:0kB
    Jan 17 12:42:42 qqorderdb02 systemd[1]: mongodbqq.service: main process exited, code=killed, status=9/KILL
    Jan 17 12:42:42 qqorderdb02 systemd[1]: Unit mongodbqq.service entered failed state.
    Jan 17 12:42:42 qqorderdb02 systemd[1]: mongodbqq.service failed.

    事后分析,主节点的内存比从节点的内存大,创建索引,主节点正常执行了,而从节点出席那了OOM(12:02),KIll后,服务自启动也失败了(12:42)。

     2.重启服务

    重启,查看mongodblog,我们会看到redo未提交的创建索引的事务

    2019-01-17T19:38:11.529+0800 I CONTROL  [initandlisten] ** WARNING: /sys/kernel/mm/transparent_hugepage/enabled is 'always'.
    2019-01-17T19:38:11.529+0800 I CONTROL  [initandlisten] **        We suggest setting it to 'never'
    2019-01-17T19:38:11.529+0800 I CONTROL  [initandlisten]
    2019-01-17T19:38:11.529+0800 I CONTROL  [initandlisten] ** WARNING: /sys/kernel/mm/transparent_hugepage/defrag is 'always'.
    2019-01-17T19:38:11.529+0800 I CONTROL  [initandlisten] **        We suggest setting it to 'never'
    2019-01-17T19:38:11.529+0800 I CONTROL  [initandlisten]
    2019-01-17T19:38:11.529+0800 I CONTROL  [initandlisten] ** WARNING: Running wiredTiger with the --nojournal option in a replica set
    2019-01-17T19:38:11.529+0800 I CONTROL  [initandlisten] **          is deprecated and subject to be removed in a future version.
    2019-01-17T19:38:11.529+0800 I CONTROL  [initandlisten]
    2019-01-17T19:38:11.592+0800 I INDEX    [initandlisten] found 1 index(es) that wasn't finished before shutdown
    2019-01-17T19:38:11.595+0800 I FTDC     [initandlisten] Initializing full-time diagnostic data capture with directory '/var/mongodbqq/db/diagnostic.data'
    2019-01-17T19:38:11.596+0800 I INDEX    [initandlisten] found 1 interrupted index build(s) on qqorderdb.weixinordersn
    2019-01-17T19:38:11.596+0800 I INDEX    [initandlisten] note: restart the server with --noIndexBuildRetry to skip index rebuilds

     但执行一会后,重启失败,退出。查看server log,报错信息如下:

    Jan 17 19:41:10 qqorderdb02 systemd[1]: mongodbqq.service stop-final-sigterm timed out. Killing.
    Jan 17 19:41:10 qqorderdb02 systemd[1]: Failed to start mongodbqq_service.
    Jan 17 19:41:10 qqorderdb02 systemd[1]: Unit mongodbqq.service entered failed state.
    Jan 17 19:41:10 qqorderdb02 systemd[1]: mongodbqq.service failed.

    查看mongodb的log,最新的信息如下:

    2019-01-17T19:41:00.001+0800 I -        [initandlisten]   Index Build: 55387600/192576426 28%
    2019-01-17T19:41:03.002+0800 I -        [initandlisten]   Index Build: 57463100/192576426 29%
    2019-01-17T19:41:06.002+0800 I -        [initandlisten]   Index Build: 59385700/192576426 30%
    2019-01-17T19:41:09.001+0800 I -        [initandlisten]   Index Build: 61549000/192576426 31%

    通过查看Server log 和 Mongodb log ,我们可以判断:启动时需要重建关闭时未完成的index,但是重建这个大集合(本案例为weixinordersn,5亿数据量,102Gsize)的索引耗时较长,超过了启动服务允许的时间。服务超时后会被killed。

    3.解决方案

    设置systemd的service超时时间,在mongodb服务中,指明TimeoutSec参数。

    TimeoutSec:定义 Systemd 停止当前服务之前等待的秒数。单位是秒,设置为0是不限制.

    例如mongodbtest.service的编写,增加 TimeoutSec=1800

    [Unit]
     
    Description=mongodbtest
    After=network.target remote-fs.target nss-lookup.target
    [Service]
    User=mongouser
    Group=mongouser
    # (open files)
    LimitNOFILE=64000
    Type=forking
    ExecStart=/data/mongodb/mongobin404/bin/mongod --config /data/mongodb/mongobin404/bin/mongodbtest.conf
    ExecReload=/bin/kill -s HUP $MAINPID
    ExecStop=/data/mongodb/mongobin404/bin/mongod --shutdown --config /data/mongodb/mongobin404/bin/mongodbtest.conf
    PrivateTmp=true
    TimeoutSec=1800
     
    [Install]
    WantedBy=multi-user.target

    4.性能下降的分析

    结合 应用超时 和 数据库监控的IO飙升的时间关联,性能下降主要分别是 创建索引 和 主从断掉导致oplog.rs 插入、查询、更新变慢导致。

    5.参考文献

    1.https://cloudblue.freshdesk.com/support/solutions/articles/44001881778

    2.https://www.cnblogs.com/f-society/p/13177614.html

    本文版权归作者所有,未经作者同意不得转载,谢谢配合!!!

  • 相关阅读:
    CF500F New Year Shopping [线段树分治,背包]
    P5344 【XR-1】逛森林[倍增优化建图,zkw线段树优化spfa]
    CF452F Permutation [哈希,树状数组]
    [NOI Online #2 提高组]子序列问题
    牛客挑战赛39题解
    #6036. 「雅礼集训 2017 Day4」编码 [前缀优化2sat]
    CF1156E Special Segments of Permutation [分治,set]
    #6198. 谢特 [后缀自动机,01trie合并,启发式合并]
    P4246 [SHOI2008]堵塞的交通 [动态图连通性]
    CF1096G Lucky Tickets [NTT,多项式快速幂]
  • 原文地址:https://www.cnblogs.com/xuliuzai/p/15156886.html
Copyright © 2020-2023  润新知