这么简单的一条SQL，100个并发时居然会发生死锁，太不可思议了。
发生死锁的SQL
update_smallrange.sql:
-
\setrandom id 1 10000
- update maintb set name = 'aaaaa12345' where id=:id;
服务端日志：
点击(此处)折叠或打开
-
2015-01-16 20:56:44.189 CST >错误: 检测到死锁
-
2015-01-16 20:56:44.189 CST >详细信息: 进程4880等待在事务 4074284上的ShareLock; 由进程4910阻塞.
-
进程4910等待在事务 4080369上的ShareLock; 由进程4880阻塞.
-
进程 4880: update maintb set name = 'aaaaa12345' where id=9692;
-
进程 4910: update maintb set name = 'aaaaa12345' where id=9692;
-
2015-01-16 20:56:44.189 CST >提示: 详细信息请查看服务器日志.
- 2015-01-16 20:56:44.189 CST >语句: update maintb set name = 'aaaaa12345' where id=9692;
但是我的case也不是每次再现，之前测试的时候几乎100%的再现概率，但是过了几天，相同的环境却一次也没再现过。
今天偶然翻了下社区的邮件列表，发现这是个已知的BUG。
http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230
这个BUG报告中的错误消息有两种，其中一种和我的是一样的
死锁错误1：
点击(此处)折叠或打开
-
2014-07-30 09:41:54 PDT PID:4729 XID:25780 ERROR: deadlock detected
-
2014-07-30 09:41:54 PDT PID:4729 XID:25780 DETAIL: Process 4729 waits for ShareLock on transaction 25779; blocked by process 4727.
-
2014-07-30 09:41:54 PDT PID:4729 XID:25780 HINT: See server log for query details.
- 2014-07-30 09:41:54 PDT PID:4729 XID:25780 STATEMENT: UPDATE "z8z6px927zu6qzzbnb5ntgghxg"."access_grants" ag SET last_issued=DEFAULT FROM "z8z6px927zu6qzzbnb5ntgghxg"."oauth_clients" oc WHERE oc.id = ag.client_id AND ag.entity_name = 'user' AND ag.entity_id = 129 AND oc.client_id = '3hp45h9d4f9wwtx7cvpus6rdb4s5kb9f' RETURNING ag.id
死锁错误2：
点击(此处)折叠或打开
-
2014-07-30 09:41:56 PDT PID:4739 XID:25806 ERROR: deadlock detected
-
2014-07-30 09:41:56 PDT PID:4739 XID:25806 DETAIL: Process 4739 waits for ExclusiveLock on tuple (1,98) of relation 16553 of database 16385; blocked by process 4738.
- 2014-07-30 09:41:56 PDT PID:4739 XID:25806 HINT: See server log for query details.
这个BUG在9.3.x(9.3.4和9.3.5)上会存在，9.0和9.1等早期版本没有问题。用我们公司的术语说就是LevelDown了。
点击(此处)折叠或打开
-
I think this is a regression as we only see the behavior under postgres 9.3.x (reproduced locally on 9.3.4 and 9.3.5 in a VMWare VM running Ubuntu 11.04, but also evident in 9.3.3 on Amazon RDS). I am unable to reproduce in the earlier versions I've been able to test against (9.0.something and 9.1.9).
好消息是已经有这个BUG的Patch出来了，相信下次的PG版本发布会解决这个问题。