On Fri, 15 Jul 2011 19:07:45 +0800, Tony Wang wrote:
No. It's depend on pooler, application server and transaction manager, for example there are possibilities to return connection which is not associated with transaction manager, so You still need to manually commit or rollback at the end of business logic. You may return C1 to poll, and I believe Your application makes this, but transaction may be uncommited. Watch your query log if You have COMMIT or ROLLBACK there, You may as well add tracking of connection id to associate query flow per connection; or check If you have auto commit turned on.On Fri, Jul 15, 2011 at 18:50, Radosław Smogura wrote:On Fri, 15 Jul 2011 18:36:19 +0800, Tony Wang wrote:Weird that I receive your each message twice. On Fri, Jul 15, 2011 at 15:33, Radoslaw Smogura wrote:Simple and obvious question right now do You call commit after transaction? If yes do you use any query or connection pooler?Yes. connection pool is used as application level, not db level. no commit after transaction is possible (Im trying to check the logic), I just cannot imagine it happened for so many users at the same time, and then calmed down for long time, and came again. I found the query I used to log locks would miss locks that relname is null. will add that, though no idea why its null------------------------ Regards, Radoslaw Smogura (mobile) ------------------------- From: Tony Wang Sent: 15 lipca 2011 03:51 To: Scott Marlowe Cc: PostgreSQL Subject: Re: Weird problem that enormous locks On Fri, Jul 15, 2011 at 08:22, Scott Marlowe wrote:On Thu, Jul 14, 2011 at 6:01 PM, Tony Wang wrote:On Fri, Jul 15, 2011 at 01:13, Scott Marlowe> wrote:On Wed, Jul 13, 2011 at 9:47 PM, Tony Wang wrote:; On Thu, Jul 14, 2011 at 10:35, John R Pierce>> > wrote:#ccc solid;padding-left:1ex"> > Its a game server, and the queries are updating users money, as class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px#ccc solid;padding-left:1ex"> > normal. > The sql is like "UPDATE player SET money = money + 100 whereblockquote> id = >> > 12345". the indexes. 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> > The > weird thing is there was another ExclusiveLoockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">uot;player" got two locks, one RowExclusiveLock and one ExclusiveLock. kquote> acquired on c solid;padding-left:1ex"> > user > tables by any PostgreSQL command." You need to figure out what part of your app, or maybe a rogue >> developer etorder-left:1px #ccc solid;padding-left:1ex"> Yeah, thats what Im trying to do Cool. In your first post you said:"gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> select pg_class.relname, pg_locks.mode, pg_locks.granted, pg_stat_activity.current_query, pg_stat_activity.query_start, pg_stat_activity.xact_start as transaction_start, age(now(),pg_stat_activity.query_start) as query_age, > age(now(),pg_st,pg_locks left outer join pg_class on (pg_locks.relation = pg_class.oid) wheree> pg_locks.pid=pg_stat_activity.procpid and > substr(pg_class.relname,1,3) != pg_ order by query_start; cial thing I can find is that there were a lotExclusiveLock, while its normal the locks are only AccessShareLock and RowExbr> So what did / does current_query say when its happening? If it says you dont have access permission then run that query as root when it happens again.As I said, its normal update like "UPDATE player SET money = money + 100 WHERE id=12345", but there are quite manyLinks: ------ [1] mailto:wwwjfy@xxxxxxxxx [2] [2] mailto:scott.marlowe@xxxxxxxxx [3] [3] mailto:wwwjfy@xxxxxxxxx [4] [4] mailto:pierce@xxxxxxxxxxxx [5] [5] http://www.postgresql.org/docs/8.4/static/explicit-locking.html [6] [6] mailto:scott.marlowe@xxxxxxxxx [7] [7] mailto:rsmogura@xxxxxxxxxxxxxxx [8]Actually I dont know what pool You use (I think PHP - I dont know much about this), but I imagine following, If You dont use auto commit or commit: 1. User A updates moneys, gets connections C1, locks his row, no commit 2. User A updates moneys again, gets connection C2, but C1 still holds lock. Regards, Radosław SmoguraAny connection pool behaves similarly. The connection C1 surely will be committed and returned after the operation finished. Having saidthat, the ONLY possible reason is some transactions hanged holding thelocks, and cause others cannot work any more, and the "ExclusiveLock" is not a problem, right?The interesting thing is, I didnt find any timeout/exception after the"lock" period ended in postgresql log, only long query time.
Regards, Radosław Smogura -- Sent via pgsql-general mailing list (pgsql-general@xxxxxxxxxxxxxx) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-general
