在 2014/7/30 1:27, Jeff Janes 写道:
It sounds like someone is bypassing your pgbouncer and connecting
directly to your database. Maybe they tried to create their own
parallelization and have a master connection going through pgbouncer
and create many auxiliary connections that go directly to the database
(probably because pgbouncer wouldn't let them create as many
connections as they wanted through it). That would explain why the
connections slowly drain away once pgbouncer is shut down.
Can you change your pg_hba.conf file so that it only allows
connections from pgbouncer's IP address? This should flush out the
culprit pretty quickly.
Cheers,
Jeff
I suspected that first. But after I checked a few things, I am quite
sure this is not someone bypassing the pgbouncer.
1. The connections were all from the host of pgbouncer.
2. The id is an application id and no human has access to it. There was
no other suspect applications running on the host of pgbouncer when the
problem happened.
3. When I found the problem and checked the connections on the host of
pgbouncer, those network connection actually didn't exist on the client
side while they were still hanging at pg server side.
--
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance
