On Tue, 2014-01-28 at 03:30 +0100, Michelle Sullivan wrote: > Ok so temporarily I set min_duration to 0 and logged everything, the log > is here:
The situation can be made much clearer by limiting to the postgres process IDs which are responding to queries from RT, namely 15537 and 15481. The longest statement duration is 1559.273ms, which is a _parse_ of a statement query, not even an execution. The next-longest query is DBD::Pg's query (which is only run once per process, FTR) at 02:11:09 to determine table information -- which takes nearly a full _second_ to return. On the largest Postgres installation which I have access to, this takes less than 100ms. I am not familiar with Bucardo as a replication strategy, so I can't comment on that. However, I strongly suspect that the answer lies amongst your replication tool and network topology. The symptoms do not match any failure modes that I'm familiar with in RT. You may wish to try simplifying your database configuration (temporarily install Pg on one of the front-ends) to see if that resolves any of the problems. - Alex
