Postgresql is showing very high costs for some queries that should otherwise be very quick. We are tweaking and it could be up and down for a while.
toshio and devrimgunduz took care of this, it was a lot of work for us all but I think we've done a lot of good things.
There are only a few things left to do here: 1. Analyze the usage patterns of the database tables to vacuum the tables. Tables that are more frequently updated and deleted from need to be vacuumed more frequently. 2. Set up a cron job to vacuum our databases according to the above analysis. 3. Once 2 is done, arrange an extended maintenance outage to perform a vacuum full on all of our databases. [*]
I'm currently testing a script that does #1.
[*] Note that even with the former, non-optimal settings only koji's tables took longer than a couple hours to perform vacuum full.
This is done and has been in productions for a while. See the vacstats script in the fedora-infrastructure repo.
Login to comment on this ticket.