[Spacewalk-list] Postgresql slowness

[John Hodrien]

On Thu, 21 Jul 2011, Silvio.Schanz at elektrobit.com wrote:

> I noticed the same behavior after my postgres migration.  My sync of the
> fedora channel slowed down.
>
> I started a repo-sync yesterday at 14:41 and I got only 300 packages until
> now (8:00).

Kill the spacewalk-repo-sync process, waiting a number of seconds and starting
it back up returned performance for a while.  By doing this I managed to push
all of CentOS6 and updates and spacewalk client in about 6 hours, when I'd not
managed to push CentOS6 base in 24 without intervention.

Chasing performance at all costs, I'd made these changes to the config.
checkpoint_segments was in response to a warning in the log, and increased
work_mem cleared some other warnings about temporary files.  The machine had
plenty of unused memory, so I was being fairly liberal.

work_mem = 100MB                        # min 64kB
synchronous_commit = off                # immediate fsync at commit
wal_buffers = 100MB                     # min 32kB
checkpoint_segments = 10                # in logfile segments, min 1, 16MB each
effective_cache_size = 1024MB

jh