Re: Something Weird Going on with VACUUM ANALYZE

Shaun Thomas <sthomas(at)optionshouse(dot)com> wrote:

> The first line also seems extremely misleading. a manually
> launched vacuum is not an autovacuum, so why is it complaining
> about an autovacuum? Shouldn't the PID doing the autovacuum emit
> the log message? This all seems a little sketchy.

The table truncation in autovacuum had severe problems in some
conditions which were causing production down time.  The fix
accidentally introduced some messages which are confusing, and
caused the statistics from autovacuum to fail to be generated more
often than had previously been the case.  These problems should be
fixed in the next minor release.  Until then, if you have a
workload which triggers these issues, the workaround is to schedule
ANALYZE jobs.

Where things are now running faster, it's hard to say how much is
from missing the analyze step, how much is due to caching issues,
and how much is actual gain from the fixes for the original
problems.

--
Kevin Grittner
EDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company