Thread: small changes to autovacuum maintainance

Attached patch takes a swing at improving the autovacuum wording in
maintenance.sgml.

On a related note I was thinking to add a note in the runtime config section
for autovacuum_enabled that stats_start_collector and stats_row_level must be
on for enabled to work, would that still go in runtime.sgml ?

On a related note to that, I am wondering if it might be better to have the
postmaster issue a fatal error on startup if autovacuum is on and stats are
off, rather than the warning it does now. ISTM it is very easy to miss that
warning (since it is only sent to stderr and not stout afaict) and looking in
pg_settings there would be no indication that anything was wrong, other than
autovacuum being set to off (perhaps if it was set to 'error, stats disabled'
or some such it might be better?)

--
Robert Treat
Build A Brighter Lamp :: Linux Apache {middleware} PostgreSQL

Robert Treat <xzilla@users.sourceforge.net> writes:
> On a related note to that, I am wondering if it might be better to have the
> postmaster issue a fatal error on startup if autovacuum is on and stats are
> off, rather than the warning it does now.

Given that you can turn off the stats after startup, the above seems
fairly useless.

If we want to be nanny-ish about this, we should implement an interlock
that forces the stats settings on permanently if autovac is enabled.
(However, past experience with trying to force interrelationships of GUC
variables hasn't been too positive...)

            regards, tom lane

Robert Treat wrote:
> Attached patch takes a swing at improving the autovacuum wording in
> maintenance.sgml.

Thanks, applied.

> On a related note I was thinking to add a note in the runtime config section
> for autovacuum_enabled that stats_start_collector and stats_row_level must be
> on for enabled to work, would that still go in runtime.sgml ?

Yes, that is config.sgml now.  Added.

> On a related note to that, I am wondering if it might be better to have the
> postmaster issue a fatal error on startup if autovacuum is on and stats are
> off, rather than the warning it does now. ISTM it is very easy to miss that
> warning (since it is only sent to stderr and not stout afaict) and looking in
> pg_settings there would be no indication that anything was wrong, other than
> autovacuum being set to off (perhaps if it was set to 'error, stats disabled'
> or some such it might be better?)

I don't think we usually fail on startup unless there is a major
problem.  When people modify postgresql.conf, they better check the
server log file after startup.

--
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073