There are a few statistics that are only kept on a per-cluster basis, such 
as those reported in pg_stat_database and pg_stat_bgwriter.  Since 
pg_stat_reset only resets per-database counts, it doesn't touch those.

I'd like to have a function that resets those.  Maybe iterates over all 
the databases while it's at it and clear the whole mess out as another 
option.  I'm curious why something resetting the cluster-wide stats isn't 
available already.  Just because nobody bothered to write it before, or is 
there some reason why this can't work that I just didn't notice during my 
brief scan of pgstat.c?

--
* Greg Smith gsmith(at)gregsmith(dot)com http://www.gregsmith.com Baltimore, MD

Greg Smith <gsmith(at)gregsmith(dot)com> writes:
> I'm curious why something resetting the cluster-wide stats isn't 
> available already.

We used to have the stats_reset_on_server_start thingy, which covered
that case.  When we removed it on the grounds that you could do the
same less klugily with pg_stat_reset(), nobody complained about
cluster-wide stats...

			regards, tom lane

On Mon, 23 Mar 2009, Tom Lane wrote:

> We used to have the stats_reset_on_server_start thingy, which covered
> that case.  When we removed it on the grounds that you could do the
> same less klugily with pg_stat_reset(), nobody complained about
> cluster-wide stats...

Not exactly sure of the sequence here, but I think that was before 
pg_stat_bgwriter was introduced.  That was late in the 8.3 cycle and one 
of the primary things I'd like have a good way to reset.  I'll dig back to 
the commit that wiped out stats_reset_on_server_stat and see what it used 
to do.  I might suggest something that puts something similar back, 
perhaps with a more targeted UI given that pg_stat_reset has taken over 
its original primary function.

--
* Greg Smith gsmith(at)gregsmith(dot)com http://www.gregsmith.com Baltimore, MD

Greg Smith wrote:
> On Mon, 23 Mar 2009, Tom Lane wrote:
> 
> > We used to have the stats_reset_on_server_start thingy, which covered
> > that case.  When we removed it on the grounds that you could do the
> > same less klugily with pg_stat_reset(), nobody complained about
> > cluster-wide stats...
> 
> Not exactly sure of the sequence here, but I think that was before 
> pg_stat_bgwriter was introduced.  That was late in the 8.3 cycle and one 
> of the primary things I'd like have a good way to reset.  I'll dig back to 
> the commit that wiped out stats_reset_on_server_stat and see what it used 
> to do.  I might suggest something that puts something similar back, 
> perhaps with a more targeted UI given that pg_stat_reset has taken over 
> its original primary function.

Added to TODO:

	Allow the clearing of cluster-level statistics
	
	    * http://archives.postgresql.org/pgsql-hackers/2009-03/msg00917.php 

-- 
  Bruce Momjian  <bruce(at)momjian(dot)us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

  + If your life is a hard drive, Christ can be your backup. +