Re: vacuum verbose relations reporting - Mailing list pgsql-admin

From Tom Lane
Subject Re: vacuum verbose relations reporting
Date
Msg-id 11312.1219794535@sss.pgh.pa.us
Whole thread Raw
In response to Re: vacuum verbose relations reporting  (Jeff Frost <jeff@frostconsultingllc.com>)
Responses Re: vacuum verbose relations reporting  (Jeff Frost <jeff@frostconsultingllc.com>)
List pgsql-admin
Jeff Frost <jeff@frostconsultingllc.com> writes:
> No I actually mean that this part:
> Current limits are:  600000 page slots, 10000 relations, using 4587 kB.
> NOTICE:  max_fsm_relations(10000) equals the number of relations checked
> HINT:  You have at least 10000 relations.  Consider increasing the

> Just tells me that I'm equal or greater than max_fsm_relations.

Yeah, because that's all that can be told from the contents of the
shared free space map: it's full, but we have no idea how many other
tables might have been competing for space in it.

> But vacuum verbose does tell me how many page_slots are in use even
> when it's greater than max_fsm_pages, so you know what value to use in
> postgresql.conf for the fsm settings.

Well, no :-(.  What that number is is the number of page slots that the
relations that are in the FSM would have liked to have --- we have the
"request" size as well as the "allocated" size for each such relation.
We have no idea how many slots the relations that aren't in FSM might
have liked to have.

This whole design is pretty awful, of course (I can say that because it
was my design :-().  There is work in progress to make the fixed-size
FSM go away entirely in 8.4, which will certainly be a boon to DBAs.

            regards, tom lane

pgsql-admin by date:

Previous
From: "Richard Broersma"
Date:
Subject: PITR - base backup question
Next
From:
Date:
Subject: Re: PITR - base backup question