| From: | Bruce Momjian <bruce(at)momjian(dot)us> |
|---|---|
| To: | Alexey Bashtanov <bashtanov(at)imap(dot)cc> |
| Cc: | pgsql-hackers(at)postgresql(dot)org |
| Subject: | Re: Autovacuum fails to keep visibility map up-to-date in mostly-insert-only-tables |
| Date: | 2014-10-09 20:58:51 |
| Message-ID: | 20141009205851.GC19877@momjian.us |
| Views: | Whole Thread | Raw Message | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
On Thu, Oct 9, 2014 at 02:34:17PM +0400, Alexey Bashtanov wrote:
> Hello!
>
> Autovacuum daemon performs vacuum when the number of rows
> updated/deleted (n_dead_tuples) reaches some threshold.
> Similarly it performs analyze when the number of rows changed in any
> way (incl. inserted).
> When a table is mostly insert-only, its visibility map is not
> updated as vacuum threshold is almost never reached, but analyze
> does not update visibility map.
>
> Why could it be a bad idea to run vacuum after some number of any
> changes including inserts, like analyze?
> Or at least make it tunable by user (add a second bunch of paramters
> to control second vacuum threshold, disabled by default)?
I agree this is a serious problem. We have discussed various options,
but have not decided on anything. The TODO list has:
https://wiki.postgresql.org/wiki/Todo
Improve setting of visibility map bits for read-only and insert-only
workloads
http://www.postgresql.org/message-id/20130906001437.GA29264@momjian.us
--
Bruce Momjian <bruce(at)momjian(dot)us> http://momjian.us
EnterpriseDB http://enterprisedb.com
+ Everyone has their own god. +
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Robert Haas | 2014-10-09 21:02:02 | Re: Deferring some AtStart* allocations? |
| Previous Message | Bruce Momjian | 2014-10-09 20:46:11 | Re: pg_upgrade, locale and encoding |