From: | Merlin Moncure <mmoncure(at)gmail(dot)com> |
---|---|
To: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
Cc: | Tim Bellis <Tim(dot)Bellis(at)metaswitch(dot)com>, Adrian Klaver <adrian(dot)klaver(at)aklaver(dot)com>, "pgsql-general(at)postgresql(dot)org" <pgsql-general(at)postgresql(dot)org>, Alvaro Herrera <alvherre(at)2ndquadrant(dot)com>, Scott Marlowe <scott(dot)marlowe(at)gmail(dot)com> |
Subject: | Re: Autovacuum stuck for hours, blocking queries |
Date: | 2017-02-21 01:40:01 |
Message-ID: | CAHyXU0x5xqAqgBQw+D1iU3DEr0H8+k5tX4LQO3THzT0C3QKiQw@mail.gmail.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-general |
On Thursday, February 16, 2017, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Tim Bellis <Tim(dot)Bellis(at)metaswitch(dot)com <javascript:;>> writes:
> > Even though this is a read only query, is it also expected to be blocked
> behind the vacuum? Is there a way of getting indexes for a table which
> won't be blocked behind a vacuum?
>
> It's not the vacuum that's blocking your read-only queries. It's the
> ALTER TABLE, which needs an exclusive lock in order to alter the table's
> schema. The ALTER is queued waiting for the vacuum to finish, and lesser
> lock requests queue up behind it. We could let the non-exclusive lock
> requests go ahead of the ALTER, but that would create a severe risk of the
> ALTER *never* getting to run.
>
> I'd kill the ALTER and figure on trying again after the vacuum is done.
>
>
I've been drilled by this and similar lock stacking issues enough times to
make me near 100% sure deferring the ALTER would be the better choice
merlin
From | Date | Subject | |
---|---|---|---|
Next Message | Sushant Pawar | 2017-02-21 03:59:42 | Re: Search on very big (partitioned) table |
Previous Message | Venkata B Nagothi | 2017-02-20 23:49:17 | Re: Streaming Replication Without Downtime |