Re: autovacuum

From: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
To: Christopher Browne <cbbrowne(at)acm(dot)org>
Cc: pgsql-admin(at)postgresql(dot)org
Subject: Re: autovacuum
Date: 2006-02-04 19:12:07
Message-ID: 20656.1139080327@sss.pgh.pa.us
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-admin

Christopher Browne <cbbrowne(at)acm(dot)org> writes:
>> does autovacuum locks tables while vacuuming?

> Of course it does; any request to access a relation will issue one or
> more locks on the relation.

This is correct in general ...

> VACUUM issues an AccessShareLock request against each relation that is
> vacuumed, which is probably nearly the same lock request your
> applications will be requesting, save for the fact that they'll also
> be submitting some RowExclusiveLock requests for individual rows of
> relations.

... but wrong in detail. Actually VACUUM takes ShareUpdateExclusive
lock, which is a bit stronger than an ordinary reader's AccessShare
lock (it does not block ordinary reads or updates, but it does block
index creation as well as other attempts to VACUUM the table). Also,
RowExclusiveLock is a table-level lock. Data-modifying commands such
as UPDATE take RowExclusiveLock on the whole table as a means of letting
other sessions know that someone is changing the table. They also
take row-level locks on the specific rows they change.

This is all discussed here:
http://www.postgresql.org/docs/8.1/static/explicit-locking.html

regards, tom lane

In response to

Browse pgsql-admin by date

  From Date Subject
Next Message Mohan Anon 2006-02-04 22:53:17 Postgres 8.1.x and MIT Kerberos 5
Previous Message Christopher Browne 2006-02-04 14:54:30 Re: autovacuum