To: Bryce Nesbitt <bryce2@obviously.com>
cc: pgsql-sql@postgresql.org
Subject: Re: AccessExclusiveLock on CREATE TABLE REFERENCES deadlocks
	(formatted better)
In-reply-to: <4AB90198.2010109@obviously.com> 
References: <4AB90198.2010109@obviously.com>
Comments: In-reply-to Bryce Nesbitt <bryce2@obviously.com>
	message dated "Tue, 22 Sep 2009 09:55:52 -0700"
Date: Tue, 22 Sep 2009 13:35:39 -0400
Message-ID: <13526.1253640939@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Bryce Nesbitt <bryce2@obviously.com> writes:
> 1) Why the AccessExclusiveLock on create table?

It has to install a trigger on the referenced table.  There has been
some discussion that maybe CREATE TRIGGER could take just ExclusiveLock
and not AccessExclusiveLock, but it hasn't been done yet; and I'm not
sure how much that would help you anyway.  It would only help if the
referenced table (contexts) is essentially read-only to the rest of
your workload, else it'll block anyhow.

> 2) Why is the foreign key check a heavy operation, since a new table
> will have zero foreign keys, it can't possibly violate the constraint yet.

It's not a heavy operation in that case.  The problem doubtless is that
it's backed up behind some other transaction that is sitting on a lock
on the contexts table.  And then everything else backs up behind it.

> 3) Other than eliminating dynamic table creation, how can this operation
> be altered?

Get rid of long-running transactions that hold locks on the contexts
table.

			regards, tom lane