Re: DeadLocks..., DeadLocks... - Mailing list pgsql-general

From Bill Moran
Subject Re: DeadLocks..., DeadLocks...
Date
Msg-id 20070614123059.838a116d.wmoran@potentialtech.com
Whole thread Raw
In response to Re: DeadLocks..., DeadLocks...  (<tom@tacocat.net>)
Responses Re: DeadLocks..., DeadLocks...  (Gregory Stark <stark@enterprisedb.com>)
List pgsql-general
In response to <tom@tacocat.net>:
>
> On 6/14/2007, "Gregory Stark" <stark@enterprisedb.com> wrote:
>
> >
> >
> ><tom@tacocat.net> writes:
> >
> >> But everyone once in a long while it seems that I hit simultaneaous
> >> execute() statements that deadlock on the insertion.
> >
> >What version of Postgres is this and do you have any foreign key constraints
> >or triggers on the table you're inserting into?
>
> Version 8.2
> This table does not have foreign key constraints on it, but it is the
> source of foreign key constraints on other tables.
> No triggers.
>
>  Is that insert the *only* DML
> >you're executing? No updates or deletes?
>
> At the time of the failure, no other DML.
> There are other's but they are on different tables.
> >
> >What do you mean by saying it deadlocks? Do you get a transaction abort with
> >an error about a deadlock detected? Or do you just mean it freezes?
>
> "deadlock detected"
> And the corresponding error I get is a primary key violation on the same
> table.
>
>
> The problem occurs when I have multiple processes acting on what appears
> to be the exact same set of information.  I can't really control the
> issue of simultaneous/parallel processing

Put an "ORDER BY" in your SELECT.

I believe the problem is that when this runs from two different places,
the DB may order the returned values in a different order for each one,
which leads to the possibility of two similar inserts deadlocking.  Unless
I misunderstand your schema, you should be able to guarantee against
deadlocking by guaranteeing that the SELECT portion will always return
rows in the same order.

--
Bill Moran
http://www.potentialtech.com

pgsql-general by date:

Previous
From: Greg Smith
Date:
Subject: Re: pg_xlog - files are guaranteed to be sequentialy named?
Next
From: Rikard Pavelic
Date:
Subject: explain analyze on a function