Re: Process locked in INSERT: how to debug? - Mailing list pgsql-general

From Csaba Nagy
Subject Re: Process locked in INSERT: how to debug?
Date
Msg-id 1093870064.3765.11.camel@coppola.ecircle.de
Whole thread Raw
In response to Process locked in INSERT: how to debug?  ("Philippe Lang" <philippe.lang@attiksystem.ch>)
List pgsql-general
Hi Philippe,

I can't answer your actual question, but just a hint: postgres is
deadlock-prone on inserts into tables referencing other tables via
foreign keys. This is a known limitation of postgres, it places an
exclusive lock on the referenced rows. For details search the archives
for "foreign key deadlock", there's plenty of discussions about it.
So my personal suspect nr. 1. would be the foreign keys.
If that's the problem, you can solve it in a number of ways:
 - serialize your inserts, as the problem appears only for concurrent
inserts - this is probably the least viable option;
 - make sure you order your inserts so that they insert the rows in the
same order of referenced parent rows (damn hard to do if you have many
foreign keys);
 - be prepared to retry each transaction involving the inserts (this is
actually the best way to do it with postgres, at least in pre-8.0
versions, as it also solves the problem of "one failed SQL rolls back
the whole transaction");
 - if you use 8.0 then you could use the new save point feature to retry
just the last failed insert, but I have no idea if that's not bundled
with some performance penalties (actually I didn't try at all 8.0, so I
just suppose it would work);
 - use a patched postgres version which does not have the foreign key
deadlock problem, but it also do not have 100% referential integrity
guarrantied for foreign key constraints. I think you can find the patch
somewhere referenced in a past mail;

HTH,
Csaba.


On Mon, 2004-08-30 at 14:26, Philippe Lang wrote:
> Hello,
>
> What is the best method to debug a database that *sometimes* gets locked
> in the INSERT mode? ('ps -afx | grep postgres' shows me a process that
> is freezed in INSERT mode.)
>
> Thanks
>
> Philippe
>
> ---------------------------(end of broadcast)---------------------------
> TIP 3: if posting/reading through Usenet, please send an appropriate
>       subscribe-nomail command to majordomo@postgresql.org so that your
>       message can get through to the mailing list cleanly


pgsql-general by date:

Previous
From: "Philippe Lang"
Date:
Subject: Process locked in INSERT: how to debug?
Next
From: Bruno Wolff III
Date:
Subject: Re: Single Row Table?