Skip site navigation (1) Skip section navigation (2)

Re: deadlock problems with foreign keys

From: Jan Wieck <janwieck(at)yahoo(dot)com>
To: Rod Taylor <rbt(at)zort(dot)ca>
Cc: Mario Weilguni <mario(dot)weilguni(at)icomedias(dot)com>,Hackers List <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: deadlock problems with foreign keys
Date: 2002-03-28 18:27:52
Message-ID: 200203281827.g2SIRqp28595@saturn.janwieck.net (view raw or flat)
Thread:
Lists: pgsql-hackers
Rod Taylor wrote:
> There was no deadlock in 7.2 with what was provided -- but the second
> transaction was blocked from doing it's thing by the lock from the
> first.  Perhaps a deadlock is caused by 'do other stuff'?
>
> I will agree that a FOR UPDATE is heavy.  There is no intention to
> update the record, we just want to ensure it's NOT updated or deleted.
> A FOR PREVENT UPDATE lock may be preferable and it should block any
> other locks while allowing the lock to be 'upgraded' in the case where
> you hold the only PREVENT UPDATE lock.  It wouldn't be exclusive to
> itself, only other types of locks.
>
>
> All that said, SET CONSTRAINTS ALL DEFERRED at the beginning of the
> transaction also caused a block on the update with the second
> transaction.  That interests me.  Why doesn't the second transaction
> go through and block the first from using COMMIT?

    SET   CONSTRAINTS   ALL   DEFERRED   only   set's  DEFERRABLE
    constraints  to  DEFERRED.   Constraints   default   to   NOT
    DEFERRABLE,  so  unless  you  explicitly  allowed it at table
    creation, you did a noop.


Jan

>
>
> --
> Rod Taylor
>
> This message represents the official view of the voices in my head
>
> ----- Original Message -----
> From: "Mario Weilguni" <mario(dot)weilguni(at)icomedias(dot)com>
> To: "Postgresql Mailinglist (E-Mail)" <pgsql-hackers(at)postgresql(dot)org>
> Sent: Thursday, March 28, 2002 9:44 AM
> Subject: [HACKERS] deadlock problems with foreign keys
>
>
> I've a severe problem with deadlocks in postgres, when using
> referential integrity it's quite easy to trigger deadlocks. I think
> the may be a bug in ri_trigger.c (discussed later). Here's some short
> example:
>
> create table languages (
>   id        integer not null,
>   name      text    not null,
>   primary key(id)
> );
>
> create table entry (
>   id        integer not null,
>   lang_id   integer,
>   sometext  text,
>   primary key (id),
>   foreign key ( lang_id ) references languages (id)
> );
>
> insert into languages values (1, 'english');
> insert into languages values (2, 'german');
>
> insert into entry values (1, 1, 'text 1');
> insert into entry values (2, 1, 'text 2');
>
>
> transaction A: begin;
> transaction A: update entry set sometext='text 1.1' where id=1;
> transaction A: .... do more time-consuming processing here...
> meanwhile, B: begin;
>            B: update entry set sometext='text 2.1' where id=2;
>
> -- both processes hang now
>
> I think this is too much locking here, because the logfile show's
> something like this:
> 'select 1 from "languages" where id=$1 for update' (2 times).
>
> Now I've a lot of tables (around 30) and use referential integrity a
> lot on ~10 columns (language, country....) , and with more fields it's
> very easy to deadlock the whole system (it happens a lot in my web
> applicaiton with ~20 concorrent users).
>
> IMHO the "select ... for update" on languages is not necessary, since
> I do not want to update "lang_id", but I might be wrong. The other
> problem is, that this will make postgres in benchmarks very slow (with
> many concurrent connections), at least if the application is not
> trivial.
>
> IMO the problem is in ri_trigger.c around line 390:
> /* ----------
> * The query string built is
> * SELECT 1 FROM ONLY <pktable> WHERE pkatt1 = $1 [AND ...]
> * The type id's for the $ parameters are those of the
> * corresponding FK attributes. Thus, SPI_prepare could
> * eventually fail if the parser cannot identify some way
> * how to compare these two types by '='.
> * ----------
> */
>
> Any ideas if this is a bug or simply strict SQL standard?
>
> Best regards,
> Mario Weilguni
>
>
> ---------------------------(end of
> broadcast)---------------------------
> TIP 5: Have you checked our extensive FAQ?
>
> http://www.postgresql.org/users-lounge/docs/faq.html
>
>
>
> ---------------------------(end of broadcast)---------------------------
> TIP 2: you can get off all lists at once with the unregister command
>     (send "unregister YourEmailAddressHere" to majordomo(at)postgresql(dot)org)
>


--

#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me.                                  #
#================================================== JanWieck(at)Yahoo(dot)com #



_________________________________________________________
Do You Yahoo!?
Get your free @yahoo.com address at http://mail.yahoo.com


In response to

pgsql-hackers by date

Next:From: Peter EisentrautDate: 2002-03-28 23:07:31
Subject: Re: Rough sketch for locale by default
Previous:From: Riebs, AndyDate: 2002-03-28 18:09:44
Subject: Re: Performance Tuning Document?

Privacy Policy | About PostgreSQL
Copyright © 1996-2014 The PostgreSQL Global Development Group