Re: Weird locking situation - Mailing list pgsql-hackers

From Tom Lane
Subject Re: Weird locking situation
Date
Msg-id 18233.1065188370@sss.pgh.pa.us
Whole thread Raw
In response to Re: Weird locking situation  (Christopher Kings-Lynne <chriskl@familyhealth.com.au>)
Responses Re: Weird locking situation
Re: Weird locking situation
List pgsql-hackers
Christopher Kings-Lynne <chriskl@familyhealth.com.au> writes:
> OK, I tried it again and it still seems buggy to me...

> I wonder if it's something to do with the tsearch trigger on food_foods?

I tried a table with a simple BEFORE trigger and it didn't fail.
But when I added a GIST index, it did:

[ install contrib/btree_gist ]
regression=# create index gindex on foo using gist (f2);
CREATE INDEX
...
regression=# UPDATE foo SET f2=now() WHERE f1=1;
ERROR:  deadlock detected
DETAIL:  Process 18122 waits for AccessExclusiveLock on relation 154635 of database 17139; blocked by process 18133.
Process 18133 waits for ShareLock on transaction 6330; blocked by process 18122.

The trouble here is that GIST indexes are not concurrency-safe.
This is on the TODO list but I fear it's not a small task ...
        regards, tom lane


pgsql-hackers by date:

Previous
From: Michael Meskes
Date:
Subject: Re: ecpg doesn't compile (datetime.h/dtime_t)
Next
From: Michael Brusser
Date:
Subject: Using backslash in query