Thread: 'Too many LWLocks held' with GIN pg_trgm_ops index [RE-SENT]

'Too many LWLocks held' with GIN pg_trgm_ops index [RE-SENT]

From
Craig Ringer
Date:
[RESENT MESSAGE - the previous one was presumably held for moderation
because of an attachment. If the prior one did in fact make it through
and I just didn't get it then sorry for the noise.]

Hi all

Following my -general post about this issue I'm supplying a test case
for a query that fails with a 'Too many LWLocks held' error.

The test data, system configuration details, etc are in the tarball that
you can obtain here:

http://www.postnewspapers.com.au/~craig/lwlock_test.tar.bz2

The README in the package provides all the detail I can think of, but
here's the summary again:

------
This is a test case for a `Too many LWLocks held' error encountered when
using pg_trgm with a GIN index in the following query:

SELECT a.message, b.message FROM po_message a, po_message b
WHERE a.id <> b.id AND a.message % b.message;

The error does not occur if a GIST index is present on
`po_message.message' instead of a GIST index, or if no index is present
at all. It only arises when the % operator is used.
------

--
Craig Ringer

Re: 'Too many LWLocks held' with GIN pg_trgm_ops index [RE-SENT]

From
Teodor Sigaev
Date:
> Following my -general post about this issue I'm supplying a test case
> for a query that fails with a 'Too many LWLocks held' error.
Thank you, I reproduced the bug and now I'm working on it.

--
Teodor Sigaev                                   E-mail: teodor@sigaev.ru
                                                    WWW: http://www.sigaev.ru/