Re: BUG #5798: Some weird error with pl/pgsql procedure

On Wed, Dec 22, 2010 at 4:53 PM, Maxim Boguk <maxim(dot)boguk(at)gmail(dot)com> wrote:
> About stack backtrace I not sure how to get it (and even worse the
> database is heavily loaded production DB so I not sure creating core
> dump would be good idea).
> Still I would like to receive any help with creating a stack backtrace.

Well, if you can reproduce this problem in a given session, you can
use pg_backend_pid() at the start of the session to get the backend
ID, and then if your build has debugging symbols (or separate
debuginfos you can install), you could attach gdb to it, set a
breakpoint at the lines in heaptuple.c that can throw this error,
reproduce the problem, and then get a backtrace. But if it's
happening randomly I don't have a lot of ideas.

I do wonder if this could be related to using TIDs that don't actually
exist in the table. That might be worth some experimentation.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company