Обсуждение: pg_clog troubles

Hi folks,

I'm dead in the water with pg_clog errors:

Mar 19 18:25:05 nexus postgres[28736]: [6] FATAL 2:  open of
/data00/pgdata/pg_clog/007D failed: No such file or directory
Mar 19 18:25:06 nexus postgres[22250]: [1] DEBUG:  server process (pid
28736) exited with exit code 2

Mar 19 22:06:53 nexus postgres[29389]: [9] FATAL 2:  open of
/data00/pgdata/pg_clog/0414 failed: No such file or directory
Mar 19 22:06:53 nexus postgres[22250]: [4] DEBUG:  server process (pid
29389) exited with exit code 2

Mar 19 22:11:04 nexus postgres[29491]: [12] FATAL 2:  open of
/data00/pgdata/pg_clog/0353 failed: No such file or directory
Mar 19 22:11:04 nexus postgres[22250]: [7] DEBUG:  server process (pid
29491) exited with exit code 2

Mar 19 22:13:34 nexus postgres[29716]: [6] FATAL 2:  open of
/data00/pgdata/pg_clog/0353 failed: No such file or directory
Mar 19 22:13:34 nexus postgres[29700]: [1] DEBUG:  server process (pid
29716) exited with exit code 2

There aren't any other error messages, just DEBUG messages about how
PostgreSQL is cleaning up after itself.

These occur running a complex-ish query, but no more complex than
others,
and less complex than many.  The query runs for about 20 seconds, then
the error occurs.

Between attempts, I shut down postgresql and ran 'ipcclean'.  No joy
there.

Here's what's in my pg_clog directory:

[postgres@nexus postgres]$ l /data00/pgdata/pg_clog/
total 8
-rw-------    1 postgres postgres     8192 Mar 19 22:05 0000

I'm running version 7.2, with Tom's vacuum analyze patch.

I searched the postgresql.org website for 'pg_clog', came up dry.  Found
some stuff in the newsgroup archives that looked similar but different
(more errors than I'm getting), and that I didn't really understand.

Any help appreciated, as always...

Gordon.
-- 
"Far and away the best prize that life has to offer is the chance to work hard at work worth doing."      -- Theodore
Roosevelt

Gordon Runkle <gar@integrated-dynamics.com> writes:
> I'm dead in the water with pg_clog errors:

> Mar 19 18:25:05 nexus postgres[28736]: [6] FATAL 2:  open of
> /data00/pgdata/pg_clog/007D failed: No such file or directory

Am I right in guessing that the range of file names you have present
in pgdata/clog/ is nowhere near 007D, let alone the others mentioned?

We have seen a couple of reports like this now, all seemingly indicating
that the clog code is being asked about the commit status of a totally-
off-the-wall transaction number.  I would really like to dig into an
example with a debugger ... but the prior reporters have been unable
to reproduce the problem on demand.  If you can provide a chance to
get this behavior under a gdb microscope, let's talk.
        regards, tom lane