Обсуждение: Addressing: ERROR: could not access status of transaction

PostgreSQL has been providing reliable service for our web hosting
company since 1997. Thanks!

Last night we got the following error during a dump of an 8.0.6
database:

  pg_dump: SQL command failed
  pg_dump: Error message from server: ERROR:  could not access status of
transaction 245900066
  DETAIL:  could not open file "/usr/local/pgsql/data/pg_clog/00EA": No
such file or directory
  pg_dump: The command was: SELECT tableoid, oid, oprname, oprnamespace,
(select usename from pg_user where oprowner = usesysid) as usename,
oprcode::oid as oprcode FROM pg_operator

Another dump run during the same time frame did not have this problem,
and running the mentioned command in 'psql' produces no error.

Research shows that this could be a corrupted tuple, but it's not
clear what it means to me if it sometimes work.

This follows behavior in the past few days where we noticed PostgreSQL
core dumping repeatedly, with this error:

   PANIC:  right sibling's left-link doesn't match

The core dumps stopped when we REINDEX'ed the table mentioned in the
PANIC statement.

We were already planning to upgrade to 8.1.x Real Soon.

Are their further insights about this new error? And is my expectation
correct that a full/dump restore into 8.1.x would address it?

Thanks!

     Mark

  . . . . . . . . . . . . . . . . . . . . . . . . . . .
    Mark Stosberg            Principal Developer
    mark@summersault.com     Summersault, LLC
    765-939-9301 ext 202     database driven websites
  . . . . . http://www.summersault.com/ . . . . . . . .

Mark Stosberg <mark@summersault.com> writes:
> Last night we got the following error during a dump of an 8.0.6
> database:
>   pg_dump: Error message from server: ERROR:  could not access status of
> transaction 245900066
> Another dump run during the same time frame did not have this problem,
> and running the mentioned command in 'psql' produces no error.

> Research shows that this could be a corrupted tuple, but it's not
> clear what it means to me if it sometimes work.

Transient errors of this kind sound like flaky hardware to me.  Might
be time to run some memory and disk diagnostics ...

> This follows behavior in the past few days where we noticed PostgreSQL
> core dumping repeatedly, with this error:
>    PANIC:  right sibling's left-link doesn't match
> The core dumps stopped when we REINDEX'ed the table mentioned in the
> PANIC statement.

That too could have been from a hardware fault during a previous index
update.  (There's a known bug in early 8.1.x releases that had this
symptom, but not in 8.0.x IIRC.)

            regards, tom lane