We figured that out. We were afraid that there was corrupt data due the counter overflow so we reverted to a backup right before the overflow. We submitted a bug this morning (bug #12990) to the Postgres bugs mailing list which has more information and questions.
Pweaver (Paul Weaver) wrote: > We started getting the following error on some transactions on our database > (when against particular rows run against the table). > > > PGError: ERROR: could not access status of transaction 283479860 DETAIL: > Could not open file "pg_multixact/members/4D6D": No such file or directory. > > We don't know what happen or how to recover. Any advise?
This looks like you overran pg_multixact/members files by wrapping around the 32 bit counter (NextMultiOffset in your pg_controldata output); a cleanup process later removed files that were still needed. This is a Postgres bug, I fear. Recovery is a bit complicated. I see that your report was made 5 days ago -- have you had any progress on getting things fixed since then? -- Álvaro Herrera http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services