On Fri, Apr 12, 2013 at 7:41 AM, Tom Lane <t...@sss.pgh.pa.us> wrote: > dan...@heroku.com writes: >> We have a somewhat high-churn table acting as a queue, and over time it's >> grown to be something like a gigabyte. I surmised it might be vanilla >> bloat, but the truth seems somewhat more exotic because both VACUUM FULL and >> CLUSTER generated absolutely no new free space. > >> In the end, ALTER TABLE and CREATE TABLE ... (LIKE) ran nearly instantly and >> got the table size down to a few hundred K from 900M. > > My money is on there being old idle transactions somewhere that kept > recently-dead rows from being reclaimable. If memory serves, VACUUM > FULL and CLUSTER will faithfully retain such rows, but of course a > manual data transfer like that wouldn't. > >> We have retained the old bloated table so we can poke at it. > > I think contrib/pgstattuple could tell you about dead tuples.
Yeah, you and Andres are on the mark, although the cause is potentially a bit less visible: hot standby feedback. Also, my misunderstanding of VACUUM FULL/CLUSTER's interaction with snapshots, which I thought more similar to DDL for no reason in particular. Sorry about the noise. -- Sent via pgsql-bugs mailing list (pgsql-bugs@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-bugs
