New VACUUM FULL
in [PgSql]
|
From: Peter Eisentraut on 6 Jan 2010 04:59 On ons, 2010-01-06 at 14:41 +0900, Takahiro Itagaki wrote: > Simon Riggs <simon(a)2ndQuadrant.com> wrote: > > > > 1. Commit your patch, as-is (you/me) > > > > I assume this is OK with you now? > > I just applied the patch with a few additional comments. Please clean up the compiler warnings: cluster.c: In function 'cluster_rel': cluster.c:789: warning: 'heapScan' may be used uninitialized in this function cluster.c:789: note: 'heapScan' was declared here cluster.c:788: warning: 'indexScan' may be used uninitialized in this function cluster.c:788: note: 'indexScan' was declared here -- Sent via pgsql-hackers mailing list (pgsql-hackers(a)postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers
From: Simon Riggs on 30 Jan 2010 09:05 On Wed, 2010-01-06 at 14:41 +0900, Takahiro Itagaki wrote: > I just applied the patch with a few additional comments. I just realised that this new feature *removes* any clustering that was previously defined on a table. Many people would see that as a bug and would say that VACUUM FULL should retain the existing clustering, if any exists. Can't remember if that was discussed already? ISTM that it will be slower if we do that, so it should be either an option or just documented as the new behaviour. -- Simon Riggs www.2ndQuadrant.com -- Sent via pgsql-hackers mailing list (pgsql-hackers(a)postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers
From: Heikki Linnakangas on 30 Jan 2010 09:38
Simon Riggs wrote: > On Wed, 2010-01-06 at 14:41 +0900, Takahiro Itagaki wrote: > >> I just applied the patch with a few additional comments. > > I just realised that this new feature *removes* any clustering that was > previously defined on a table. Hmm, that's an overstatement. If the table was in order before, it will be in the same order after VACUUM FULL, all empty gaps and dead tuples are just removed. It also doesn't clear the indisclustered field in pg_index, so the next time you run CLUSTER it will cluster the table just fine. I'm guessing that you mean that VACUUM FULL doesn't reorder the table like CLUSTER does. I think that's OK, it has never done that. In fact the current situation is already an improvement, the new VACUUM FULL doesn't reshuffle the table and destroy old ordering like the old one does. -- Heikki Linnakangas EnterpriseDB http://www.enterprisedb.com -- Sent via pgsql-hackers mailing list (pgsql-hackers(a)postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers |