On Tue, Apr 8, 2008 at 2:49 PM, Zdenek Kotala <[EMAIL PROTECTED]> wrote:
> Mikko Partio napsal(a): > > On Tue, Apr 8, 2008 at 1:45 PM, Zdenek Kotala <[EMAIL PROTECTED]> > > wrote: > > > > Mikko Partio napsal(a): > > > > > > Should I try remove one of the duplicate rows from pg_class? > > > Try it with caution. You should use ctid column to refer to exact row. > > > > > > Try before: > > > > > > select oid, ctid, xmin, xmax, cmin, cmax, relname from pg_class where > > > oid > > > = 294397; > > > > > > If both row are identical or not. > > > > > > > > > > # select oid, ctid, xmin, xmax, cmin, cmax, relname from pg_class where > > oid > > = 294397; > > oid | ctid | xmin | xmax | cmin | cmax | relname > > > > --------+-----------+------------+------------+------+------+-------------- > > 294397 | (1125,36) | 3944654557 | 0 | 35 | 35 | abc > > 294397 | (1124,55) | 3943984380 | 3943984642 | 3 | 3 | abc > > (2 rows) > > > > Which one should I remove? What is the worst case scenario for the > > delete? > > The database is 1,5TB so I would rather not restore it from backups :-) > > > > (1124,55) is deleted one. This row should not be visible. But it seems a > problem with visibility - transaction wraparound. Do you have any warning > message in the log about vacuuming? Or clog corruption when transaction > 3943984642 is marked as rollbacked. > There are no such messages in the logfile, but autovacuum is working hard to prevent xid wraparound (we have an import process which consums huge amounts of xids). # select max(age(relfrozenxid)) from pg_class where relkind = 'r' ; max ------------ 204500712 (1 row) How can I tell if there is clog corruption? Thanks a lot for your help. Regards Mikko