> -----Original Message-----
> From: Oleg Bartunov [mailto:oleg@sai.msu.su]
> Sent: Saturday, October 16, 1999 11:27 PM
> To: Hiroshi Inoue
> Cc: hackers@postgreSQL.org
> Subject: RE: [HACKERS] vacuum of permanently updating database
>
>
> Hiroshi,
>
> thank you for the message. I'll try current tree but if
> it's a bug (probable ?) why don't try to fix it for 6.5.3 ?
>
Yes it's a bug.
But as for the 1st bug,it requires a lot of changes to fix.
Seems Bruce and Tom have thought that it's dangerous
to apply them to REL6_5.
As for the 2nd bug,it is fixed easily by the following patch.
If there's no objection,I would commit it into REL6_5.
[snip]
> >
> > Could you try the current tree ?
> >
> > As far as I see,there are 2 possibilities.
> >
> > 1. Relation cache invalidation mechanism is much improved
> > by Tom in the current tree.
> > In your case,index tuples may be inserted into invalid index
> > relation and vanish.
> >
> > 2. If vacuum aborts after the internal commit,the transaction
> > status is changed to be ABORT. This causes inconsistency.
> > I have changed not to do so in the current tree.
> >
Regards.
Hiroshi Inoue
Inoue@tpf.co.jp
*** xact.c 1999/09/10 07:57:08 1.4
--- xact.c 1999/09/10 08:25:15
***************
*** 736,742 **** * this transaction id in the pg_log relation. We skip it * if no one shared buffer was
changedby this transaction. */
! if (SharedBufferChanged) TransactionIdAbort(xid); ResetBufferPool();
--- 736,742 ---- * this transaction id in the pg_log relation. We skip it * if no one shared buffer was
changedby this transaction. */
! if (SharedBufferChanged && !TransactionIdDidCommit(xid)) TransactionIdAbort(xid);
ResetBufferPool();
