Re: AW: [HACKERS] RE: [GENERAL] Transaction logging

From: Vadim Mikheev <vadim(at)krs(dot)ru>
To: Zeugswetter Andreas IZ5 <Andreas(dot)Zeugswetter(at)telecom(dot)at>
Cc: "'hackers(at)postgresql(dot)org'" <hackers(at)postgresql(dot)org>
Subject: Re: AW: [HACKERS] RE: [GENERAL] Transaction logging
Date: 1999-07-19 03:40:46
Message-ID: 37929E3E.D56894BF@krs.ru
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Zeugswetter Andreas IZ5 wrote:
>
> > 2. all datafiles have to be backed up _before_ log backup
> > due to WAL logic: changes must be written to log before
> > they'll be written to on-disk data pages.
> >
> When I was talking about pg_log, I meant pg_log as it is now.
> As I understand it, it only stores commit/rollback info for each used xtid
> and no other info.

Actually, I would like to implement WAL as it's done in other systems.
There would be no more pg_log with xact statuses as now. But for
the first implementation it's easy to leave pg_log as is (UNDO
is hard to implement). In any case WAL will be _single_ source
about everything - what's changes and what transactions were
commited/aborted. From this point of view pg_log will be just
one of datafiles: on recovery changes (commit/abort statuses)
will be applied to pg_log just like to other datafiles.

> PS: for rollback you need the before image of rows, I would keep
> this in a separate place like Oracle (rollback segment) and Informix
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Oracle places rollback segments in memory to speedup abort/MVCC.
Before images are in WAL and used to restore rollback segments
on recovery.

> (physical log) since this info does not need to go to the
> rollforward tape.

Vadim

In response to

Browse pgsql-hackers by date

  From Date Subject
Next Message Vadim Mikheev 1999-07-19 03:50:41 WAL: parallel logging
Previous Message Bruce Momjian 1999-07-19 02:41:25 Please add to 6.5.1