RE: Disable WAL logging to speed up data loading - Mailing list pgsql-hackers

Hi, Sawada-San


On Monday, December 28, 2020 7:12 PM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
> On Mon, Dec 28, 2020 at 4:29 PM osumi.takamichi@fujitsu.com
> <osumi.takamichi@fujitsu.com> wrote:
> > On Monday, December 28, 2020 2:29 PM Masahiko Sawada
> <sawada.mshk@gmail.com> wrote:
> > > On Thu, Dec 3, 2020 at 12:14 PM osumi.takamichi@fujitsu.com
> > > <osumi.takamichi@fujitsu.com> wrote:
> > > >
> > > > I've made a new patch v05 that took in comments to filter out WALs
> > > > more strictly and addressed some minor fixes that were discussed
> > > > within past few days.
> > > > Also, I changed the documentations, considering those modifications.
> > >
> > > From a backup management tool perspective, how can they detect that
> > > wal_level has been changed to ‘none' (and back to the normal)? IIUC
> > > once we changed wal_level to none, old backups that are taken before
> > > setting to ‘none’ can be used only for restoring the database to the
> > > point before the LSN where setting 'wal_level = none'. The users can
> > > neither restore the database to any points in the term of 'wal_level
> > > = none' nor use an old backup to restore the database to the point
> > > after LSN where setting 'wal_level = none’. I think we might need to
> > > provide a way to detect the changes other than reading
> XLOG_PARAMETER_CHANGE.
> > In the past, we discussed the aspect of backup management tool in [1]
> > and concluded that this should be another patch separated from this
> > thread because to compare the wal_level changes between snapshots
> > applies to wal_level = minimal, too. Please have a look at the "second idea"
> > in the e-mail in the [1] and responses to it.
> >
> 
> Thank you for telling me about the discussion!
> 
> The discussion already started on another thread? I think it might be better to
> consider it in parallel, if not started yet. We can verify beforehand that the
> proposed solutions will really work well together with backup management
> tools. And from the user perspective, I wonder if they would like to see this
> feature in the same release where wal_level = none is introduced. Otherwise,
> the fact that some backup management tools won’t be able to work together
> with wal_level = none will be a big restriction for users. That patch would
> probably not be a blocker of this patch but will facilitate the discussion.
I don't think the new thread is created already.

By the way, I checked documents and user manuals of backup management tools like
pgBackRest, EDB's BART and pg_probackup respectively.
These tools work when wal_level is higher than minimal
because these use physical online backup or wal archiving and thus
they don't need to work together with wal_level=minimal or none.

> the fact that some backup management tools won’t be able to work together
> with wal_level = none will be a big restriction for users.
Accordingly, it turned out that current situation or introducing wal_level=none
doesn't become restriction for users from the backup management tools perspective.
Any comments ?

Best Regards,
    Takamichi Osumi