On 09/17/2015 11:15 PM, Guillaume Lelarge wrote:
> Le 18 sept. 2015 5:23 AM, "Adrian Klaver" <adrian.klaver@aklaver.com
> <mailto:adrian.klaver@aklaver.com>> a écrit :
> >
> > On 09/17/2015 05:37 PM, Michael Chau wrote:
> >>
> >> 1)
> >>
> >> In Production, I have a DB2 which is replicated partially using
> Londiste
> >> from DB1.
> >>
> >>
> >> Well I think the above needs more explanation to help understand how the
> >> DB2 backup got into this state and possibly prevent it in the future.
> >>
> >> A: So, the DB1 has several schemas in the database. We use Londiste to
> >> replicate just one of the schemas to DB2. The table in question is in
> >> that schema. Backup is done on both DB1 and DB2.
> >>
> >> 2)
> >> I make file-system backups nightly on both DBs.
> >>
> >> How is that done exactly?
> >>
> >> A: To backup:
> >>
> >> 1) pg_start_backup()
> >> 2) tar up the files under the data directory
> >> 3) pg_stop_backup()
> >>
> >> To restore on test server:
> >>
> >> 1) Just untar the tar ball, then start up Postgres. Of course the data
> >> directory is empty beforehand.
> >>
> >> This has been working for almost 2 years without any problem until last
> >> Monday. I remember that I just ran vacuum analyze that table on both DB1
> >> and DB2 that morning. But, I don't think that it harms anything.
> >
> >
> > Well it looks fairly straight forward, to me at least.
> >
>
> Do I miss something obvious? Because this is to me the wrong way to do
> the restore. You need to apply WAL files archived between
> pg_start_backup and pg_stop_backup to get consistent data files.
Would that not be taken care of by the tar data directory/ untar data
directory?
I would think if it was a WAL issue the OP could never get the server to
start and get to the point the query failed on a single table and
column. All that being said, I think the OP would be better served by
pg_basebackup:
http://www.postgresql.org/docs/9.4/static/app-pgbasebackup.html
>
--
Adrian Klaver
adrian.klaver@aklaver.com
