Re: BUG #15638: pg_basebackup with --wal-method=stream incorrectly generates WAL segment created during backup

Thank you for checking. Bianary diffs I had were small and at the tail even
though there were no changes made to database so this makes sense that
these diffs were logically meaningless.

Were you able to confirm that with fetch method slave created with
--wal-method=fetch had record in pg_wal/archive_status directory with .done
extension after running pg_basebackup and no record with
--wal-method=stream? I'm also wondering if PG11 has the same behavior.

On Thu, Feb 21, 2019 at 10:28 AM Maksim Milyutin <milyutinma(at)gmail(dot)com>
wrote:

> On 2/19/19 9:08 PM, Andre Piwoni wrote:
>
> By the way, you can assert this structure, including now explained diff in
> WAL segment taken during backup, for fetch vs stream WAL method by creating
> master and then creating new slave by taking base backup using both methods:
>
> pg_basebackup -h new_slave -D /var/lib/pgsql/10/data -v -P -R -U
> replication --wal-method=stream
> pg_basebackup -h new_slave -D /var/lib/pgsql/10/data -v -P -R -U
> replication --wal-method=fetch
>
>
> I have verified on different versions of PostgreSQL the identity of
> conflict segments received from master node and from pg_basebackup
> --wal-method=stream. On PG11+ they are binary identical. However on PG10
> and lower the segments are binary different. The logical content are the
> same while meaningless tails differ.
>
> In case of pg_basebackup --wal-method=fetch the segments are entirely the
> same. And this case doesn't produce any problem in archiving.
> I think the backup tools as well as custom archive scripts have to take
> account of these points.
>
> <http://www.webmdhealthservices.com/>
>
> --
> Regards,
> Maksim Milyutin
>
>

--

*Andre Piwoni*

Sr. Software Developer, BI/Database

*Web*MD Health Services

Mobile: 801.541.4722

www.webmdhealthservices.com