On Mon, Apr 7, 2014 at 3:04 AM, Michael Paquier
<michael.paquier@gmail.com> wrote:
> On Fri, Apr 4, 2014 at 9:57 PM, Michael Paquier
> <michael.paquier@gmail.com> wrote:
>> For 9.4, clearly yes, this would change the semantic of recovery and
>> this is not something wise to do at the end of a development cycle.
>> For 9.5 though, this is a different story. It clearly depends on if
>> this is though as useful enough to change how recovery fetches WAL
>> files (in this case by scanning existing repslots). There are other
>> things to consider as well like for example: do we reset the
>> restart_lsn of a repslot if needed WAL files are not here anymore or
>> abort recovery? I haven't worked much with repslots though...
> Coming back to that, I am still wondering if for the time being it
> would not be better to add in pg_basebackup documentation that
> replication slot information is not added in a backup, per se the
> patch attached.
Not sure if this is exactly the right way to do it, but I agree that
something along those lines is a good idea. I also think, maybe even
importantly, that we should probably document that people using
file-copy based hot backups should strongly consider removing the
replication slots by hand before using the backup.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company