On Mon, Sep 26, 2011 at 2:23 PM, Guillaume Lelarge
<guillaume@lelarge.info> wrote:
> And I would agree (that the current behaviour is broken). Using a
> database name as a flag to replication connection was a false good idea.
> But, actually, I failed to find a better one.
Well, that may or may not be a good idea, but that's a separate issue.
pg_hba.conf has a very specific way of working, and the replication
'database' doesn't work that way -- it should follow the same rules
the other databases do since it's stored in the same area and should
implicitly use the same mechanics. A cleaner way of doing it might
have been to introduce a separate area for virtual databases for
example (and this might have mitigated pain for the non-zero chance
for users that already have a database named 'replication').
Maybe it's too late to change it now, though :(, we should do
something about the current situation, even if the best we can come up
with is putting a clearly worded disclaimer into the docs. I still
think it's better to make 'all' work though.
merlin