On Wed, Feb 22, 2017 at 4:12 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Fujii Masao <fujii@postgresql.org> writes:
>> Fix connection leak in DROP SUBSCRIPTION command.
>> Previously the command forgot to close the connection to the publisher
>> when it failed to drop the replication slot.
>
> If there's a bug here, this seems like an extremely unreliable way of
> fixing it. What if an error gets thrown before you reach that ereport?
>
> In other words, this coding is assuming that the walrcv_command()
> subroutine cannot throw an error, which I would consider dangerous
> even if it were a fixed subroutine. If it's a hook that's doing
> unknown stuff, that seems a completely untenable assumption. You
> really need either to hook the cleanup action into normal error
> recovery, or to use a PG_TRY block.
To be honest, I have thought about using PG_ENSURE_ERROR_CLEANUP()
when seeing the thread. If other ERROR messages are generated in the
future that the current fix would be unreliable.
--
Michael