Re: DROP SUBSCRIPTION hangs if sub is disabled in the same transaction

On Sat, Sep 16, 2017 at 9:52 PM, Peter Eisentraut
<peter(dot)eisentraut(at)2ndquadrant(dot)com> wrote:
> On 9/15/17 13:35, Arseny Sher wrote:
>> Peter Eisentraut <peter(dot)eisentraut(at)2ndquadrant(dot)com> writes:
>>
>>> Here is a simple patch that fixes this, based on my original proposal
>>> point #4.
>>
>> I checked, it passes the tests and solves the problem. However, isn't
>> this
>>
>> + if (slotname || !subenabled)
>>
>> is a truism? Is it possible that subscription has no slot but still
>> enabled?
>
> Yeah, we could just remove the _at_commit() branch entirely. That would
> effectively undo the change in 7e174fa793a2df89fe03d002a5087ef67abcdde8,
> but I don't see any other choice for now. And the practical impact
> would be quite limited.
>

Yeah, we can remove only _at_commit() branch, but other part of the
commit is still valid for ALTER SUBSCRIPTION DISABLE.

>> Besides, we can avoid stopping the workers if subscription has no
>> associated replication origin, though this probably means that
>> subscription was broken by user and is not worth it.
>
> Right, it seems not worth addressing this case separately.
>

Once we got this patch, DROP SUBSCRIPTION is not transactional either
if dropping a replication slot or if the subscription got disabled in
a transaction block. But we disallow to do DROP SUBSCRIPTION in a
transaction block only in the former case. In the latter case, we
adopted such non-transactional behaviour. Since these behaviours would
be complex for users I attached the documentation patch explaining it.

Regards,

--
Masahiko Sawada
NIPPON TELEGRAPH AND TELEPHONE CORPORATION
NTT Open Source Software Center