Re: logical replication syntax (was DROP SUBSCRIPTION, query cancellations and slot handling)

On 02/05/17 15:14, Petr Jelinek wrote:
> On 02/05/17 15:10, Tom Lane wrote:
>> Robert Haas <robertmhaas(at)gmail(dot)com> writes:
>>>> On Thu, Apr 20, 2017 at 7:46 AM, Petr Jelinek
>>>> <petr(dot)jelinek(at)2ndquadrant(dot)com> wrote:
>>>>> DROP SUBSCRIPTION mysub NODROP SLOT;
>>
>>>> I'm pretty uninspired by this choice of syntax.
>>
>> Actually, this command has got much worse problems than whether you like
>> the spelling of its option: it shouldn't have an option in the first
>> place. I put it to you as an inviolable rule that no object DROP command
>> should ever have any options other than RESTRICT/CASCADE and IF EXISTS.
>> If it does, then you don't know what to do when the object is recursed
>> to by a cascaded drop.
>>
>> It's possible that we could allow exceptions to this rule for object types
>> that can never have any dependencies. But a subscription doesn't qualify
>> --- it's got an owner, so DROP OWNED BY already poses this problem. Looks
>> to me like it's got a dependency on a database, too. (BTW, if
>> subscriptions are per-database, why is pg_subscription a shared catalog?
>> There were some pretty schizophrenic decisions here.)
>
> Because otherwise we would need launcher process per database, not pretty.
>
>>
>> So ISTM we need to get rid of the above-depicted syntax. We could instead
>> have what-to-do-with-the-slot as a property of the subscription,
>> established at CREATE SUBSCRIPTION and possibly changed later by ALTER
>> SUBSCRIPTION. Not quite sure what to call the option, maybe something
>> based on the concept of the subscription "owning" the slot.
>>
>
> So what do you do if the upstream does not exist anymore when you are
> dropping subscription?
>

Let me expand, if we don't drop the slot by default when dropping
subscription, we'll have a lot of users with dead slots laying around
holding back WAL/catalog_xmin, that's really bad. If we do drop by
default like now, we need option to not do that, neither RESTRICT, nor
CASCADE fit that.

--
Petr Jelinek http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services