On 2022-Aug-26, Jelte Fennema wrote:
> I indeed don't think this problem is unique to subscriptions, but it seems
> better to at least have this problem in a few places less (not making perfect
> the enemy of good).
>
> If someone has a more generic way of solving this for other commands too,
> then that sounds great, but if not then slowly chipping away at these cases
> seems better than keeping the status quo.
>
> Attached is a new patch where ALTER SUBSCRIPTION ... OWNER TO ... can
> now also be executed concurrently with the other subscription commands.
Would it work to use get_object_address() instead? That would save
having to write a lookup-and-lock function with a retry loop for each
object type.
--
Álvaro Herrera Breisgau, Deutschland — https://www.EnterpriseDB.com/