Re: pg_restore causing deadlocks on partitioned tables

I wrote:
>> (2) ALTER TABLE ONLY ... ADD CONSTRAINT on a partition root tries to get
>> AccessExclusiveLock on all child partitions, despite the ONLY.

> The cause of this seems to be that ATPrepSetNotNull is too dumb to
> avoid recursing to all the child tables when the parent is already
> attnotnull. Or is there a reason we have to recurse anyway?

I wrote a quick patch for this part. It seems pretty safe and probably
could be back-patched without fear. (I also noticed that
ATSimpleRecursion is being unnecessarily stupid: instead of the
demonstrably not-future-proof relkind check, it could test relhassubclass,
which is not only simpler and less likely to need future changes, but
is able to save a scan of pg_inherits in a lot more cases.)

As far as I can tell in some quick testing, this fix is sufficient to
resolve the complained-of deadlock. It'd still be a good idea to fix the
TRUNCATE side of things as well. But that would be hard to back-patch
because removing ri_PartitionCheck, or even just failing to fill it,
seems like a potential ABI break for extensions. So my proposal is
to back-patch this, but address the ResultRelInfo change only in HEAD.

regards, tom lane