On Mon, Aug 30, 2021 at 09:16:51PM -0400, Joe Conway wrote:
> On 8/30/21 8:22 PM, Tom Lane wrote:
>> Yeah, this single data point is not enough justification to blame
>> dynahash.c (which is *extremely* battle-tested code, you'll recall).
>> I'm inclined to guess that the looping is happening a few stack levels
>> further up, in the logical-decoding code (which is, um, not so much).
>
> Heh, fair point :-)
It looks something is messed up with the list of invalidation messages
to process in this code path. Maybe some incorrect memory context
handling?
--
Michael