On Wed, Oct 20, 2021 at 9:01 PM Kyotaro Horiguchi <horikyota.ntt@gmail.com> wrote: > > lastOverflowedXid is the smallest subxid that possibly exists but > possiblly not known to the standby. So if all top-level transactions > older than lastOverflowedXid end, that means that all the > subtransactions in doubt are known to have been ended.
Thanks for the patch! I verified that it appears to reset lastOverflowedXid properly.
Is it right time to register the patch in the current commit fest, right? (How to do that?)
On a separate note, I think it would be really good to improve observability for SLRUs -- particularly for Subtrans SLRU and this overflow-related aspects. pg_stat_slru added in PG13 is really helpful, but not enough to troubleshoot, analyze and tune issues like this, and the patches related to SLRU. Basic ideas:
- expose to the user how many pages are currently used (especially useful if SLRU sizes will be configurable, see
- Andrew Borodin also expressed the idea to extend pageinspect to allow seeing the content of SLRUs
- a more specific thing: allow seeing lastOverflowedXid somehow (via SQL or in logs) - we see how important it for standbys health, but we cannot see it now.