On 26/11/2025 23:15, Álvaro Herrera wrote:
> On 2025-11-26, Heikki Linnakangas wrote:
>
>> What happens if you replay the WAL generated with old binary, without
>> this patch, with new binary? It's not good:
>
> Maybe this needs a new record identifier, separating old wal from that generated by the new code?
One downside of that is that the new WAL record type would be unreadable
by older versions. We recommend upgrading standbys before primary, but
it'd still be nicer if we could avoid that.
Maybe we can make RecordNewMultiXact() tolerate the missing page, in
this special case of replaying WAL and the multixid being at the page
boundary.
- Heikki