logical replication seems broken

Hello,

I am seeing errors in replication in a test program that I've been running for years with very little change (since 2017, really [1]).

The symptom:
HEAD-replication fails (most of the time) when cascading 3 instances (master+2 replicas).

HEAD-replication works with 2 instances (master+replica).

I have also compiled a server on top of 4ad31bb2ef25 (avoiding some recent changes) - and this server runs the same test program without failure; so I think the culprit might be somewhere in those changes.  Or (always possible) there might be something my testing does wrong - but then again, I do this test a few times every week, and it never fails.

This weekend I can dig into it some more (make a self-sufficient example) but I thought I'd mention it already. perhaps one of you see the light immediately...


Erik Rijkers

[1] https://www.postgresql.org/message-id/flat/3897361c7010c4ac03f358173adbcd60%40xs4all.nl