I'm back at looking into this again, after a rather exhausting week. I
think I have a better grasp of what is going on in this code now, and it
appears to me that we should change the locking so that active_pid is
protected by ReplicationSlotControlLock instead of the slot's spinlock;
but I haven't analyzed the idea fully yet and I don't have the patch
done yet either. I'm going to report, hopefully with a fix committed
this time, on Monday at 19:00 CLT.
--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services