Simon Riggs wrote:
> I'm surprised that a network roundtrip takes less time than the
> backend takes to mark clog and then queue for the SyncRepLock.
I'm not surprised by that at all. Some of our replication involves
Gb or faster connections on the same LAN segment (sometimes on the
same switch).
That's rather beside the point though -- on a busy system a process
can get starved long enough to make some seemingly improbably timings
occur. PostgreSQL shouldn't fall over just because the load gets
heavy enough to cause such timings. A race condition is a race
condition.
-Kevin