> I don't believe pg_postmaster_start_time() is reset by a
> backend-crash-and-restart sequence ... but the stats collector data will
> be. You might want to look into the postmaster log for evidence of
> such a crash.
I'm sure that the backend has neither restarted nor crashed at that moment.
Anyway, I've also confirmed this in the server log, I don't find any
evidence or trace of this happened.
Is there any other possible reasons that could make autovacuum daemon to
update it to NULL value?
NOTE: "log_min_messages" is set to NOTICE.