On Fri, 4 Apr 2008, Tom Lane wrote:
> The actual advice I'd give to a DBA faced with such a case is to
> kill -ABRT the bgwriter and send the stack trace to -hackers.
And that's a perfect example of where they're trying to get to. They
didn't notice the problem until after the crash. The server didn't come
back up (busy processing WALs) and that downtime was caught by a
monitoring system. At that point it was too late to collect debugging
information on what was wrong inside the server processes that might have
given a clue what happened.
If they'd have noticed it while the server was up, perhaps because the
"last checkpoint" value hadn't changed in a long time (which seems like it
might be available via stats even if, as you say, the background writer is
out of its mind at that point), they could have done such a kill and
collected some actual useful information here. That's the theory at
least.
--
* Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD