> Well, that makes sense: if the shutdown took more than a > minute then the > "stop" script action would give up waiting, and then the > "start" action > would see the postmaster running and go away happy. (It's a bit odd > that "service start" actions are supposed to treat "already running" > as OK, but I've been told that that's required by the Linux Standards > Base and I can't change it.) thanks, that's good to know. i never realized that's how it behaves. > The real question here is why'd it take so long to stop? It should be > using "mode fast" which'd kick out active queries. indeed a mystery -- from looking at the query log there didn't seem to be ANY active queries at the time. seems that > 85% of the session IDs (84 total) for which i got lines like "2007-04-23 03:05:48 PDT [26987] FATAL: the database system is shutting down" did not even have any preceding query activity. hard to debug retroactively -- it's just annoying that i don't understand what was different about this restart. and, yes, i confirmed that it is using fast mode: 212 $SU -l postgres -c "$PGENGINE/pg_ctl stop -D '$PGDATA' -s -m fast" > /dev/null 2>&1 < /dev/null george
