We have a nightly restart of one PG database. Today it failed and I
can't seem to understand why and how to prevent this in the future (nor
can I reproduce the problem).
We have a line in a shell script that calls "/etc/init.d/postgresql
restart". In the shell script's log from this invocation I have:
Stopping postgresql service: [FAILED]
Starting postgresql service: [ OK ]
The outcome of this was that the service was not running (despite the "[
OK ]" on the second line). The query log from that run ends with:
2007-04-23 03:03:59 PDT [23265] LOG: received fast shutdown request
2007-04-23 03:03:59 PDT [23265] LOG: aborting any active transactions
2007-04-23 03:03:59 PDT [26749] FATAL: terminating connection due to
administrator command
<... snipped more lines like the one above ...>
2007-04-23 03:03:59 PDT [24090] LOG: could not send data to client:
Broken pipe
2007-04-23 03:03:59 PDT [26700] FATAL: terminating connection due to
administrator command
<... snipped more lines like the one above ...>
2007-04-23 03:04:13 PDT [26820] FATAL: the database system is
shutting down
<... snipped more lines like the one above ...>
2007-04-23 03:06:10 PDT [23269] LOG: shutting down
2007-04-23 03:06:10 PDT [23269] LOG: database system is shut down
2007-04-23 03:06:13 PDT [23267] LOG: logger shutting down
<end of log>
So it looks like the STOPPING of the service actually succeeded, albeit
it took a while (more than the usual sessions open?). The STARTING is
the one that actually failed (is that because the STOP was still in
process?). The question is why -- in a RESTART situation
wouldn't/shouldn't the START part wait for the STOP part to complete
(regardless of how long it takes)?
Also what can we do to avoid this in the future? We can issue a separate
STOP and then a START X minutes later, but how long an X? It would seem
that a RESTART is really what I want...
TIA,
George
