I've just gone live using postgres 6.3.2 on Linux/Intel (using the
Red Hat postgresql-6.3.2-4 RPM) for a mail server cluster with a
web front end for users. postmaster runs on a frontend server and
has connections to it from Apache/mod_perl processes (cached
connections via Apache::DBI). It's been fine during light testing
but this morning the first taught course on it started and about 30
people tried to log in at once (I'm told). The symptoms were that
current connections hung and my attempt to use psql to connect
generated the following error messages:
[~]frontend1% psql herald
NOTICE: SIAssignBackendId: discarding tag 2147483487
Connection to database 'herald' failed.
FATAL 1: Backend cache invalidation initialization failed
[~]frontend1% psql herald
NOTICE: SIAssignBackendId: discarding tag 2147483487
Connection to database 'herald' failed.
FATAL 1: Backend cache invalidation initialization failed
I ended up having to kill and restart the postmaster and all
surviving postgres processes (even doing that gave bind errors for
a while: perhaps postgres doesn't use SO_REUSEADDR?). Please can
someone tell me if this problem is as serious as it looks and where
I should look to try to fix it? I can't afford to have such outages
(and how frequent are they likely to be?) and 4000 students are
going to show up on Thursday and start hammering the heck out of it.
--Malcolm
--
Malcolm Beattie <mbeattie@sable.ox.ac.uk>
Unix Systems Programmer
Oxford University Computing Services