Ah, so it could be that the dead pool node isn't the cause of my "end
of file errors"- I was theorising that the attempts to reconnect to
the downed door on the pool node were the cause of my problems. It
could just be a coincidence, our ganglia shows a high rate of
transfers starting just before our pool node died, the load could have
caused the pool node to die for some reason then gone on to cause the
"end of file" problem later.
Sorry if that was a bit rambly, kinda thinking out loud. Did you
manage to grab an `ls' of the loginBroker before the restart to see if
it had a overly large number of connections listed?
cheers,
Matt
|