High CPU usage + BIND not responding

[Jim Reid]

>>>>> "Conny" == Conny Brunnkvist / dimac net <c.brunnkvist at speednames.com> writes:

    Conny> I run BIND 8.2.3-REL on a Red Hat Linux-boxen with a couple
    Conny> of thousand zones, and now I want to use query-logging for
    Conny> examining the usage plus be able to extract some
    Conny> statistics.

Fine, but how is this related to the problem below?

    Conny> If a zone is added/changed (->named.conf updated), an
    Conny> update runs and the zone/zones in question is reloaded via
    Conny> 'ndc reload <zone>'.  Thing is, it works fine for a few
    Conny> updates it seems, but at some point the processor usage of
    Conny> named runs away to a 99-100%, the server stops answering
    Conny> queries (althoug logs keep being written to), and I'm
    Conny> forced to kill and restart it.

If you add or remove zone{} statements, you should just "ndc reconfig"
the server. Don't do a reload because that makes the server re-read
and load every zone file it has. ["ndc reload zone-name" is OK, but
I've no idea what happens if the zone isn't already known to the name
server.] If there are lots of zones or the files are big, reloading
can take a long time. And while it's doing that, the server can't
answer queries. Try putting a system call trace on the name server
when this CPU usage increases and see what it's doing. My guess is
it's opening, reading and closing lots of zone files.