DNS Cache, database backends...

[Jim Reid]

>>>>> "Forrest" == Forrest Aldrich <forrie at navipath.com> writes:

    Forrest> There are often legitimate reasons that require the
    Forrest> server to be rebooted -- mostly to do with kernel updates
    Forrest> and other facets of the OS that require rebooting.  Just
    Forrest> a part of the work.

These are valid reasons for rebooting. But a properly configured name
server should not need rebooting frequently. All it should run is the
name server, an NTP daemon and SSH. These bits of software and the
kernel should be stable enough to not require changing all that often.
A reboot should only be needed for fixing any important kernel bugs or
if the whole OS is upgraded.

FWIW, one of the name servers I used to run was up for almost 3 years
before it was rebooted. [In fact, a new box was installed when the
beancounters decided the old hardware should be retired because it had
been depreciated in the accounts. The box and the server were still
running just fine but the accountants wanted written-off hardware to
be replaced.] The name server on that box was only restarted when a
new version of BIND was installed: typically every 6-8 months or so
when a new release came out. I don't see any reason why somebody else
couldn't or shouldn't operate their name servers in a similar
way. Personal opinion: if someone is stopping and starting production
name servers frequently, there's something seriously wrong with their
procedures and business processes. So if your users have a zillion
stub resolvers pointing at your name server(s), you just don't stop
and start them often (if at all).