Problems with ntp on one of our systems were traced to using a DNS cache
which came up after ntp tried to poll its servers.

D'OH!

We'd switched dns on the machine to DJBDNS dnscache running under
daemontools. dnscache was taking about 10 seconds to come up after
rc.local (where ntpd is started) completed. Apparently ntpd is pretty
graceless about retrying servers when dns is unavailable.

ntpd log errors were:

13 Dec 04:48:59 ntpd[276]: signal_no_reset: signal 17 had flags 4000000
13 Dec 04:49:01 ntpd[276]: signal_no_reset: signal 14 had flags 4000000

Two instances of ntpd would be present. The second one would hang.

Solution was to put another nameserver in /etc/resolve.conf in addition
to the local cache. I should have had a backup nameserver listed anyway.

My pain has proven most educational.

ScottJ