Basic logic of NTP should not require constant use of ntpdate and restarting of ntpd
Binary package hint: ntp
There are number of issues surrounding this package, and the fact that the number of places ntpdate is called and ntpd service is restarted.
This seems to be causing huge problems and frankly I am questioning the underlying logic here. Surely ntpd should behave appropriately. I therefore propose:
1. Servers which cannot be contacted should be regularly re-tried. This means if ntpd is started early, without network port being available, etc etc it will auto-recover and sync as and when it can.
2. The minimum acceptable drift before ntp stops re-syncing should be abandoned or made optional with the default to re-sync regardless of drift. I understand why that was done originally, but if I run ntpd I am saying 'please sync my computer to the time on these other systems'. That means I want it to happen - not to silently fail to do so if I happen to have had a clock drift by a couple of mins leaving me to take remedial action using ntpdate by hand.
By changing these aspects of ntp we make it more reliable, completely remove the need to prat about with ntp with ifup/down scripts and dhcp scripts etc etc. Put simply it would just *work* instead of being a pain.