Red Hat Bugzilla – Bug 23530
RFE: start ntpd earlier
Last modified: 2007-04-18 12:30:34 EDT
Could ntpd be started a bit earlier? This would
guarantee correct time for more init scripts/programs.
# chkconfig: - 26 74
# chkconfig: - 55 10
(this would still guarantee that ntpd is found - this
config would mean starting it right after and stopping
it just before netfs.)
ntpd usually takes some minutes before correcting system time, so moving it in
the bootup sequence doesn't make any difference.
Not true. ntpdate sets the time efficiently, synchronization an PLL lock is what
takes 5+ minutes.
All dependent on configuration, which we cannot ship with enabled by default,
Yes, ntpdate does immediately set system time. But ntpd (which we are talking
about) does not.
Am I screwed somewhere? ;>
If /etc/ntp/step-tickers is used, ntpdate is called immediately at
ntpd startup (see the startup script). No delay here (hm, don't
think this is documented; should be, I suppose ...). I used to
call ntpdate in rc.local until I found out that this is provided in the
ntpd startup script ...
What's needed is to set the system accurately and early as possible, not to
a server. FWIW, ntpd guarantees monotonicity (i.e. time never goes backward),
ntpdate does not, but that's another reason to run ntpdate if possible as early
:) Seems to be perfectly ok with my request (yes, some time ago
our time server admin who's using one of my Linux servers told
me about all that stuff ...).
Starting ntpd early means starting ntpdate early, right?
Yup. Howvere, the real problem is that we can't distribute with a pre-configured
(hate the name, <shrug>) server.
changed in ntp 4.0.99k-7 and later.