On Wed, 21 Aug 2024, Abhijith Sethuraj wrote:

[CAUTION: Non-UBC Email]I spoke too soon, it failed over to "foobar" right now. 
What's the
recommended way to make this happen faster?

Why should it? It has no idea whether pps disappeared because of death, or
because it is was temporarily indiposed. It has trained the system clock so
that it is tracking UTS with high accuracy. Ie your system clock is better
than almost any other clock you have. Why should it transfer to a possibly
worse clock just because of a temporary glitch in the better clock?

YOu might know that you put a bullet through the PPS. chrony does not and
will wait a while to see if it comes back to life before going to someone
else.


Thanks,
Abhijith

On Wed, Aug 21, 2024 at 3:13 AM Abhijith Sethuraj <[email protected]> 
wrote:
      Thanks, that's very helpful.

Here's what I see:

S Name/IP Address        Auth COpts EOpts Last Score     Interval  Leap
=======================================================================
* 1PPS                      N -P--- -P---    1   1.0    -7ns   +31ns  N
D foobar                    N -P--- -P---    3   1.0 -1550ns +2091ns  N

Going through the man pages, "foobar" is in "D" state when its root distance 
(in a recent
measurement) is greater than combinelimit * root distance of current best 
source's root
distance. So, I tried increasing combinelimit to a higher value and it started 
combining.
The 1PPS signal is coming from a time server appliance that has a good clock 
and uses a GNSS
ref clock as well; "foobar" is another good source in the same site. In such a 
case, is it
best to combine both 1PPS and "foobar" sources; or should I just use 1PPS (if 
available) and
fallback to other sources if 1PPS isn't available (that is, don't forcefully 
combine them)?

Also, one thing that I noticed is that if I stop the 1PPS signals, it's not 
falling back to
"foobar"; 1PPS remains as the best source. Should I add something in the 
configs to make
this work?

S Name/IP Address        Auth COpts EOpts Last Score     Interval  Leap
=======================================================================
* 1PPS                      N ----- -----  228   1.0  -200us  +200us  N
D foobar                    N ----- -----    1   1.0 -2650ns +1040ns  N


Thanks in advance,
Abhijith


On Tue, Aug 20, 2024 at 3:45 PM Miroslav Lichvar <[email protected]> wrote:
      On Tue, Aug 20, 2024 at 03:23:03PM +0530, Abhijith Sethuraj wrote:
      > Hello,
      >
      > We have a 1PPS input to the NIC and we're planning to run chrony to use
      > 1PPS as an additional refclock, in addition to getting time of day via 
NTP.
      > When we tried this out, we're not entirely sure if things work fine as 
the
      > source that gives time of day shows up as "unusable" in `chronyc 
sources`.

      The "?" character in the sources means a different state than those
      previously listed. Run chronyc selectdata -v to get the actual
      selection state.

      --
      Miroslav Lichvar


      --
      To unsubscribe email [email protected]
      with "unsubscribe" in the subject.
      For help email [email protected]
      with "help" in the subject.
      Trouble?  Email [email protected].


Reply via email to