Re: Clock selection algorithm
chrony-users-Cohe+2zZ/ftmR6Xm/[email protected]
| Newsgroups | gmane.comp.time.chrony.user |
|---|---|
| Organization | PE1CHL |
| Message-ID | <[email protected]> |
I have changed the logging config like this: # tunable parameters driftfile /var/lib/chrony/ntp.drift logdir /var/log/ntpstats -log statistics measurements tracking tempcomp +log refclocks i.e. I was logging some other things but I wasn't really using that. After restarting chrony (I tried this on two different systems) I have not yet seen the problem re-occur. Looking in the refclocks logs I only see some offsets of the e-03 order during the first minutes after startup, not thereafter. I will monitor the situation further to see if there it happens again. PS: there are some issues with this mailinglist making it troublesome in this era of SPF and DMARC. 1. it sends copies of the mail back to the submitter 2. it sends out mails with a From: header containing the original submitter's e-mail address. That is not really feasible anymore. It would be better when the mailinglist sends out mail messages with a From: address referring to an address on the mailinglist server and mentioning the submitter only in the Full Name. On 3/7/22 10:33, Miroslav Lichvar wrote: > On Thu, Mar 03, 2022 at 07:23:30PM +0100, chrony-users-Cohe+2zZ/ftmR6Xm/[email protected] wrote: >> But sometimes, probably due to interrupt latency, the offset of the PPS refclock has >> some spikes (e.g. spiking to 1 millisecond) and in this situation the PPS clock >> is sometimes marked as an outlier (x) because all the NTP refclocks are closer to the >> system time than this PPS sample is. Chrony selects an NTP refclock and increases >> the stratum. It spends some time (like 5-10 minutes) tracking the NTP servers before >> it reselects the PPS source. This then alerts our monitoring system. > 1 millisecond is a lot for an interrupt delay. If you enable the > refclocks log, how many subsequent samples do you see impacted? If > it's not too many, you could simply increase the polling interval and > let the median filter take care of the outliers. > > If that doesn't work and you want to force chronyd to keep the PPS > refclock selected, you can set a larger delay for the refclock (e.g. > delay 0.005) to make it overlap with the other sources and avoid the > falseticker status. > -- To unsubscribe email chrony-users-request-kWFZVVI9zxvPqho9SqqRMmD2FQJk+8+b@public.gmane.org with "unsubscribe" in the subject. For help email chrony-users-request-kWFZVVI9zxvPqho9SqqRMmD2FQJk+8+b@public.gmane.org with "help" in the subject. Trouble? Email [email protected]