The time tracing service ntpd/chrony has a maxpoll option can be used to
ask the time daemons to check the upstream time server more frequently even tough
they might assume less is enough.
As you can see most time keeping device used in consumer electronics is not so accurate on its own: http://www.best-microcontroller-projects.com/ppm.html
The issue can be worst in a virtual machine and the drift /accuracy can change over time (temperature, load, aging),
The time demons also can miss information about how good the current machine clock the early times in a CI jobs.
When operator is keen to have more accurate time, the more frequent consultation with a time server (or special hardware) is required. The maxpoll prevents the time daemon to choose less frequent synchronization.
The default value is 1024 sec,
my current goal is reach less 30ms difference across virtual machines for CI jobs.
For example a 20ppm clock accuracy can lead to +-20 ms (40ms across 2 machine) in 1024 sec.
Patch on review: https:/