Feature #17597
openleader should recognize its own clock skew instead of assuming the whole cluster is skewed
0%
Description
In a cluster with 5 monitors where only one of the monitors had a crashed ntpd, the check for determining clock skew incorrectly reported that the other four mons had clock skew (where ntp was running fine). From the point of view of the monitor with the crashed ntpd this might be true, but now for a cluster as a whole. After restarting the failed ntpd the clock skew issue was resolved. It looks that the logic determining clock skew could be improved on this point. Better warning/error messages help to pinpoint/resolve the issue faster.
Updated by Stefan Kooman over 7 years ago
Version info, all cluster nodes running:
ceph --version
ceph version 10.2.3 (ecc23778eb545d8dd55e2e4735b53cc93f92e65b)
Updated by Joao Eduardo Luis over 7 years ago
- Tracker changed from Bug to Feature
- Subject changed from ceph -s "Clock skew detected" reports the wrong monitor to leader should recognize its own clock skew instead of assuming the whole cluster is skewed
- Category set to Monitor
This behaviour exists by design. All clock skews are reported as seen by the leader.
I agree, the logic could certainly be improved to provide more helpful messages, but this is a feature rather than a bug imo.