Actions
Bug #3714
closedosd: new peering code does not consume osdmaps prior to booting
% Done:
0%
Source:
Community (user)
Tags:
Backport:
Regression:
Severity:
Reviewed:
Affected Versions:
ceph-qa-suite:
Pull request ID:
Crash signature (v1):
Crash signature (v2):
Description
Previously when we handled the old osdmaps catching up (pre-MOSDBoot) we'd do advance_map and the pgs would update their past_intervals. The new code does not force them to do this catch-up prior to booting and joining the cluster, which means an osd that has been down for a long time will mark itself up and only then start chewing through maps. In paravoid's case, it has been long enough that the peering thread times out and suicides.
Actions