Actions
Bug #58049
closedmon:stretch-cluster: mishandled removed_ranks -> inconsistent peer_tracker leading to unable to form quorum
Status:
Resolved
Priority:
Urgent
Assignee:
Category:
Stretch Clusters
Target version:
-
% Done:
0%
Source:
Tags:
backport_processed
Backport:
pacific,quincy
Regression:
No
Severity:
3 - minor
Reviewed:
Affected Versions:
ceph-qa-suite:
Component(RADOS):
Pull request ID:
Crash signature (v1):
Crash signature (v2):
Description
First encountered in the downstream: https://bugzilla.redhat.com/show_bug.cgi?id=2142674
When we failover monitors many times in the stretch cluster, there are instances where Ceph becomes
unresponsive due to monitors not being able to form a quorum.
We have investigated and concluded that this is due to how we mishandled `removed_ranks` in MonMap which
leads to inconsistent peer_tracker which then leads to deadlock election state of the monitor, which means
they cannot form a quorum -> ceph becomes unresponsive.
Actions