Bug #49158
doc: ceph-monstore-tools might create wrong monitor store
0%
Description
I tried to recover mon quorum from OSDs with the following document.
There are three mons a, b, and c in my cluster.
```
$ kubectl get svc -n ceph-hdd
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT AGE
...
rook-ceph-mon-a ClusterIP 10.68.74.234 <none> 6789/TCP,3300/TCP 24h
rook-ceph-mon-b ClusterIP 10.68.56.39 <none> 6789/TCP,3300/TCP 24h
rook-ceph-mon-c ClusterIP 10.68.112.111 <none> 6789/TCP,3300/TCP 24h
...
```
Here `CLUSTER-IP` field means an IP address of each mon.
I ran `ceph-monstore-tool` and the names of mons are changed by mistake.
```
...
0: v1:10.68.56.39:6789/0 mon.a # Should be "mon.b"
1: v1:10.68.74.234:6789/0 mon.b # Should be "mon.a"
2: v1:10.68.112.111:6789/0 mon.c # Should be "mon.c"
...
```
It's due to this tool's assumption that mons' names are sorted by IP address. However, it's not always true after the following PR.
History
#1 Updated by Neha Ojha over 2 years ago
- Status changed from New to Fix Under Review
- Pull request ID set to 39288
#2 Updated by Satoru Takeuchi over 2 years ago
This problem was fixed in the following PR.