Backport #17285
ceph-mon leaks in MDSMonitor when ceph-mds process is running but MDS is not configured
History
#1 Updated by Igor Podoski over 1 year ago
PR https://github.com/ceph/ceph/pull/10238 is fixing this leak at hammer branch. After v10.0.5 there was massive MDSMonitor code reorganization and bug probably was fixed.
#2 Updated by Nathan Cutler over 1 year ago
- Tracker changed from Bug to Backport
#3 Updated by Nathan Cutler over 1 year ago
- Description updated (diff)
original description¶
Memory is leaking every time when you see in monitor log:
mon.0@0(leader).mds e1 warning, MDS mds.? XXX.XXX.XXX.XXX:6800/1494583 up but filesystem disabled
Is not very noticeable, however when cluster is in unstable state, e.g. network cards are dropping random packets, monitors tend to fall into election loop. When two monitors are marking themselves as leaders for a while and then start election from the beginning, MDS messages are forwarded from one to another (to leader). After a while amount of MDS messages is quite big and when they hit monitor which is leader (at that point of time) memory leak is very visible and it leads to OOM kill.
This will only occur on disabled / not configured MDS.
Bug lasts until ceph v10.0.5.
#4 Updated by Nathan Cutler over 1 year ago
- Status changed from New to In Progress
#5 Updated by Nathan Cutler over 1 year ago
- Description updated (diff)
#6 Updated by Nathan Cutler over 1 year ago
- Subject changed from ceph-mon leaks in MDSMonito when ceph-mds process is running but MDS is not configured to ceph-mon leaks in MDSMonitor when ceph-mds process is running but MDS is not configured
#7 Updated by Nathan Cutler over 1 year ago
- Status changed from In Progress to Resolved
- Target version set to v0.94.10