ceph lossy error messages in log

[Andrey Epifanov]

Detailed bug description:

A lot of messages in CEPH log files produced events in monitoring system (LMA)

September 14th 2016, 13:39:33.000 ceph22 system.syslog ceph-osd EMERGENCY 2016-09-14 20:39:33.762155 7f81ffc99700 0 -- 10.15.16.38:6802/1206747 >> 10.15.16.31:0/1050813 pipe(0x7f82f30a5000 sd=1454 :6802 s=0 pgs=0 cs=0 l=1 c=0x7f82e0004520).accept replacing existing (lossy) channel (new one lossy=1)

The conclusion after the investigations of this issues by L3 CEPH team:

By default policy set in osd for any communications except with other osds is lossy: https://github.com/ceph/ceph/blob/v0.80.9/src/ceph_osd.cc#L393
So any reconnection from anybody, but an OSD will be marked as replacing lossy connection.
This message is normal.

So, please add exception in alarm system for this event.

Steps to reproduce:
 N/A
Expected results:
 No errors in Monitoring system
Actual result:
 Alarm in Monitoring
Reproducibility:
 Always
Workaround:
 Ignore alarms or modification alarm settings
Impact:
 Operator confusing
Description of the environment:
 Operation system: Ubuntu
 MOS: 8.0
Additional information:
 Installed set of LMA plugins for MOS 8.0

[Simon Pasquier (simon-pasquier)]

The problem is that all Ceph messages are logged with the EMERGENCY level. See bug #1482215 for the details.