OpenNebula RAFT HA questions


(Stefan Kooman) #1

As far as I know I have correctly setup the ONE HA cluster (have followed the documentantion). I have set up several clusters. I always end up with one host with a VOTE of “-1”. Is that expected behaviour? I would expect to see all hosts vote for the same leader.

HA & FEDERATION SYNC STATUS
ID NAME STATE TERM INDEX COMMIT VOTE FED_INDEX
0 host1 leader 69 1067254 1067254 0 1066982
1 host2 follower 69 1067254 1067254 0 1066982
2 host3 follower 69 1067254 1067254 -1 1066982

Besides that, I see that once in a while the leader changes to a diffrent host. I have verbose (debug) logging enabled in oned.conf. I cannot find the reason for this. I have “XMLRPC_TIMEOUT_MS=0” as according to this important note “We have detected a problem in the OpenNebula configuration that can be easily fixed changing the value RAFT/XMLRPC_TIMEOUT_MS to 0. After modifying /etc/one/oned.conf you should restart OpenNebula service.”. Is this still needed in ONE 5.4.3?


(Ruben S. Montero) #2

Hi

  1. Yes it is normal, note that to become a leader you need a majority 2 out 3 in your case. So in your example means that 0 + 1 voted for 0, vote from 2 was not considered

  2. You may see leader changes because of load peaks, so heartbeats may delay.

Cheers