Exhibitor zookeeper ensemble with 2 leaders
maverick2202 opened this issue · comments
We have 5 node zookeeper ensemble started by exhibitor. We have seen sometimes when a zookeeper restarts (due to machine failure) that there are 2 leader running for a transient time. Also, we have seen that znode exists on 1 out of 5 zookeeper but not all during that time that exhibitor is restarting zookeeper.
Is there a way to force quorum on exhibitor/zookeeper so that it doesn't except writes.
Example:
[
{
"code": 3,
"description": "serving",
"hostname": "192.0.3.22",
"isLeader": false
},
{
"code": 3,
"description": "serving",
"hostname": "192.0.3.20",
"isLeader": true
},
{
"code": 3,
"description": "serving",
"hostname": "192.0.3.21",
"isLeader": true
}
]
In the 3 node cluster, 2 nodes are leader for very short period of time and then it goes back to 1. However, during this short time won't there be a problem as data can be inconsistent.
Unfortunately Exhibitor does support things like denying writes during certain failure scenarios.
This is really more of a zookeeper problem in terms of its handling of node failures.