We have a 3 node master-data ES cluster . We face random high cpu utilization in random nodes though there is no much of activity happening in ES.
This high cpu utilization and Load stays forever until the node is restarted by killing the ES process.
Service restart also doesnt work in this case
We are using ES 5.6 and openjdk version "1.8.0_144"
Other problem is when we restart the ES node the shards are not evenly distributed and gets assigned on single node.
I am pasting the output of _node/hot_threads query
I am not that expert in ES.