We have a cluster composed of 3 masters and 4 data nodes.
Each data node's jvm has 30Go memory allowed. The disk usage on 4 data nodes is around 17 To. There is actually no indexes closed.
We see 2 things :
- data node jvm is frequently high ( > 80%).
- we already had circuit breaker problem twice
My questions are:
- What makes jvm using so many memory?
- Do opened indices are playing a role in that ?
- Any method to lower this jvm usage ?