Check elasticsearch logs for OOM - Out of memory event. That can kill entire node.
Increase jvm settings for that node first.
Second - check if there are shards greater or even close to JVM.
Reindex them into smaller shards.
Avoid for future creating shards close to jvm.
