I am testing out elasticsearch 2.3.3 performance and found out the index rate drops after a couple of hours (to a few hunder document/sec on each node). I have to restarted the cluster to bring the index rate back. Our setup is to have a client node accepting bulk request from logstash and sends out to a 40 data nodes cluster. They run on raid-0 spinning disks. I tried to watch the thread pool. There is always one node at a time to show bulk rejected value 5. Any insight how to improve the index performance?