I'm running two single node ES cluster: one for pre-production and one for development. In my monitoring and stats system I've mentioned that since the update to 5.3.1 the pre-production system's power usage went up. I've checked the host: system load when idle went up: from near to 0 to somewheren between 50-100% of one core (0.5-1.0 in top). It's on the pre-production box only: development is still at a idle CPU load near 0 (same OS + config, same ES + Kibana + Logstash + Xpack config, same dataset, same non-existent load except when I'm testing). The only difference between the two boxes is the number of cores (24 vs 8) and the number of shards (24 vs 8).
The process causing the load is java. Everything in /var/log/elasticsearch looks clean. Any smart tips about how to figure out what's cuasing the extra CPU cycles?