We have a Solr cluster (v6.1.0) running in Standalone mode comprised of 1 master node and 2 slave nodes. Index mode is Direct with two threads. We have daily scheduled full indexes at off hours. Sometimes (not every day), after the full index completes, the cpu utilization stays at 25% indefinitely until the master node is eventually restarted. Logs on master node show only info messages referring to flip/flop every 3 seconds.
The only way to make the cpu go back down is to restart the master node. After restart cpu is between 2-10% except during indexing but normally returns to 2% after indexing except when the problem condition occurs.