ES Version: 5.5.1
We use Elasticsearch mainly for term aggregation. In one of our index, we see a spike in field data memory usage which corresponds with a spike in CPU utilization and load average too. All the fields in that index are either keyword, integer or float. From reading the documentation I know that only 'text' fields use field data for sorting and aggregation. Since we do not have any text field in our index then why there is a spike in field data memory usage. Also, we haven't set indices.fielddata.cache.size so cache is unbounded, there is no eviction from field data.
Any pointer in understanding this will be quite helpful.