I have an search-heavy ES cluster with 5 data nodes and 3 masters. There is a single index with 5 shards and 1 replica each. There are about 3500 search operations per minute. 99% of the time it handles them really well and is incredibly fast. The CPU usage for each data node typically hovers around 35%. There are small indexing operations (all using the _bulk apis to index in batches of 1500) happening semi-frequently without issue (10,000 or less docs). However, occasionally I'll have a batch of 100-150k documents (again, indexed in batches of 1500 at a time) which renders the cluster useless for searches for a few minutes until after it's finished. The CPU on each node will spike to 99% for the duration.
This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.