Longer search queue when we have low search thread utilization

We are seeing slow query performance on elastic-search. This coincide with increase in search queue size but at the same time active search threads as a percentage of search thread pool size is pretty low. This indicate that even though we have many threads that could be assigned to the search request queuing on the node, they are left idle.

One explanation to this could be that the default max concurrent search size on the multi-search request is the reason why we are seeing low thread utilization in search thread pool. Searches in these multi search request are the ones that are queued for thread bottlenecked by the max concurrent search size setting on the multi-search request. Does this make sense? Is there any other other reason why this should happen?

Not really, no, tasks are only placed in a queue if all search threads are active, but you seem to be saying that there are items in the queue and idle threads at the same time? Can you share the output of GET /_cat/thread_pool/search?v?

You are right. It did not make sense. Thanks. Turns out that the metrics were rolled up avg vs max. So in fact were were hitting 100% active thread when the queue size started to accumulate.

1 Like

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.