Increased QPS erratic CPU behavior

Hi !

I'm actually testing ElasticSearch 0.90.x and go very
erratic behavior while increasing query rate. We are running many
benchmarks to push the 2 nodes cluster to the limit, with our standard
query which is consistent mainly of a querystring + terms filter.

Basically, when we run the cluster at a query rate of <26 everything seems
fine, the cpu usage is at ~50%, no iowait at all and pretty decent response
time (~150ms). But thats when we go up to 27 QPS, the CPU usage seems to go
through the roof (100%), still no IO wait but the response time goes up to

While doing some CPU profiling, we noticed that the FixedBitSet methods
(or, and, ) usage seems to have grown a lot (about 10x more). The 2
cpu snapshots are attached to this message.

So, how come that an increase of 1 query per second can completely
overwhelm the cluster like that ? There is nothing else changed, only an
increase of the rate into our benchmark. This seems like a weird behavior
and it's kind of blocking at the moment.

Jérôme Gagnon

You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to
For more options, visit