Query performance- filter aggregation with the same filter improve performance drastically

[alexs1]

Can anyone explain why the left (pseudo) query runs above 5s the while the right part in 30ms?

The filters in the example below, are identical (aggs/ query filters).

isn't the aggregation part suppose to run on filtered data? (result value is the same)

[spinscale (Alexander Reelsen)]

Which Elasticsearch version are you running on?

Also please make sure that the request cache does not skew your calculations and try both queries with that cache disabled.

[alexs1]

thanks @spinscale

version: 7.13.2
added request_cache=false to both ..

[nik9000 (Nik Everett)]

I think you are hitting a bug I caused in 7.13. This is the fix, coming in 7.14: https://github.com/elastic/elasticsearch/pull/74260

When we found the bug I scrambled to get an emergency "off switch" into the system for 7.13.2 so you can try it: https://github.com/elastic/elasticsearch/pull/73620

Sorry for the trouble!

[ustrauss (Uri Strauss)]

We hit the same bug on version 7.13.2, it took us 2 weeks to understand where it's coming from and how to work around it. Due to high latency, it was crashing the cluster on specific reports that were generated by users.