Context

I am using the motor for async pipeline aggregation to perform vector search.
It shows a weird behavior, when a new query is used for vector search, it takes too much time around 10 to 180 seconds sometimes, if I send the same query again sequentially it takes less than a second. 

The performance for the queries seen before is high and for new queries it is dramatically bad.

I am using the latest version of python and python-motor async client.

My atlas vector database is hosted at an Atlas local deployment on Linux machine having 8 GB RAM and only the single atlas container on that machine.

My vector field have 1024 embedding size (dunzhang/stella_en_400M_v5) is used to create embeddings. And having 2 filters in vector index (type = int). using limit_users=5000 during the search.

Definition of done

Test the problem and find out that the issue is with motor or Atlas search.

Resolve the issue as soon as possible. It is a severe performance issue.

Pitfalls

There are no risks.