Is the full scan threshold on a per segment/shard/collection basis?

[fhalde]

I am trying to determine what's the cause of my queries taking 30-60s to complete

The database is quite small in that it could almost fit in the provisioned 32Gi memory (per node)

The collection has a total of 30k points with dense and sparse vectors totalling to 60k vectors. Split across 12 shards, 24 segments and replication factor of 3 spread across 3 nodes with 4vCPU and 32Gi Memory

Indeed the shard & segments are over provisioned but would it add this much latency? The RPS is 3 req/s and we use filters in queries. I can confirm we only filter on fields that have a payload index

How would you recommend I approach debugging this? Happy to share more details

[fhalde]

Qdrant version 1.15.1

[joein]

Hey @fhalde
Could you provide an example of your request?

[timvisee]

Also, are you using Qdrant Cloud/Docker deployment, or are you using local mode in Python?