qdrant-scaling-query-volume

You can reduce inter-shard data transfer in Qdrant by applying a per-shard sub-sampling scheme that asks each shard for a smaller limit and merges results to achieve the requested total limit.

Per-shard sub-sampling in Qdrant is a strategy that asks each shard for a smaller result limit and merges them, leveraging a Poisson-based safety factor to limit data transfer while preserving result quality and overall latency.

Sub-sampling is applicable when queries hit high limits on auto-sharded collections with a common shard key where exact results are not strictly required, relying on random independent data distribution to keep results representative.

You can scale Qdrant query volume across shards by implementing a Poisson-based sub-sampling strategy with a safety factor, which limits inter-shard data transfer to preserve result quality and maintain overall query latency.

You should not use sub-sampling when exact vector search results are strictly required, as the strategy relies on auto-sharding with random data distribution to approximate representative results rather than guaranteeing precise accuracy.