I’m unhappy with Qdrant filtering performance

That’s understandable—filtering performance in Qdrant can vary a lot depending on how selective the filter is, how many points you have, and your indexing configuration.

To help you improve it, the main things to check are:

1. Filter selectivity

   • If the filter matches a large portion of the collection, Qdrant still has to do more work.
   • Very broad filters often perform much worse than highly selective ones.

2. Payload indexing

   • Make sure the payload fields used in filters are indexed.
   • In Qdrant, good payload indexing is often the difference between fast and slow filtering.

3. Collection size and shard layout

   • Large collections, especially with many shards or high write churn, can affect filter latency.
   • If your workload is skewed, shard distribution may matter.

4. Vector search + filter interaction

   • If you’re doing ANN search with a filter, the engine may need to search more candidates to satisfy the filter.
   • Tight filters can help; medium-selectivity filters can sometimes be worst because they reduce candidate pool without enough pruning.

5. Condition structure

   • Complex nested conditions, many OR clauses, and large should arrays can slow things down.
   • Simpler filters usually perform better.

6. Disk vs memory pressure

   • If the node is under memory pressure or storing too much on disk, latency can worsen.

If you want, I can help you diagnose it concretely. If you paste:

• your filter JSON,
• collection size,
• which fields are indexed,
• and an example latency you’re seeing,

I can tell you what’s likely causing the slowdown and how to fix it.