Pavel Borisov

Hi @ankane , Commit 3df5655b305f03 PG16.2 in the previous comment results are for gcc11, now I added measurements for GCC13 ``` gcc -v Target: aarch64-linux-gnu Configured with: ../src/configure -v --with-pkgversion='Ubuntu...

TL;DR The feasible way is to have filtering data in the same index with vectors. @jkatz I may be wrong but I suppose that the main problem of using bitmap...

@jkatz It's an interesting idea! Though there is one thing that I don't quite catch: how to combine "nearness" in two different spaces. Furthermore, nearness in embedding HNSW component is...

@jkatz > It may expand the search radius, but we may not be seeing the best available vectors for a filter. For example, let's say that vectors with a specific...

I suppose the case when post-filtering depletes all (or most) of ann-found tuples is completely legit. Though considering how often are the related complaints I'd suggest it to be mentioned...

Can the syntax like `select * from emb_table where xx with (ivfflat.probes=10)` as mentioned in a comment above be implemented inside the extension at all? Doesn't it need modification of...

I consider massive updates/deletes for pgvector a rare use case. It could be recommended to set ef_search with overhead to the number of tuples in the limit clause rather than...

Maybe it's related to commit 67681306. Invalidation in recovery of descrs (o_invalidate_descrs()) that clears hash entry is added and commented out at the same time.

https://github.com/orioledb/orioledb/tree/inval-descr-recovery seems to fix this invalid read in Valgrind1 test in CI

at [02bc078](https://github.com/orioledb/orioledb/commit/02bc078a30f71940500be7eb63ff86badb17a3fa) Bucket size increased from 18.1Gb to 26.1 GB (1.4x) over 16 start-stops with the number of files 180000 I.e. increase of s3 bucket is lower than previously (1.4x...