You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Previously it was impossible for a doc id of the initial insert into HNSW to be larger than its size. But since we added the ability to add objects without a vector and introduce the vector later it is now possible to import more than 25,000 objects before ever making use of the HNSW index. In that case, we now need to check if the initial size is large enough and adjust it accordingly.
Goals
The initial insert will work with any doc id regardless of initial index size
Add tests to prevent regression
Notes
While we can reproduce this with a stress test, I think we should also be able to reproduce this with a very simple integration test in the adapters/repos/db/vector/hnsw package. A simple insert with doc id 25,000 on an empty index should already reproduce this.
The text was updated successfully, but these errors were encountered:
Problem
Previously it was impossible for a doc id of the initial insert into HNSW to be larger than its size. But since we added the ability to add objects without a vector and introduce the vector later it is now possible to import more than 25,000 objects before ever making use of the HNSW index. In that case, we now need to check if the initial size is large enough and adjust it accordingly.
Goals
Notes
While we can reproduce this with a stress test, I think we should also be able to reproduce this with a very simple integration test in the
adapters/repos/db/vector/hnsw
package. A simple insert with doc id25,000
on an empty index should already reproduce this.The text was updated successfully, but these errors were encountered: