GitHub user polyzos added a comment to the discussion: Some questions on
LanceDB Lakehouse Integration
@vamossagar12 You're correct that vector index creation is not currently
configurable through Fluss. The LanceDB integration today is a storage/tiering
layer and it handles writing Fluss log records into Lance datasets and atomic
fragment appends, but does not invoke any of LanceDB's index creation APIs.
As it stands, users would need to create indexes separately through LanceDB's
API after the table is created via Fluss. I agree this creates friction. A
natural improvement would be to expose index configuration through table
properties (e.g., `lance.index.type`, `lance.index.metric`,
`lance.index.num_partitions`) and have Fluss call LanceDB's index creation API
as part of table creation. This is something we should track for improvement.
The other two are also not supported.. In general we should strengthen more our
vector/lance support and we need to keep track of what can be improved and we
welcome new idea, so thanks all lot for sharing all the above.
GitHub link:
https://github.com/apache/fluss/discussions/2877#discussioncomment-16184232
----
This is an automatically sent email for [email protected].
To unsubscribe, please send an email to: [email protected]
