Rebuild absent embeddings from source passages to support model changes
This CL schedules embeddings to be recomputed for any rows in the passages table that don't have corresponding rows in the embeddings table, keyed on url_id. This rebuild process happens once when the SQL database is initialized, and partial progress can lead to full recovery even when interrupted by shutdown. Rebuilds can be disabled by forcing the "RebuildEmbeddings" feature parameter to false, and model version changes can be simulated to delete the embeddings data by forcing the "DeleteEmbeddings" feature parameter to true. Models can change any time, even without deploying a new Chromium binary, and without this CL the history data would not be available for embeddings search, but with this CL the history data gets rebuilt to become searchable. Bug: 345798663 Change-Id: I46a45d5d8d55ac6a38b70df985044b610f23abef Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5606010 Reviewed-by:John Lee <johntlee@chromium.org> Code-Coverage: findit-for-me@appspot.gserviceaccount.com <findit-for-me@appspot.gserviceaccount.com> Commit-Queue: Orin Jaworski <orinj@chromium.org> Reviewed-by:
Orin Jaworski <orinj@chromium.org> Cr-Commit-Position: refs/heads/main@{#1312838}
Loading
Please register or sign in to comment