MB-69881: Improved APIs and perf optimizations for vector search #2270
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
CascadingRadium
requested review from
Likith101,
Thejas-bhat,
abhinavdangeti,
capemox,
Copilot and
maneuvertomars
Contributor
There was a problem hiding this comment.
Pull request overview
This PR re-architects vector search to improve memory efficiency and reduce garbage collection pressure. The changes replace slice-based eligible document tracking with bitsets, achieving up to 64× memory reduction per segment, and optimize the iterator pattern to eliminate per-call allocations in the unadorned postings iterator.
Key changes:
- Replaced slice-based eligible document tracking (
[]uint64) with bitsets, reducing memory from8N bytestoN/8 bytesper segment - Introduced iterator-based API for eligible documents that directly translates to bitset iteration at the storage layer
- Fixed garbage creation in
UnadornedPostingsIteratorby reusing a single struct instance instead of allocating perNext()call - Optimized bytes read tracking to skip computation in no-scoring mode
Reviewed changes
Copilot reviewed 8 out of 8 changed files in this pull request and generated 4 comments.
Show a summary per file
| File | Description |
|---|---|
| index/scorch/snapshot_vector_index.go | Introduces bitset-based eligible document storage and iterator API, replacing the previous slice-based approach |
| index/scorch/unadorned.go | Changes UnadornedPosting from uint64 to struct with pointer receivers and adds reusable struct fields to iterators to eliminate per-call allocations |
| index/scorch/snapshot_index_tfr.go | Adds conditional bytes read tracking via updateBytesRead flag to skip computation in no-scoring mode |
| index/scorch/snapshot_index.go | Initializes updateBytesRead flag based on scoring requirements |
| index/scorch/optimize_knn.go | Removes requiresFiltering flag and updates to use new SegmentEligibleDocuments API |
| index/scorch/optimize.go | Sets updateBytesRead to false for unadorned term field readers |
| index/scorch/snapshot_index_vr.go | Updates InterpretVectorIndex call to remove filtering parameter |
| index_test.go | Updates expected bytes read values to reflect the optimization that skips unnecessary computation |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
abhinavdangeti
changed the title
Contributor
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
abhinavdangeti
previously approved these changes
Jan 12, 2026
abhinavdangeti
approved these changes
Jan 12, 2026
abhinavdangeti
changed the title
Likith101
approved these changes
Jan 13, 2026
CascadingRadium
added a commit
that referenced
this pull request
Jan 20, 2026
- Use a `bitset` to track eligible documents instead of a slice of `N uint64s`, reducing memory usage from `8N bytes` to `N/8 bytes` per segment (up to `64×` reduction) and improving cache locality. - Pass an iterator over eligible documents that iterates the bitset directly, allowing direct translation into a bitset of eligible vector IDs in the storage layer and eliminating the need for a separate slice intermediary. - Fix garbage creation in the `UnadornedPostingsIterator`, which previously allocated a temporary struct per Next() call to wrap a doc number and satisfy the `Postings` interface; the iterator now returns a single reusable struct (one-time allocation) consistent with the working of the `PostingsIterator` in the storage-layer. - Avoid unnecessary `BytesRead` statistics computation when executing searches in no-scoring mode, removing redundant work as a micro-optimization. --------- Co-authored-by: Abhinav Dangeti <[email protected]>
CascadingRadium
added a commit
that referenced
this pull request
Jan 21, 2026
- Backport the following commits into `v8.0.x-couchbase` via cherry-pick
- #2270
- #2224
- #2272
---------
Co-authored-by: Abhinav Dangeti <[email protected]>
Co-authored-by: Copilot <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
bitsetto track eligible documents instead of a slice ofN uint64s, reducing memory usage from8N bytestoN/8 bytesper segment (up to64×reduction) and improving cache locality.UnadornedPostingsIterator, which previously allocated a temporary struct per Next() call to wrap a doc number and satisfy thePostingsinterface; the iterator now returns a single reusable struct (one-time allocation) consistent with the working of thePostingsIteratorin the storage-layer.BytesReadstatistics computation when executing searches in no-scoring mode, removing redundant work as a micro-optimization.