[SVS][PoC] 2-steps vectors update in SVSTiered

[rfsaliev]

Proof-of-concept for 2-steps SVS index update

This PR is the PoC to demonstrate potential Tiered SVS index update performance improvements by using 2-steps index update capability implemented in intel/ScalableVectorSearch#348.

Which issues this PR fixes

1. #...
2. MOD...

Main objects this PR modified

1. ...
2. ...

Mark if applicable

• This PR introduces API changes
• This PR introduces serialization changes

---

Note

Medium Risk
Changes tiered update locking and SVS search/add integration against new prebuilt libraries; incorrect two-phase ordering or ID translation could affect correctness under concurrent queries.

Overview
PoC for shorter exclusive locks during tiered SVS flushes by splitting backend inserts into compute-then-commit using SVS add_points_compute_changes / add_points_commit, wired through new SVSIndexBase::makeAddVectorsChanges and applyAddVectorsChanges.

TieredSVSIndex::updateSVSIndex now runs vector preprocessing and change computation while holding only a shared mainIndexGuard, then upgrades to an exclusive lock solely for applyAddVectorsChanges instead of calling addVectors under the writer lock for the whole operation.

For non-multi SVS indices, single-query TopK avoids the batch QueryResult search API and uses MutableVamanaIndex scratchspace search with translate_internal_id on neighbors.

Build: cmake/svs.cmake points prebuilt SVS downloads at nightly pr330 tarballs, collapses Clang libstdc++ selection to GCC 11+, drops the glibc 2.26 GCC binary path, and raises the minimum glibc for GCC prebuilts to 2.28.

Benchmarks: adds BM_TopKSearchDuringUpdate to stress parallel TopK queries during a background tiered update.

^{Reviewed by Cursor Bugbot for commit c81cfa8. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Backend deletes under shared lock

High Severity

updateSVSIndex calls makeAddVectorsChanges while holding only a shared mainIndexGuard lock, but that path still runs deleteVectorsImpl (via impl_->delete_entries) before the exclusive lock and applyAddVectorsChanges. Concurrent backend searches also use shared locks, so they can observe deleted labels and a mutating graph before new points are committed.

Additional Locations (1)

• src/VecSim/algorithms/svs/svs.h#L286-L291

 

^{Reviewed by Cursor Bugbot for commit ba22610. Configure here.}

[codecov]

Codecov Report

❌ Patch coverage is 91.80328% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 97.05%. Comparing base (45daaf3) to head (c81cfa8).
⚠️ Report is 2 commits behind head on main.

Files with missing lines	Patch %	Lines
src/VecSim/algorithms/svs/svs.h	91.22%	5 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #988      +/-   ##
==========================================
- Coverage   97.12%   97.05%   -0.07%     
==========================================
  Files         141      141              
  Lines        8164     8225      +61     
==========================================
+ Hits         7929     7983      +54     
- Misses        235      242       +7     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

There are 2 total unresolved issues (including 1 from previous review).

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit c81cfa8. Configure here.}

[cursor]

Index deletes under shared lock

High Severity

makeAddVectorsChanges calls deleteVectorsImpl, which mutates the SVS backend via delete_entries and markIndexUpdate. Tiered updateSVSIndex invokes this while holding only a shared mainIndexGuard, so concurrent topKQuery paths can read the same impl_ and race with those writes.

Additional Locations (1)

• src/VecSim/algorithms/svs/svs_tiered.h#L715-L717

 

^{Reviewed by Cursor Bugbot for commit c81cfa8. Configure here.}