Benchmark

[danielebarbaro]

@ezimuel I’m just trying to understand how to run a more effective performance benchmark.

- this is a "simple"draft

• this branch comes from feat/update-delete

[Copilot]

Pull request overview

This PR adds document delete/update support to the core VectorDatabase (including persistence behavior), and introduces a new unified benchmark runner plus a GitHub Actions workflow to run/compare benchmarks.

Changes:

• Add deleteDocument() and updateDocument() to VectorDatabase, including persistence + tests and documentation.
• Implement soft-delete tracking in the HNSW index and a BM25 removal API to keep text search consistent after deletes.
• Replace the legacy benchmark scripts/report generator with a new benchmark runner (benchmark/run.php) and add CI benchmarking workflow.

Reviewed changes

Copilot reviewed 14 out of 14 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
src/VectorDatabase.php	Adds delete/update APIs, persists deleted-node state, and changes count() to active-doc count.
src/HNSW/Index.php	Adds soft-delete set, filters deleted nodes from search results, and persists deleted IDs via export/import state.
src/BM25/Index.php	Adds removeDocument() for delete/update support.
tests/VectorDatabaseTest.php	Adds unit tests for delete/update behavior in memory.
tests/PersistenceTest.php	Adds persistence tests for delete/update and doc-file removal expectations.
README.md	Documents the new delete/update APIs and persistence expectations.
composer.json	Moves PHPVector\\Benchmark\\ autoloading into production autoload (needed for --no-dev installs).
benchmark/run.php	New benchmark CLI entrypoint (profiles/scenarios/output formats).
benchmark/Benchmark.php	New benchmark implementation (insert/search/update/delete/recall/persistence).
benchmark/ResultFormatter.php	New Markdown + GitHub-benchmark JSON formatting.
benchmark/Report.php	Removes legacy report generator.
benchmark/benchmark.php	Removes legacy benchmark entrypoint.
.github/workflows/benchmark.yml	Adds benchmark workflow for PR comparison + main history storage.
.github/workflows/test.yml	Bumps cache/codecov GitHub Action versions.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[ezimuel]

I left a minor comment on a CI settings for deployments.
Moreover, I would like to understand why we are using markdown as output for the benchmark. I checked the output on Actions and it doesn't seem to render markdown, see here. Maybe, we can create a comment on the PR with the benchmark output or do you have other proposals? Thanks.

[danielebarbaro]

@ezimuel you’re right, Markdown isn’t rendered in the Actions output.
I was just using it to read the results more comfortably 😇 .

I’m now working on posting the benchmark results as a comment on the PR, with an "automatic comparison" against the base branch.