perf: amortize per-sample startup and post comments on fork PRs

[JacksonWeber]

What

Two perf-CI improvements, both motivated by the recent perf-regression job taking ~14 minutes per PR.

1. Make perf tests faster

Diagnosis from a recent run: 6 scenarios × 3 samples = 18 child Node processes, each paying ~14.7s of pure overhead (Node startup + importing applicationinsights' large dependency tree + useAzureMonitor() init) for only 6s of actual measurement.

Fix: refactor bench.mjs to accept --samples N and take all N samples back-to-back inside one child process (warmup runs once, then N timed durations). runBenchmarks.mjs now spawns 1 child per scenario instead of samples × scenarios. OpenTelemetry global-state isolation is preserved because each scenario still gets a fresh process.

Expected savings: ~3 min off each of the two Run X benchmarks steps.

2. PR comments on fork PRs

The previous inline sticky-pull-request-comment step was gated on head.repo.full_name == github.repository so fork PRs never received a comment.

Modeled on microsoft/opentelemetry-distro-python's split:

• performance.yml now runs without pull-requests: write and uploads baseline.json, candidate.json, perf-comparison.md, pr-number.txt as the perf-results artifact.
• performance-comment.yml (new) triggers via workflow_run after Performance completes, runs in the trusted base-repo context with pull-requests: write, downloads the artifact, validates the PR number is a positive integer, regenerates report.md from the JSON using the base branch's own trusted comparePerf.mjs (so attacker-supplied markdown is never posted under the writable token), and posts the sticky comment.

3. Docs

test/performanceTests/README.md updated to describe the workflow split (with security model) and the single-child-per-scenario speedup.

Verification

• Locally verified bench.mjs + runBenchmarks.mjs end-to-end with --samples 2 --duration 1 --warmup 1 --scenarios OtelLogTest,OtelSpanTest: produces the expected JSON shape and comparePerf.mjs consumes it unchanged.
• Both workflow YAML files parse cleanly.

[Copilot]

Pull request overview

This PR improves performance CI by reducing benchmark runtime overhead and enabling performance comparison comments to be posted on fork-based pull requests in a safer, split-workflow model.

Changes:

• Refactors the perf harness to take multiple samples per scenario within a single child process (bench.mjs + runBenchmarks.mjs) to amortize Node/module initialization costs.
• Splits PR commenting into a new workflow_run workflow that runs in the base-repo context, downloads perf artifacts, regenerates markdown from JSON using trusted code, and posts a sticky PR comment.
• Updates perf-test documentation to describe the workflow split and security model.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
test/performanceTests/runBenchmarks.mjs	Passes --samples through to the bench child and aggregates per-sample ops/s into summary stats.
test/performanceTests/bench.mjs	Collects N back-to-back timed samples in one process and outputs a samples[] JSON shape.
test/performanceTests/README.md	Documents the CI workflow split and rationale/security model.
.github/workflows/performance.yml	Removes PR write permissions and uploads perf outputs + PR number as an artifact for follow-on commenting.
.github/workflows/performance-comment.yml	New workflow to download artifact in base context, regenerate report from JSON, and post sticky PR comment.

---

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