#4089 chore(bench): upgrade CodSpeed for Node 24

Summary (AI generated)

CI uses Node 24 with CodSpeed 5.7.1, whose report warns that this runtime is untested. Upgrade @codspeed/vitest-plugin and its core to 6.1.1, which supports Node 24. Benchmark cases, product source, the CodSpeed workflow and Node version are unchanged.

Also fix the serial backend-plugin job's bootstrap overlap after bun install was killed with exit 137 while Supabase started in parallel. Finish image extraction, dependency installation and database startup sequentially. Capture resource diagnostics after bootstrap failure and skip log upload when Edge startup never produced a log path. The kernel kill reason was not captured; this is not a proven OOM diagnosis.

Merged main through c09ee7ca0f, retaining its release versions and preproduction test repairs. The final diff is package.json, bun.lock and this one test-job bootstrap change.

Motivation (AI generated)

This separates measurement-tool calibration from the KV rollout in #4038. The SDK 6.0 release documents a one-time simulation calibration change on Node 24. SDK 6.1.1 further changes V8 compilation during analysis.

The initial report on da11021c03 flags three regressions, CRC32, semver validation and updater support windows, plus software/SDK-major differences. Independent source review confirmed all three benchmark bodies, implementations and their functional dependencies match main. These cross-SDK differences are inconclusive as product performance evidence. Acknowledgements scoped to this SDK-migration report remain pending; tracking and thresholds are unchanged.

After this lands, the main push can collect a supported SDK 6 baseline. Refresh #4038 after that baseline completes, using matching measurement settings. CodSpeed baseline selection.

Business Impact (AI generated)

Restore supported benchmark instrumentation for the CI runtime and keep calibration changes separate from claims about KV latency or cost. No product performance improvement is claimed. The independent local DB/KV HTTP measurements remain historical evidence in #4038.

The bootstrap change removes concurrent installer/database resource contention in the observed failing job so its tests can run.

Test Plan (AI generated)

• Frozen Bun install, backend lint/typecheck and diff checks on the final changes.
• YAML validation: all other jobs, job limits, test commands and sharding remain unchanged; image extraction/install/start are sequential; empty log upload is guarded.
• All 8 release-workflow unit tests passed.
• All 60 focused MCP/write-guard/preproduction-isolation tests passed on preceding head da11021c03.
• All 43 benchmarks across five files completed under Node 24.22.3 with plugin 6.1.1 setup/teardown before the main refreshes. This is a local compatibility smoke test, not an instrumented CPU comparison.
• Existing serial-job logs confirm install exited 137, Supabase started successfully and plugin tests never ran. The artifact upload also failed because its path was empty; the guard fixes that secondary failure.
• All 47 other checks passed on 92545f962a, with 26 skipped matrix/template entries. The serial backend-plugin job now runs and passes after the bootstrap change; backend/Cloudflare/CLI/Playwright/lint/type/SQL execution gates are green. Socket reports no new dependency alerts; Sonar passed with zero new issues and zero security hotspots. CodSpeed's calibration comparison is the sole remaining failed check.
• Review and acknowledge only this SDK migration's calibration findings. A matching SDK 6 main baseline and refreshed feature-PR run remain required before attributing CPU differences to product code.
• Automatic review after CI is clear.

Generated with AI

---

^{Open in [code]smith · Need help on this PR? Tag @codesmith-bot with what you need.}