Separate concurrency benchmark lifecycle measurements

[reneklacan]

Summary

• split concurrency benchmarks into end-to-end and steady-state suites
• prestart and gate steady-state workers so startup and shutdown stay outside throughput timing
• increase high-concurrency workloads until samples take roughly 2-5 seconds
• report item throughput and extend the CI timeout for the larger repeated suite

Measurements

High-concurrency cases now take approximately 2.2-3.7 seconds locally, up from roughly 0.3-0.5 seconds.

Verification

• cargo fmt --all --check
• cargo test --workspace (200 passed, 3 ignored)
• cargo clippy --all-features --workspace --all-targets -- -D warnings
• repeated steady-state benchmark samples with no queued or in-flight jobs left behind
• workflow YAML parsing and git diff --check

---

Note

Low Risk
Changes are limited to benchmark harness code and CI timeout; no production queue/runtime behavior is modified.

Overview
Reworks oxana concurrency Divan benchmarks so throughput numbers separate full runtime lifecycle from steady-state processing, and CI can finish the heavier suite.

End-to-end benches still time starting the runtime, draining the queue, and shutdown. Steady-state benches pre-start workers, block the first worker cohort until setup completes, then only time the gated job batch; shutdown and RunStats checks run in SteadyStateCleanup after Divan stops the timer. Cases are parameterized via BenchmarkCase (worker count + job count), with larger job counts at high concurrency so samples run ~2–5s instead of sub-second.

Benchmark setup uses enqueue_list instead of per-job enqueue, reports items/sec via Divan ItemsCount, and asserts success counts through shared assert_stats. The Test workflow job timeout rises from 10 to 30 minutes to match repeated cargo bench in CI.

^{Reviewed by Cursor Bugbot for commit 79dd9f3. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

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

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

^{Reviewed by Cursor Bugbot for commit 79dd9f3. Configure here.}

[cursor]

Measure path can hang forever

Medium Severity

measure waits only on the finished semaphore and never watches the runtime task. prepare already uses select! so a setup-time runtime exit fails fast, but if the runtime exits during the timed section before every job calls record_completion, wait_until_finished blocks forever. Cleanup never runs, so CI sits until the 30-minute timeout instead of surfacing the runtime failure.

Additional Locations (1)

• oxana/benches/concurrency.rs#L314-L320

 

^{Reviewed by Cursor Bugbot for commit 79dd9f3. Configure here.}