Skip to main content
Use this guide to run performance benchmarks consistently, compare cold vs warm behavior, and collect reproducible numbers.

​
What to run

isol8 includes multiple benchmark suites with different goals:
  • bun run bench for cold container spawn latency
  • bun run bench:pool for warm-pool steady-state latency
  • bun run bench:detailed for phase-level timings
  • bun run bench:tti for ComputeSDK-style cold TTI
  • bun run bench:tti:pool for ComputeSDK-style warm TTI

​
Local benchmark workflow

  1. Use a quiet machine and close heavy background processes.
  2. Run each suite sequentially (avoid parallel benchmark runs).
  3. Repeat runs and compare medians, not single outliers.
bun run bench
bun run bench:pool
bun run bench:detailed
bun run bench:tti
bun run bench:tti:pool

​
GitHub Actions benchmark workflow

You can run benchmarks on a GitHub-hosted Linux runner by manually triggering Production Tests (Manual).
  • workflow input: runBenchmarks=true
  • runner: ubuntu-latest
  • benchmark step runs only if production tests pass
  • benchmark command in workflow: bun run bench:cli

​
Reading results correctly

  • Compare warm and cold separately.
  • Use bench:detailed to locate where time is being spent (create, start, run, cleanup).
  • For cross-project comparisons, align environment first (host class, architecture, runner type, and iteration count).

​
Practical regression gate

Use warm TTI and warm-pool averages as primary gates for interactive use-cases, and track cold numbers as secondary guardrails.