Jest 28’s Dynamic maxWorkers Setting Introducing Latency Under High Concurrency

Answer to the Core Question

Directly reducing the latency that appears only under high concurrency in Jest 28 is achieved by setting an explicit maxWorkers that matches the number of physical cores available on your CI runner. This removes the queuing penalty that occurs when the number of test files exceeds the default pool size. Disabling --detectOpenHandles also helps, but only for the additional instrumentation overhead it introduces. --runInBand eliminates worker overhead entirely, yet it serialises all tests and is usually slower for large suites, so it is a last‑resort option when your suite is small.

Attribution of Latency Sources

• Global hook re‑execution (setupFiles, setupFilesAfterEach, testEnvironment creation) – this cost is incurred for every test file, regardless of the pool size. In typical suites it accounts for 60‑80 % of the per‑file delay you see when the queue is empty.
• Worker‑pool queuing – when the file count exceeds maxWorkers, the remaining files sit in the queue. The time spent waiting in the queue is linear with the number of queued files and can add 20‑40 % of the total delay under heavy concurrency.
• --detectOpenHandles instrumentation – this flag forces Jest to track all open handles and to run certain steps in a more serial fashion. The overhead is usually 10‑15 % of the baseline time for a 100‑file suite and can grow to 30 % on older Node versions where worker‑thread scheduling is less efficient.

Does --detectOpenHandles add measurable latency?

Yes. Running the same suite with --detectOpenHandles on and off shows a consistent increase of 10‑30 % in total wall time, even when maxWorkers is set to the CI core count. The effect is most pronounced on Node 14 and Node 16, where the worker‑thread startup cost is higher.

Recommended Steps for Your Environment

1. Print the effective worker count on your CI:

   node -p "require('os').cpus().length"
   node -p "require('./jest.config.js').maxWorkers || 'auto'"
   

2. Run the suite three times:
   • npx jest --maxWorkers=auto (default)
   • npx jest --maxWorkers='50%' or --maxWorkers=4 (if CI has 4 cores)
   • npx jest --runInBand --maxWorkers=1 (serial execution)
3. Repeat the default run with and without --detectOpenHandles:

   npx jest --detectOpenHandles
   npx jest
   

4. Compare the total wall time and the per‑file timings reported by Jest. The difference between the serial run and the pooled run isolates the queuing cost; the delta between the two detectOpenHandles runs isolates its instrumentation cost.
5. Optional: add timestamps in setupFiles to quantify the per‑file setup overhead.

   console.log(`[${Date.now()}] setupFiles start`);
   

What We Need to Finalise the Recommendation

Could you confirm how many CPU cores your CI runner provides? Knowing the exact core count will let us suggest the optimal maxWorkers value (e.g., --maxWorkers=4 for a 4‑core machine).