Stress testing

Stress testing is meant for concurrency bugs, especially race conditions in FFI / C integrations.

Run it with:

acton test stress

How stress mode runs

• Runs one test function/actor at a time.
• For that test, starts multiple concurrent workers of the same test in one process.
• Worker count defaults to roughly 1.5 * nr_wthreads so workers must share RTS worker threads.
• Override it with --stress-workers N when you want a specific level of oversubscription.
• Stress runs are always fresh (no test-result cache reuse).
• By default, stress runs for up to 5 seconds per test (--max-time 5000).
• Continuous mode is available with --max-time 0.
• In continuous mode, --max-iter is unbounded unless you set it.
• The default stress --min-time is 1 second unless overridden (used for calibration; stress run length is controlled by --max-time).

Worker scheduling

Stress workers are split into:

• A no-drift cohort (at least 2 workers, scaling to roughly 25% of workers)
• A staggered-drift cohort (small per-iteration microsecond offsets)

This combines synchronized overlap windows with evolving offsets over time.

Per-worker live stats

In an interactive terminal, stress mode shows one live status line per worker.

• sync worker line: wN sync RUN/DONE ... @ RATE/s cur=0us tot=0us
• drift worker line: wN drift RUN/DONE ... @ RATE/s cur=DRIFTus tot=TOTALus

The main test line also carries a compact worker summary:

• total: ... RUNS runs in DURATIONms @ RATE/s
• worker mix: workers=N (sync=S drift=D)
• iteration estimate, granularity and observed coverage: iter~...ms coarse~...ms sweep=... calib=... cov=SEEN/TOTAL(PCTpct)
  • iter~ is a moving estimate of one test iteration duration
  • coarse~ is the current phase coarseness (lower is finer)
  • in continuous mode (--max-time 0), coarse~ decreases stepwise over time as sweep depth increases
  • cov is observed occupied phase bins at current coarseness (resets when coarseness is refined)

Example:

racy_sum: RUN : 2400 runs in 1500.000ms @ 1600.0/s | workers=8 (sync=2 drift=6) iter~2.340ms coarse~0.037ms sweep=64 cov=37/64(57.8pct)

Use normal test flags to tune duration/iterations, for example:

acton test stress --max-time 30000 --min-iter 100

Pin a higher worker count explicitly:

acton test stress --stress-workers 24 --max-time 30000

Run continuously until interrupted:

acton test stress --max-time 0

Press Ctrl-C to stop and print the partial stress result collected so far.