Test execution

Run the complete host/C# bank from the host repository:

bash test/scripts/run-all.sh

Run the complete Rust target bank from tsonic-rust:

npm test

Both use the host’s shared resource-budget and bounded-process runner. CPU affinity, CPU quotas and available memory determine the worker count. The budget leaves room for the desktop and coordinator, bounds native child builds, disables swap for the test scope, and retains complete logs under the running repository’s .temp/test-runs/. A worker count is not a limit on all native threads.

Configuration

Variable	Meaning
TSONIC_TEST_CPUS	Effective CPU budget; default 85% of available CPUs, at least one
TSONIC_TEST_MEMORY_MIB	Total test memory ceiling in MiB
TSONIC_TEST_WORKERS	Concurrent test files/tasks; must fit CPU and memory budgets
TSONIC_TEST_WORKER_MEMORY_MIB	Memory reservation per worker
TSONIC_TEST_CHILD_JOBS	Cargo jobs and .NET processor budget per native child
TSONIC_TEST_HEAP_MIB	Node heap limit per worker, below its memory reservation
TSONIC_TEST_TIMEOUT_SECONDS	Complete-run deadline; default 7,200 seconds
TSONIC_TEST_HEARTBEAT_SECONDS	Progress interval; default 60 seconds
TSONIC_TEST_TASKS_MAX	Scope process/thread ceiling; default 2,048
TSONIC_TEST_LOG_SIZE_MAX_BYTES	Complete-log ceiling; default 64 MiB

For example, use twelve host workers within a 32 GiB ceiling:

TSONIC_TEST_CPUS=16 TSONIC_TEST_MEMORY_MIB=32768 TSONIC_TEST_WORKERS=12 \
  bash test/scripts/run-all.sh

Native integrations need more resources than lightweight policy tests. Rust’s default reservation is 4 GiB per worker with two Cargo jobs; the host default is 2 GiB with one native child job. Increase reservations when observed peaks require them. Invalid or unbudgetable selections fail before tests begin.

The host schedules independent groups together, but retains explicit exclusive tasks and per-group limits. Package unit modules run separately; discovery and reporting must account for each authored test file exactly once. A full suite collects all failures. Do not start a subsequent suite while known failures remain; an aborted or filtered run is never full certification.

Short utilization calibration

Calibration is an explicitly partial run, not a way to certify a subset:

TSONIC_TEST_CALIBRATION_SECONDS=120 \
TSONIC_TEST_CALIBRATION_WARMUP_SECONDS=60 \
TSONIC_TEST_MIN_CPU_PERCENT=60 \
  bash test/scripts/run-all.sh

After warmup, the runner measures the entire test cgroup’s CPU consumption over the selected window, writes a .calibration.json report, and stops the scope. The percentage is relative to the assigned CPU budget: 60% of twenty CPUs means twelve CPU-seconds per wall-clock second. Exit 125 means a completed calibration or an insufficient window; inspect the report to distinguish them. Exit 126 means the utilization target was missed. Neither is a certification pass. Test failures observed during calibration still need investigation.

Use the same variables with Rust’s npm test. Remove calibration variables for the full run. Keep network, shared-state and performance-sensitive test isolation; do not weaken assertions or remove checks to raise CPU utilization.