Load and failure testing

scripts/stress.sh puts a real nyxd under load and through failures, then checks that it recovers and does not leak. It runs in CI (the Stress workflow: weekly, on demand, and on pull requests that touch the supervisor, logs or runtime), rootful and rootless, and writes its measurements to the job summary.

make build build-nyx
sudo scripts/stress.sh        # rootful: every scenario
scripts/stress.sh             # rootless: everything but the OOM scenario

Scenarios

Scenario	What it does	Fails when
Many at once	Starts `STRESS_CONTAINERS` (50) long-running containers from `STRESS_PARALLEL` (8) clients, execs into some, removes them all	any start/exec/remove fails; containers, bundles or logs are left behind
Churn	Runs `STRESS_CHURN` (200) short-lived containers	a container did not run; daemon RSS grows by more than `STRESS_RSS_GROWTH` MiB (40) or its open files by more than `STRESS_FD_GROWTH` (16); bundles are left behind
Log flood	A container writes as fast as it can for `STRESS_LOG_SECONDS` (20) with `-log-max-size 1m -log-max-files 2`	logs on disk exceed the limit; `nyx logs --tail` takes 2 s or more
Out of memory (rootful)	A Compose service with a 32 MiB memory limit allocates 1 GiB, next to a second service	the hog is not OOM-killed (exit 137); the other service or the daemon stops
Daemon crash	`SIGKILL`s nyxd with 10 containers running and starts it again	the containers are not all re-adopted within 30 s, or exec into them fails
Large image (optional)	`STRESS_BIG_IMAGE=<ref>` pulls and runs a large image	the pull or run fails

It reports the daemon's RSS at idle, under load and at its peak, start/remove/churn times, log sizes and re-adopt time. Set STRESS_REPORT=<file> for a Markdown copy.

Run it by hand on the Actions tab (Stress → Run workflow) with a different number of containers, churn count or a large image to pull.

Scenarios​

Scenarios