Docker smoke test loop

Docker smoke-test loop

When to use

• CI needs to verify a freshly built image actually boots and accepts connections.
• You have a sentinel line your server prints on ready (e.g. CRAFT_SERVER_URL=...).
• Want cleanup guaranteed even on failure.
• Different CI environments have different tooling (sometimes no CLI, no node, no nc).

How it works

1. Deterministic container name + token: CONTAINER_NAME="craft-smoke-$$" (PID of the script, avoids collisions when tests run in parallel), TOKEN="smoke-test-$(openssl rand -hex 16)".
2. Cleanup trap before anything else:

   cleanup() { docker logs "$CONTAINER_NAME" 2>/dev/null || true; docker rm -f "$CONTAINER_NAME" 2>/dev/null || true; }
   trap cleanup EXIT
   

   Guarantees logs flush to CI output and container is removed even if the script exits abnormally. The || true swallows errors when the container never got created.
3. Run the container detached: docker run -d --name ... -p PORT:9100 -e TOKEN ....
4. Readiness loop polls two conditions each second:
   • Still running? docker inspect -f '{{.State.Running}}' "$CONTAINER_NAME" - if no, dump logs and exit 1 immediately.
   • Ready? docker logs "$CONTAINER_NAME" 2>&1 | grep -q "CRAFT_SERVER_URL=" - match the server's own stdout signal. Timeout defaults to 30s via SMOKE_TEST_TIMEOUT env.
5. Layered test invocation:
   • If bun + apps/cli/src/index.ts exist, run the real CLI validator.
   • Else try a tiny node-based WebSocket ping: new WebSocket(url, { headers: { 'x-craft-token': TOKEN }}).
   • Else just nc -z 127.0.0.1 $PORT to confirm the port is listening.
   • Else warn but don't fail - container-is-running counts as a weak smoke signal.

Example

#!/usr/bin/env bash
set -euo pipefail
IMAGE="${1:?Usage: smoke.sh <image>}"
CONTAINER="smoke-$$"
TOKEN="$(openssl rand -hex 16)"
trap 'docker logs "$CONTAINER" 2>/dev/null || true; docker rm -f "$CONTAINER" 2>/dev/null || true' EXIT

docker run -d --name "$CONTAINER" -p 9100:9100 -e "TOKEN=$TOKEN" "$IMAGE"

for _ in $(seq 30); do
  docker inspect -f '{{.State.Running}}' "$CONTAINER" | grep -q true || { docker logs "$CONTAINER"; exit 1; }
  docker logs "$CONTAINER" 2>&1 | grep -q "SERVER_URL=" && break
  sleep 1
done

nc -z 127.0.0.1 9100 && echo OK

Gotchas

• docker logs | grep is cheap but re-reads the whole log each iteration. Fine for small outputs, not for logs-per-second servers.
• The readiness sentinel must be on stdout or stderr and flushed promptly - buffered output breaks the probe.
• trap ... EXIT fires on normal exit AND signals; combined with set -e it's the cleanest cleanup pattern.
• Use $$ not $RANDOM for the container name in CI - deterministic per test-runner PID makes debugging easier.