Smoke-check the running platform

When to use

You have applied a change to the running stack and need to confirm the platform actually serves traffic. This is distinct from the automated smoke suite, which runs in CI against an ephemeral testcontainers stack. This playbook is the manual walk you do against a stack you can reach, after a change or deploy.

A green test suite proves the code compiles and units behave. It does not prove the running stack serves traffic, the feed is flowing, or the WebSocket fans out.

How it works

Walk the five layers in order. Stop and diagnose at the first failing layer; a later layer rarely passes when an earlier one is broken. The stack is reached through Traefik on http://localhost; prefer the proxied http://localhost/api/<service>/... routes over direct service ports when checking from outside containers.

1. Fleet is up

deno run --allow-run --allow-env scripts/status.ts --once

Every core service (gateway, market-sim, ems, oms, journal, user-service, frontend, redpanda, postgres) should report running. Idle-safe groups (algos, analytics, microstructure, aux) may legitimately be stopped; treat those as failures only if the change under test touches them.

2. Gateway answers HTTP

curl -sf http://localhost/api/gateway/health

Healthy response: {"service":"gateway","version":...,"status":"ok"}. A non-2xx or a hang means Traefik routing or the gateway process is the problem, not anything downstream.

Per-dependency readiness (needs an auth cookie):

curl -s http://localhost/api/gateway/ready

3. WebSocket connects on the correct path

The frontend connects to /ws/gateway, not /ws. This path has regressed before. Confirm the client path still matches the server route:

grep -n "GATEWAY_WS_URL" frontend/src/store/middleware/gatewayMiddleware.ts

A bare HTTP GET to the route returns 426 Upgrade Required when wired correctly; a 404 means the route is wrong:

curl -s -o /dev/null -w "%{http_code}\n" http://localhost/ws/gateway

4. Market-data feed is flowing

HTTP snapshot of current prices:

curl -s http://localhost/api/market-sim/prices | head -c 400

A populated price map means the simulator is producing. To confirm it reaches Kafka, consume a few ticks from inside the redpanda container:

docker compose exec -T redpanda rpk topic consume market.ticks -n 5

An empty or stalled market.ticks with a healthy market-sim points at the broker or the producer wiring, not the simulator.

5. Logs are clean

Identify unhealthy services first, then tail the ones the change touched, looking for crash loops, unhandled rejections, or repeated reconnects:

deno run --allow-run --allow-env scripts/status.ts --once
docker compose logs --tail=50 <service>

Reporting

State, per layer, what was run and what was observed. If a layer failed, name it and quote the exact output rather than declaring the platform healthy on the strength of CI. If only some layers were checked (for example a frontend-only change that skipped Kafka), say which were skipped and why.