Health check¶

Prerequisites

A running Meridian instance (Docker Compose)
curl

What you'll have after

A reliable way to confirm Meridian is up, plus the exact command to use in a Docker/Kubernetes health probe.

The endpoint¶

curl -s http://localhost:3011/api/cra/health

Expected:

{"status":"ok"}

A 200 with "status":"ok" means the HTTP server is up and the core is initialised. The health endpoint is intentionally unauthenticated so probes do not need a token.

Use it as a container health probe¶

In docker-compose.yml:

services:
  meridian:
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:3011/api/cra/health"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 20s

In Kubernetes:

livenessProbe:
  httpGet:
    path: /api/cra/health
    port: 3011
  initialDelaySeconds: 20
  periodSeconds: 30
readinessProbe:
  httpGet:
    path: /api/cra/health
    port: 3011
  initialDelaySeconds: 10
  periodSeconds: 10

What health does not tell you¶

Health confirms the service is up. It does not confirm that:

your LLM tier (Ollama/DeepSeek/Anthropic) is reachable,
your S3/MinIO audit target is writable,
your custom rules file parsed correctly.

To confirm those end-to-end, submit a known diff and read the RFC back (Your first analysis). If Gate 3 silently does nothing, the LLM tier is likely misconfigured — see LLM cost control.

Troubleshooting¶

Symptom	Likely cause	Fix
Connection refused	Container not started / wrong port	`docker compose ps`; check `PORT` and the published port
`200` but `status` not `ok`	Core failed to initialise	`docker compose logs meridian` for the startup error
Health ok, analyses never finish	LLM tier unreachable	Verify `OLLAMA_BASE_URL` / API keys; check egress in air-gapped setups
Health probe flapping in k8s	`start_period`/`initialDelaySeconds` too short on cold start	Increase the initial delay

Next: ENV reference