Maintenance

This document provides guidance on checking the health and status of the Dataset Profiler service.

Health Checks

Readiness Probe

A lightweight readiness probe that does not depend on Ray or Redis. This is the endpoint Kubernetes should point its readiness/liveness probe at, so a transient Ray or Redis outage never deregisters the pod from the Service:

# Check readiness
curl -X GET "https://{host}/monitoring/ready"

Expected response:

{
  "status": "ready"
}

The root endpoint GET / is an equivalent pure-HTTP liveness target.

Dependency Health Check

Monitor the health of the service and its dependencies (Redis and Ray). This endpoint is purely diagnostic: it always returns HTTP 200 from a live pod and reports a dependency being down inside the response body (status: degraded) rather than as a non-200 status:

# Check dependency health
curl -X GET "https://{host}/monitoring/health-check"

Expected response:

{
  "status": "healthy",
  "redis": {
    "status": "healthy",
    "message": "Connected to Redis server successfully."
  },
  "ray": {
    "status": "healthy",
    "message": "Ray cluster reachable with 3 alive node(s)."
  }
}

When a dependency is unavailable the overall status becomes degraded and the affected dependency reports its own status and an explanatory message.

Ray Cluster Health Check

Check the health of the Ray cluster:

# Check Ray cluster status
ray status