- Explicitly define memory and CPU `requests` and `limits` to prevent pods from starving the node.
- Define distinct liveness and readiness probes rather than relying entirely on application crashes.
- Always declare resource `requests` (for node scheduling) and `limits` (for throttling) for CPU and Memory in Kubernetes Deployments.
- Implement HTTP Readiness Probes (to accept traffic) apart from Liveness Probes (to restart deadlocked apps).
- Keep Pod replicas ≥ 2 in Production to allow Rolling Updates without dropping zero-downtime traffic.