Kubernetes Deployments declaratively manage Pod replicas with rolling update strategies. Resource requests (scheduling guarantee) and limits (enforcement ceiling) must be set for all containers. Liveness probes restart unhealthy pods; readiness probes gate traffic routing — the most impactful reliability levers. HPA (Horizontal Pod Autoscaler) scales replicas based on CPU/memory or custom metrics. PodDisruptionBudgets ensure availability during voluntary disruptions.
Zero-downtime rolling update with maxSurge and maxUnavailable.
Without readiness probes, Kubernetes sends traffic to pods before they've finished startup. This causes errors during every deployment. A /health/ready endpoint checking DB connectivity is essential.
Setting requests << limits (burstable QoS) means your pod can steal CPU from neighbours. Setting requests = limits (guaranteed QoS) gives predictable performance and prevents throttling surprises.
Without a PDB, node drains during Kubernetes upgrades may terminate all replicas of a deployment simultaneously. spec.minAvailable: 2 (or 50%) ensures capacity during voluntary disruptions.
Sign in to share your feedback and join the discussion.