Kubernetes for DevOps: Pods, Deployments, Services & More

Kubernetes (K8s) has become the standard runtime fabric for containerised workloads. Whether you are deploying a single microservice or coordinating hundreds of them across multiple availability zones, K8s gives you a consistent, declarative API to describe what should run, where it should run, and how it should behave. This guide walks through the core concepts a DevOps engineer needs day-to-day — cluster architecture, workload primitives, networking, configuration, observability, and troubleshooting — with working YAML and kubectl commands throughout. To see where Kubernetes fits in the broader learning journey, check the Kubernetes roadmap.

Why Kubernetes?

Section 1 of 17 · ~1 min

Running containers in production with plain Docker quickly exposes a set of operational problems: who restarts a crashed container, how do you roll out a new image version without downtime, how do you balance traffic across multiple replicas, and how do you ensure a container on a node with 64 GB RAM does not starve a neighbouring container of memory?

Kubernetes solves these problems through the desired-state model: you tell the cluster what the end state should look like (a YAML manifest), and controllers continuously reconcile actual state toward that goal. This gives you:

• Self-healing — crashed containers are restarted; failed nodes trigger Pod rescheduling.
• Horizontal scaling — add replicas with one command or automatically via HPA.
• Rolling updates and rollbacks — new versions roll out gradually; one command reverts them.
• Service discovery and load balancing — Pods get stable DNS names without static IPs.
• Declarative configuration — your entire infrastructure lives in version-controlled YAML.

Note: Kubernetes does not replace a CI/CD pipeline. It is the runtime target; tools like Argo CD, Flux, or Jenkins deploy to it.

Cluster Architecture

Section 2 of 17 · ~1 min

Understanding the architecture prevents a lot of confusion when things go wrong.

Control Plane

The control plane is the brain of the cluster. In managed services (EKS, GKE, AKS) it is operated for you; in self-managed setups you are responsible for its reliability.

Component	Role
kube-apiserver	Single entry point for all cluster operations. Validates and persists objects to etcd. Every kubectl command talks to this.
etcd	Distributed key-value store. Holds all cluster state. Losing etcd without a backup loses your cluster.
kube-scheduler	Watches for new Pods with no assigned node and picks the best node based on resource requests, taints, affinity rules, and topology spread.
kube-controller-manager	Runs core controllers: ReplicaSet, Deployment, Node, Endpoints, ServiceAccount, and others. Each controller reconciles its resource type.
cloud-controller-manager	(Cloud clusters only) Interfaces with cloud provider APIs — creates load balancers, routes, storage volumes.

Worker Nodes

Every node that runs workloads must have three components:

Component	Role
kubelet	Agent on each node. Receives PodSpec from the API server and instructs the container runtime to start/stop containers. Reports node and Pod status.
kube-proxy	Maintains iptables/ipvs rules that implement Service networking — forwarding traffic to the correct Pod IPs.
Container runtime	Actually runs containers. Kubernetes supports any CRI-compatible runtime: containerd (most common), CRI-O, or Docker (via cri-dockerd shim).

Tip: In production, the control plane nodes are typically tainted so user workloads do not schedule on them. Managed services do this automatically.

How a Pod Gets Scheduled

1. You run kubectl apply -f pod.yaml.
2. kube-apiserver validates the manifest and stores it in etcd.
3. kube-scheduler notices a Pod with no nodeName and selects a node.
4. kubelet on that node watches the API server, sees the Pod assigned to it, pulls the image, and starts the container.
5. kube-proxy updates network rules so the Pod is reachable via its Service.

Pods

Section 3 of 17 · ~2 min

A Pod is the smallest schedulable unit in Kubernetes. It wraps one or more containers (built from Docker for DevOps images) that share:

• Network namespace — all containers in a Pod share the same IP address and port space.
• Storage — volumes are declared at the Pod level and mounted into individual containers.
• Lifecycle — all containers start and stop together with the Pod.

Minimal Pod Manifest

apiVersion: v1
kind: Pod
metadata:
  name: web-pod
  labels:
    app: web
spec:
  containers:
    - name: nginx
      image: nginx:1.27
      ports:
        - containerPort: 80

kubectl apply -f pod.yaml
kubectl get pods
kubectl describe pod web-pod
kubectl delete pod web-pod

Pod Lifecycle

A Pod moves through these phases:

Phase	Meaning
Pending	Pod accepted by the cluster; containers not yet started (pulling images, waiting for node).
Running	At least one container is running or starting.
Succeeded	All containers exited with code 0 (typical for Jobs).
Failed	At least one container exited with non-zero code and will not restart.
Unknown	Node communication lost; state cannot be determined.

Multi-Container Patterns

Pods can host multiple containers. The three canonical patterns are:

• Sidecar — a helper container that augments the main one (e.g., a log-shipper or a service mesh proxy like Envoy/Istio).
• Init container — runs to completion before the main containers start (e.g., database migration, config rendering from a vault).
• Ambassador — a proxy container that abstracts external services (e.g., a Redis proxy in front of a remote cluster).

apiVersion: v1
kind: Pod
metadata:
  name: app-with-sidecar
spec:
  initContainers:
    - name: init-migrate
      image: migrate-tool:v1
      command: ["./migrate", "--up"]
  containers:
    - name: app
      image: myapp:v2
      ports:
        - containerPort: 8080
    - name: log-shipper
      image: fluent/fluent-bit:3.0
      volumeMounts:
        - name: app-logs
          mountPath: /var/log/app
  volumes:
    - name: app-logs
      emptyDir: {}

Why You Rarely Create Bare Pods

Bare Pods are not rescheduled if the node they are on fails. You should almost always use a higher-level workload controller (Deployment, StatefulSet, DaemonSet) that manages Pods on your behalf.

ReplicaSets and Deployments

Section 4 of 17 · ~1 min

ReplicaSet

A ReplicaSet ensures a specified number of identical Pod replicas are running at any time. It uses a label selector to identify the Pods it owns. If a Pod dies, the ReplicaSet creates a replacement.

Note: You almost never create ReplicaSets directly. Deployments create and manage ReplicaSets for you, adding rollout history.

Deployments

A Deployment wraps a ReplicaSet and adds rollout logic: you update the Pod template and the Deployment gradually replaces old Pods with new ones according to the configured strategy.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: web-deployment
  namespace: production
spec:
  replicas: 3
  selector:
    matchLabels:
      app: web
  strategy:
    type: RollingUpdate
    rollingUpdate:
      maxUnavailable: 1
      maxSurge: 1
  template:
    metadata:
      labels:
        app: web
    spec:
      containers:
        - name: web
          image: myapp:v2
          ports:
            - containerPort: 8080
          resources:
            requests:
              cpu: "250m"
              memory: "256Mi"
            limits:
              cpu: "500m"
              memory: "512Mi"

Key fields:

• selector.matchLabels — the Deployment manages any Pod with these labels. Must match template.metadata.labels.
• strategy.rollingUpdate.maxUnavailable — how many Pods can be unavailable during a rollout (absolute or percentage).
• strategy.rollingUpdate.maxSurge — how many extra Pods can be created above the desired count during a rollout.

# Apply or update
kubectl apply -f deployment.yaml

# Watch rollout progress
kubectl rollout status deployment/web-deployment

# Scale manually
kubectl scale deployment web-deployment --replicas=5

# View rollout history
kubectl rollout history deployment/web-deployment

# Rollback one version
kubectl rollout undo deployment/web-deployment

Tip: Always set resources.requests and resources.limits. Without them, the scheduler cannot make good placement decisions and your Pods get the lowest QoS class.

Services

Section 5 of 17 · ~1 min

A Service gives a stable DNS name and IP to a dynamic set of Pods selected by a label selector. Even as Pods are created and destroyed, the Service’s IP (ClusterIP) and DNS name stay constant.

Service Types

Type	Accessibility	Typical Use
ClusterIP (default)	Inside the cluster only	Internal microservice communication
NodePort	Cluster nodes’ IPs on a static port (30000–32767)	Dev/test external access without a cloud LB
LoadBalancer	External IP via cloud provider’s load balancer	Production external traffic
ExternalName	DNS alias for an external service	Pointing in-cluster workloads at external endpoints

ClusterIP Example

apiVersion: v1
kind: Service
metadata:
  name: web-service
  namespace: production
spec:
  selector:
    app: web
  ports:
    - protocol: TCP
      port: 80
      targetPort: 8080
  type: ClusterIP

Pods in the same namespace can reach this service at web-service:80; pods in other namespaces use web-service.production.svc.cluster.local:80.

LoadBalancer Example

apiVersion: v1
kind: Service
metadata:
  name: web-lb
  namespace: production
  annotations:
    service.beta.kubernetes.io/aws-load-balancer-type: "nlb"
spec:
  selector:
    app: web
  ports:
    - protocol: TCP
      port: 443
      targetPort: 8080
  type: LoadBalancer

How kube-proxy Implements Services

kube-proxy watches the API server for Service and Endpoints objects. When a Service is created, it programs iptables rules (or IPVS rules in IPVS mode) on every node so that traffic to the ClusterIP is DNAT’d to one of the healthy Pod IPs in the Endpoints list.

kubectl get svc -n production
kubectl describe svc web-service -n production
kubectl get endpoints web-service -n production

Caution: A Service with no matching Pods will have an empty Endpoints list and traffic will be silently dropped. Always check kubectl get endpoints when debugging connectivity.

Ingress

Section 6 of 17 · ~1 min

A Service of type LoadBalancer creates a separate cloud load balancer per service — expensive at scale. An Ingress is a cluster-wide HTTP(S) router that multiplexes many Services behind a single load balancer using hostname and path rules.

Ingress Controller

An Ingress object is just a configuration resource. You also need an Ingress controller — a Pod that reads Ingress objects and programs a reverse proxy. Common choices:

• NGINX Ingress Controller (most widely deployed)
• AWS Load Balancer Controller (for ALB on EKS — see AWS for DevOps Engineers)
• Traefik (dynamic config, good for smaller clusters)
• Istio Gateway (service mesh, advanced traffic management)

Ingress Manifest with TLS

apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: web-ingress
  namespace: production
  annotations:
    nginx.ingress.kubernetes.io/rewrite-target: /
    nginx.ingress.kubernetes.io/ssl-redirect: "true"
spec:
  ingressClassName: nginx
  tls:
    - hosts:
        - app.example.com
      secretName: tls-app-example-com
  rules:
    - host: app.example.com
      http:
        paths:
          - path: /api
            pathType: Prefix
            backend:
              service:
                name: api-service
                port:
                  number: 80
          - path: /
            pathType: Prefix
            backend:
              service:
                name: web-service
                port:
                  number: 80

The tls.secretName field references a Kubernetes Secret of type kubernetes.io/tls that holds the certificate and key. Tools like cert-manager automate TLS certificate provisioning and renewal via Let’s Encrypt.

kubectl get ingress -n production
kubectl describe ingress web-ingress -n production

Tip: Set ingressClassName explicitly. Older clusters used annotations (kubernetes.io/ingress.class), but the ingressClassName field is the current standard.

ConfigMaps

Section 7 of 17 · ~1 min

A ConfigMap stores non-sensitive configuration as key-value pairs or files. Pods consume ConfigMaps as environment variables or as mounted files.

Creating a ConfigMap

apiVersion: v1
kind: ConfigMap
metadata:
  name: app-config
  namespace: production
data:
  LOG_LEVEL: "info"
  MAX_CONNECTIONS: "100"
  config.yaml: |
    server:
      host: 0.0.0.0
      port: 8080
    database:
      pool_size: 10

kubectl apply -f configmap.yaml
kubectl get configmap app-config -n production -o yaml

Consuming a ConfigMap as Environment Variables

spec:
  containers:
    - name: app
      image: myapp:v2
      envFrom:
        - configMapRef:
            name: app-config

Or inject individual keys:

      env:
        - name: LOG_LEVEL
          valueFrom:
            configMapKeyRef:
              name: app-config
              key: LOG_LEVEL

Consuming a ConfigMap as a Mounted File

spec:
  containers:
    - name: app
      image: myapp:v2
      volumeMounts:
        - name: config-volume
          mountPath: /etc/app
  volumes:
    - name: config-volume
      configMap:
        name: app-config
        items:
          - key: config.yaml
            path: config.yaml

The file /etc/app/config.yaml inside the container will contain the YAML defined in the ConfigMap.

Note: When you update a ConfigMap that is mounted as a volume, Kubernetes eventually (within ~60 seconds) updates the file in the container. Environment variables injected via envFrom are not updated — the Pod must be restarted.

Secrets

Section 8 of 17 · ~1 min

Secrets store sensitive data — passwords, API tokens, TLS certificates, SSH keys. They are similar to ConfigMaps but:

• Values are base64-encoded (not encrypted by default, but can be with encryption at rest).
• Access is controlled by RBAC — separate from ConfigMap access.
• They integrate with external secret stores (HashiCorp Vault, AWS Secrets Manager) via CSI drivers or operators.

Secret Types

Type	Use
Opaque (default)	Arbitrary key-value data
kubernetes.io/tls	TLS certificate and key
kubernetes.io/dockerconfigjson	Docker registry credentials
kubernetes.io/service-account-token	SA tokens (auto-managed)
kubernetes.io/ssh-auth	SSH private keys
kubernetes.io/basic-auth	Username/password pairs

Creating an Opaque Secret

apiVersion: v1
kind: Secret
metadata:
  name: db-credentials
  namespace: production
type: Opaque
data:
  username: YWRtaW4=        # base64("admin")
  password: c3VwZXJzZWNyZXQ= # base64("supersecret")

# Encode a value
echo -n "supersecret" | base64

# Create from literal (easier)
kubectl create secret generic db-credentials \
  --from-literal=username=admin \
  --from-literal=password=supersecret \
  -n production

# View (values are base64)
kubectl get secret db-credentials -n production -o yaml

# Decode a value
kubectl get secret db-credentials -n production \
  -o jsonpath='{.data.password}' | base64 --decode

Consuming a Secret in a Pod

spec:
  containers:
    - name: app
      image: myapp:v2
      env:
        - name: DB_USERNAME
          valueFrom:
            secretKeyRef:
              name: db-credentials
              key: username
        - name: DB_PASSWORD
          valueFrom:
            secretKeyRef:
              name: db-credentials
              key: password

Encryption at Rest and RBAC

By default, Secrets in etcd are stored base64-encoded but not encrypted. Enable encryption at rest via the API server’s --encryption-provider-config flag using an aescbc or kms provider.

Use RBAC to restrict which service accounts can get, list, or watch Secrets:

apiVersion: rbac.authorization.k8s.io/v1
kind: Role
metadata:
  name: secret-reader
  namespace: production
rules:
  - apiGroups: [""]
    resources: ["secrets"]
    resourceNames: ["db-credentials"]
    verbs: ["get"]

Caution: Never commit Secret manifests with real values to version control. Use sealed secrets (Bitnami Sealed Secrets), External Secrets Operator, or a Vault CSI driver to inject secrets at runtime.

Namespaces

Section 9 of 17 · ~1 min

Namespaces provide a mechanism for isolating groups of resources within a single cluster. They are soft multi-tenancy: they scope names, RBAC, and resource quotas, but do not provide strong network isolation by themselves (for that, use NetworkPolicies).

Built-in Namespaces

Namespace	Purpose
default	Default for resources with no namespace specified
kube-system	Kubernetes control plane components (CoreDNS, kube-proxy, etc.)
kube-public	Publicly readable resources (ConfigMap with cluster info)
kube-node-lease	Node heartbeat lease objects

Common Namespace Operations

# List all namespaces
kubectl get namespaces

# Create a namespace
kubectl create namespace staging

# Run commands in a specific namespace
kubectl get pods -n staging

# Set a default namespace for your context
kubectl config set-context --current --namespace=staging

# View resources across all namespaces
kubectl get pods --all-namespaces
# or
kubectl get pods -A

ResourceQuota and LimitRange

Namespaces support ResourceQuota (cap total resource usage per namespace) and LimitRange (set default and max limits per container).

apiVersion: v1
kind: ResourceQuota
metadata:
  name: team-quota
  namespace: staging
spec:
  hard:
    requests.cpu: "4"
    requests.memory: 8Gi
    limits.cpu: "8"
    limits.memory: 16Gi
    count/pods: "20"

Resource Requests and Limits

Section 10 of 17 · ~2 min

Getting resource configuration right is one of the highest-leverage tuning tasks in Kubernetes.

The Concepts

• Request: the amount of CPU/memory the scheduler reserves on a node for this container. The node must have at least this much available for the Pod to be scheduled there.
• Limit: the maximum CPU/memory the container is allowed to use. Exceeding the CPU limit causes throttling; exceeding the memory limit causes the container to be OOM-killed.

resources:
  requests:
    cpu: "250m"      # 250 millicores = 0.25 CPU core
    memory: "256Mi"
  limits:
    cpu: "500m"
    memory: "512Mi"

CPU is expressed in cores (1, 0.5) or millicores (500m). Memory uses binary suffixes (Ki, Mi, Gi).

QoS Classes

Kubernetes assigns one of three Quality of Service classes to every Pod based on its resource configuration:

QoS Class	Condition	Eviction Priority
Guaranteed	Every container has requests == limits for both CPU and memory	Last to be evicted
Burstable	At least one container has a request or limit, but not Guaranteed	Middle priority
BestEffort	No container has any request or limit	First to be evicted under pressure

Tip: For production workloads, aim for Guaranteed or Burstable QoS. Use kubectl describe pod <name> and look for QoS Class: in the output.

Setting the Right Numbers

Right-sizing requests and limits requires profiling your actual workload under load. Size CPU/memory requests and limits with the Kubernetes Resource Calculator.

Common mistakes:

• Setting limits much higher than requests creates “noisy neighbour” risk.
• Setting limits equal to very low requests may cause OOMKills under normal spikes.
• Not setting requests at all results in BestEffort Pods that are first to go under node pressure.

Vertical Pod Autoscaler (VPA) and Horizontal Pod Autoscaler (HPA)

• HPA scales the number of replicas based on CPU utilisation, memory, or custom metrics.
• VPA adjusts the requests/limits of existing containers (requires Pod restarts).

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: web-hpa
  namespace: production
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: web-deployment
  minReplicas: 2
  maxReplicas: 20
  metrics:
    - type: Resource
      resource:
        name: cpu
        target:
          type: Utilization
          averageUtilization: 60

Health Probes

Section 11 of 17 · ~1 min

Kubernetes uses three probe types to determine container health. Without probes, a container is considered healthy the moment it starts — even if the application inside is not ready yet.

Probe Types

Probe	Kubernetes Action on Failure
Liveness	Restart the container. Use for detecting deadlocks — processes that are running but stuck.
Readiness	Remove the Pod from Service endpoints. Traffic stops flowing to it, but the container is not restarted.
Startup	Disables liveness and readiness until it succeeds. Use for slow-starting apps to prevent premature liveness restarts.

Probe Mechanisms

Each probe type can use one of three mechanisms:

• httpGet — HTTP GET request; success = 2xx/3xx response.
• tcpSocket — TCP connection; success = port is open.
• exec — run a command inside the container; success = exit code 0.

Full Probe Example

spec:
  containers:
    - name: app
      image: myapp:v2
      ports:
        - containerPort: 8080
      startupProbe:
        httpGet:
          path: /healthz
          port: 8080
        failureThreshold: 30
        periodSeconds: 10
      readinessProbe:
        httpGet:
          path: /ready
          port: 8080
        initialDelaySeconds: 5
        periodSeconds: 10
        failureThreshold: 3
        successThreshold: 1
      livenessProbe:
        httpGet:
          path: /healthz
          port: 8080
        initialDelaySeconds: 30
        periodSeconds: 15
        failureThreshold: 3
        timeoutSeconds: 5

Key parameters:

Parameter	Default	Meaning
initialDelaySeconds	0	Seconds after container start before first probe
periodSeconds	10	How often to probe
failureThreshold	3	Failures before action is taken
successThreshold	1	Successes needed to transition back to healthy
timeoutSeconds	1	Seconds to wait for probe response

Tip: Always implement separate /healthz (liveness) and /ready (readiness) endpoints in your application. The readiness endpoint should check downstream dependencies (database reachability, cache connection); the liveness endpoint should only check the application itself.

Rollouts and Rollbacks

Section 12 of 17 · ~1 min

Rolling Update Strategy

The RollingUpdate strategy (default for Deployments) replaces old Pods with new ones incrementally. The Recreate strategy terminates all old Pods before starting new ones — useful when your app cannot run two versions simultaneously.

# Trigger a rollout by updating the image
kubectl set image deployment/web-deployment \
  web=myapp:v3 -n production

# Or edit the manifest and apply
kubectl apply -f deployment.yaml

# Watch the rollout in real time
kubectl rollout status deployment/web-deployment -n production

# Pause a rollout mid-way (canary-style)
kubectl rollout pause deployment/web-deployment -n production

# Resume
kubectl rollout resume deployment/web-deployment -n production

Rollback

# View rollout history with change-cause annotations
kubectl rollout history deployment/web-deployment -n production

# View a specific revision
kubectl rollout history deployment/web-deployment \
  --revision=2 -n production

# Roll back to previous version
kubectl rollout undo deployment/web-deployment -n production

# Roll back to a specific revision
kubectl rollout undo deployment/web-deployment \
  --to-revision=2 -n production

Tip: Annotate your rollouts with a change-cause so history is readable: add kubernetes.io/change-cause: "bump to v3, adds auth middleware" to the Deployment’s metadata.annotations before applying.

Deployment Revision History

By default, Kubernetes retains 10 revisions. Control this with spec.revisionHistoryLimit.

spec:
  revisionHistoryLimit: 5

Observability

Section 13 of 17 · ~1 min

A Kubernetes cluster generates a rich stream of metrics, logs, and events. Building good visibility requires three layers:

Logs

kubectl logs reads from the container’s stdout/stderr (captured by the container runtime and available via the API server):

# Follow logs from a running Pod
kubectl logs -f deployment/web-deployment -n production

# Show last 100 lines
kubectl logs web-pod --tail=100 -n production

# Logs from a specific container in a multi-container Pod
kubectl logs web-pod -c log-shipper -n production

# Logs from a previous (crashed) container instance
kubectl logs web-pod --previous -n production

For persistent log aggregation, deploy a log collector DaemonSet (Fluent Bit, Fluentd) that ships container logs to a backend (Elasticsearch, Loki, CloudWatch Logs).

Metrics and Prometheus

The Kubernetes ecosystem is deeply integrated with Prometheus. The metrics-server provides basic CPU/memory metrics for HPA and kubectl top. For full observability, deploy the kube-prometheus-stack (Prometheus + Alertmanager + Grafana) or a managed equivalent.

Key scrape targets:

• kube-state-metrics — Deployment replicas, Pod status, resource requests/limits, PVC state.
• node-exporter — node-level CPU, memory, disk, network.
• cadvisor — per-container CPU and memory (built into kubelet).
• Your application’s /metrics endpoint (expose Prometheus format).

Understanding Prometheus query syntax is essential for writing alerts and dashboards. Use the PromQL Explainer to translate complex queries into plain English. When building scrape configurations and relabelling rules, the Prometheus Relabel Tester lets you test rules against live label sets without touching production.

Events

Kubernetes Events are the first place to look when a Pod is not starting correctly:

kubectl get events -n production --sort-by='.lastTimestamp'
kubectl get events -n production --field-selector reason=OOMKilled

kubectl Essentials

Section 14 of 17 · ~1 min

Command	What It Does
kubectl get pods -n <ns>	List Pods in a namespace
kubectl get pods -A	List Pods in all namespaces
kubectl describe pod <name> -n <ns>	Full details, events, conditions
kubectl logs <pod> -n <ns>	Stdout/stderr from the main container
kubectl logs <pod> -c <container>	Logs from a specific container
kubectl logs <pod> --previous	Logs from the last crashed container instance
kubectl exec -it <pod> -- bash	Interactive shell in a running container
kubectl exec <pod> -- <cmd>	Run a one-off command
kubectl apply -f <file>	Declaratively create or update resources
kubectl delete -f <file>	Delete resources defined in a manifest
kubectl delete pod <name>	Force-delete a single Pod
kubectl scale deployment <name> --replicas=N	Scale a Deployment
kubectl rollout status deployment/<name>	Watch rollout progress
kubectl rollout undo deployment/<name>	Roll back to previous version
kubectl rollout history deployment/<name>	View revision history
kubectl set image deployment/<name> <c>=<img>	Update a container image
kubectl port-forward svc/<name> 8080:80	Forward a local port to a Service
kubectl port-forward pod/<name> 8080:8080	Forward a local port to a Pod
kubectl top pods -n <ns>	Current CPU/memory usage (requires metrics-server)
kubectl top nodes	Node CPU/memory usage
kubectl get events -n <ns> --sort-by=.lastTimestamp	Recent cluster events
kubectl explain deployment.spec.strategy	Inline API documentation
kubectl diff -f <file>	Preview what apply would change
kubectl get all -n <ns>	All common resources in a namespace
kubectl label pod <name> key=value	Add or update a label
kubectl annotate pod <name> key=value	Add or update an annotation
kubectl config get-contexts	List available kubeconfig contexts
kubectl config use-context <context>	Switch cluster context

Tip: Install kubectl plugins via krew (the kubectl plugin manager). Useful plugins include kubectl-neat (clean up get -o yaml output), kubectl-ctx and kubectl-ns (fast context/namespace switching), and kubectl-tree (visualise owner references).

Troubleshooting

Section 15 of 17 · ~2 min

CrashLoopBackOff

A container keeps crashing and Kubernetes keeps restarting it with exponential backoff.

Diagnose:

# See the current state and restart count
kubectl get pod <name> -n <ns>

# Check events for the Pod
kubectl describe pod <name> -n <ns>

# Check logs from the current or previous instance
kubectl logs <name> -n <ns>
kubectl logs <name> -n <ns> --previous

Common causes: application throws an unhandled exception at startup; missing environment variable or config file; entrypoint command is wrong; dependency (database, cache) not reachable.

ImagePullBackOff / ErrImagePull

The container image cannot be pulled.

Diagnose:

kubectl describe pod <name> -n <ns>
# Look for events like: Failed to pull image ... unauthorized

Common causes: image name or tag is wrong; registry is private and no imagePullSecret is configured; credentials in the pull secret have expired; rate-limit on Docker Hub.

Fix a missing imagePullSecret:

kubectl create secret docker-registry regcred \
  --docker-server=registry.example.com \
  --docker-username=<user> \
  --docker-password=<token> \
  -n production

Then reference it in the Pod spec:

spec:
  imagePullSecrets:
    - name: regcred

Pending Pods

A Pod stays in Pending and never starts.

Diagnose:

kubectl describe pod <name> -n <ns>
# Look for events under "Events:" — scheduler messages explain why

Common causes:

• Insufficient resources — no node has enough CPU or memory to satisfy the requests. Check kubectl top nodes and kubectl describe nodes.
• Taint/toleration mismatch — the target nodes are tainted but the Pod has no matching toleration.
• NodeSelector or affinity — the Pod requires specific labels on the node that no current node has.
• PVC not bound — a PersistentVolumeClaim is still Pending (waiting for a PV).

OOMKilled

A container was killed by the kernel OOM killer because it exceeded its memory limit.

Diagnose:

kubectl get pod <name> -n <ns>
# Look for: OOMKilled in the REASON column

kubectl describe pod <name> -n <ns>
# Look for: Last State: Terminated Reason: OOMKilled

Fix: Increase the memory limit (or request, to get a Guaranteed QoS). Profile the application’s actual memory use under load. Use the Kubernetes Resource Calculator to size requests and limits based on observed usage.

Debugging Toolkit

When kubectl logs is not enough:

# Open a shell in the running container
kubectl exec -it <pod> -n <ns> -- sh

# Run a temporary debug Pod on a specific node
kubectl debug node/<node-name> -it --image=ubuntu:22.04

# Attach an ephemeral debug container to a running Pod (K8s 1.23+)
kubectl debug -it <pod> -n <ns> \
  --image=busybox:1.36 \
  --target=<container-name>

# Copy a file from a Pod to local
kubectl cp <ns>/<pod>:/path/to/file ./local-file

Common Kubernetes Interview Questions

Section 16 of 17 · ~3 min

Q: What is the role of the etcd in a Kubernetes cluster? A: etcd is the cluster’s source of truth — a distributed, consistent key-value store where all cluster state (objects, configs, secrets) is persisted. If etcd is lost without a backup, the cluster’s configuration is lost. The API server is the only component that reads and writes etcd directly.

Q: How does a Kubernetes Service discover which Pods to send traffic to? A: Services use label selectors to identify target Pods. The Endpoints controller (or EndpointSlice controller in newer clusters) watches for Pods matching the selector and maintains an Endpoints/EndpointSlice object. kube-proxy reads these to program iptables/IPVS forwarding rules.

Q: What is the difference between a StatefulSet and a Deployment? A: A Deployment is for stateless workloads — Pods are interchangeable. A StatefulSet is for stateful applications (databases, queues): it gives each Pod a stable, ordered hostname (pod-0, pod-1), stable storage via PVCs that survive Pod rescheduling, and ordered, graceful start/stop.

Q: What is a DaemonSet? A: A DaemonSet ensures that one copy of a Pod runs on every node (or a subset via nodeSelector/tolerations). Common uses: log collectors (Fluent Bit), monitoring agents (node-exporter), CNI plugins, security agents.

Q: How does Kubernetes handle persistent storage? A: Via the PersistentVolume (PV) / PersistentVolumeClaim (PVC) abstraction. A PVC is a user’s request for storage (size, access mode). Kubernetes binds it to a matching PV (a piece of actual storage — EBS volume, NFS share, etc.) either statically or dynamically via a StorageClass and a CSI driver.

Q: What is a Kubernetes Namespace and when should you use multiple namespaces? A: A Namespace is a virtual cluster within a cluster. Use separate namespaces for teams, environments (dev/staging/production on the same cluster), or application boundaries. Apply ResourceQuotas and RBAC per namespace to enforce isolation.

Q: What is a Taint and Toleration? A: Taints are set on nodes to repel Pods. kubectl taint nodes node1 key=value:NoSchedule prevents any Pod without a matching Toleration from being scheduled on that node. Used to dedicate nodes for specific workloads (GPU nodes, spot nodes).

Q: What happens when you run kubectl apply -f deployment.yaml? A: kubectl sends the manifest to the API server, which validates it and merges it with the existing object (three-way merge using the last-applied-configuration annotation). If the Deployment’s Pod template changed, the Deployment controller creates a new ReplicaSet and performs a rolling update.

Q: What is the difference between kubectl apply and kubectl create? A: create is imperative — it fails if the object already exists. apply is declarative — it creates the object if missing or updates it if it exists, using a three-way merge. Prefer apply for GitOps and CI/CD pipelines.

Q: What is a Kubernetes Job vs a CronJob? A: A Job creates one or more Pods and tracks their completion. It retries failed Pods until the desired completions count is reached, then the Job is marked complete. A CronJob schedules Jobs on a cron expression, creating a new Job object on each trigger.

Note: Prefer a day-by-day path? This is covered in Mission 90 Days 66–74 — a free 90-day guided DevOps program with browser terminal missions.

What’s Next

Section 17 of 17 · ~1 min

Coming from containers? See Docker for DevOps.

Running on AWS (EKS)? See AWS for DevOps Engineers.