Files
k8-multicluster/CONTEXT-SWITCHING.md

4.3 KiB

Kubernetes Context Management

Patterns for managing multiple cluster contexts.

Understanding Contexts

A kubeconfig context combines:

  • Cluster: API server URL and CA certificate
  • User: Authentication credentials
  • Namespace: Default namespace (optional)

Context Naming Conventions

By Environment

production-us-east-1
production-eu-west-1
staging-us-east-1
development

By Purpose

management-cluster
workload-cluster-1
workload-cluster-2
monitoring-cluster

By Team

team-a-dev
team-a-prod
team-b-dev
platform-admin

MCP Commands

# List all contexts
list_contexts_tool()

# View kubeconfig (sanitized)
kubeconfig_view()

Context in Every Command

All kubectl-mcp-server tools accept context parameter:

# Explicit context (recommended)
get_pods(namespace="default", context="production")
get_pods(namespace="default", context="staging")

# Current context (implicit)
get_pods(namespace="default")

Multi-Cluster Patterns

Compare Environments

# Get deployments from both clusters
prod_deps = get_deployments(namespace="app", context="production")
staging_deps = get_deployments(namespace="app", context="staging")

Cross-Cluster Secret Sync

# Read secret from source
secret = get_secrets(namespace="app", context="source-cluster")

# Apply to target (via manifest)
apply_manifest(secret_yaml, namespace="app", context="target-cluster")

Parallel Health Checks

contexts = ["prod-1", "prod-2", "staging"]
for ctx in contexts:
    nodes = get_nodes(context=ctx)
    pods = get_pods(namespace="kube-system", context=ctx)
    # Report status

Rolling Deployment Across Clusters

# Deploy to staging first
install_helm_chart(name="app", chart="./app",
                   namespace="prod", context="staging")

# Verify
get_pods(namespace="prod", label_selector="app=myapp", context="staging")

# Deploy to production regions sequentially
for region in ["us-east", "us-west", "eu-west"]:
    install_helm_chart(name="app", chart="./app",
                       namespace="prod", context=f"production-{region}")
    # Wait for healthy
    get_pods(namespace="prod", context=f"production-{region}")

Access Control Per Context

Separate Kubeconfigs

# Different files per environment
export KUBECONFIG=~/.kube/production.yaml
export KUBECONFIG=~/.kube/development.yaml

Service Account Per Cluster

# Different SAs for different access levels
# Production: read-only SA
get_pods(namespace="app", context="production-readonly")

# Development: full access SA
apply_manifest(manifest, namespace="app", context="development-admin")

Cluster API Management

For managing cluster lifecycle:

# List CAPI-managed clusters
capi_clusters_list_tool(namespace="capi-system")

# Get workload cluster kubeconfig
kubeconfig = capi_cluster_kubeconfig_tool(
    name="workload-cluster-1",
    namespace="capi-system"
)

# Use the kubeconfig to access workload cluster
# (save to file, add to contexts)

Federation Patterns

Kubefed Resources

Cross-cluster resource propagation with KubeFed.

Cluster Mesh (Cilium)

# Check Cilium cluster mesh status
cilium_nodes_list_tool(context="cluster-1")
cilium_nodes_list_tool(context="cluster-2")

Multi-Cluster Service Mesh (Istio)

istio_proxy_status_tool(context="primary")
istio_proxy_status_tool(context="remote")

Best Practices

  1. Always Use Explicit Context

    # Good: explicit
    get_pods(context="production")
    
    # Risky: implicit
    get_pods()  # Which cluster?
    
  2. Naming Convention

    • Consistent format: <env>-<region>
    • Short but descriptive
  3. Read-Only for Production

    • Default production context: read-only
    • Separate admin context for changes
  4. Audit Cross-Cluster Operations

    • Log which clusters are accessed
    • Track who made changes
  5. Separate Credentials

    • Different SAs per environment
    • Shorter token lifetimes for prod

Troubleshooting

Context Not Found

list_contexts_tool()
# Verify context name is correct

Authentication Failure

# Check credentials
# Token expired?
# Certificate valid?

Network Issues

# API server reachable?
# VPN connected?
# Firewall rules?