Controller-Runtime Patterns

This skill covers patterns for building Kubernetes controllers using controller-runtime with Milo's multi-cluster runtime provider.

Overview

Datum Cloud services can use controller-runtime when:

•etcd storage is sufficient (no custom storage backend needed)
•Standard CRUD and reconciliation semantics fit the use case
•The service needs to operate across multiple project control planes

Decision guidance: Read k8s-apiserver-patterns/architecture-decision.md to understand when to use this approach vs aggregated API servers.

Key Files

File	Purpose
`multicluster-provider.md`	Using Milo's multi-cluster runtime provider
`reconciler-patterns.md`	Standard reconciliation patterns
`status-conditions.md`	Status condition conventions

Milo Multi-Cluster Runtime

Milo provides a multi-cluster runtime provider that enables controllers to operate across the dynamic fleet of project control planes.

Location: milo/pkg/multicluster-runtime/

Architecture

code

┌─────────────────────────────────────────────────────────────┐
│                    Milo Control Plane                       │
│  ┌───────────────────────────────────────────────────────┐  │
│  │              Multi-Cluster Manager                    │  │
│  │  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐   │  │
│  │  │ Controller  │  │ Controller  │  │ Controller  │   │  │
│  │  └──────┬──────┘  └──────┬──────┘  └──────┬──────┘   │  │
│  └─────────│────────────────│────────────────│──────────┘  │
│            │                │                │              │
│            ▼                ▼                ▼              │
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐         │
│  │  Project A  │  │  Project B  │  │  Project C  │         │
│  │  Control    │  │  Control    │  │  Control    │         │
│  │  Plane      │  │  Plane      │  │  Plane      │         │
│  └─────────────┘  └─────────────┘  └─────────────┘         │
└─────────────────────────────────────────────────────────────┘

Discovery Modes

Mode	Setting	Connection	Best For
Internal	`InternalServiceDiscovery: true`	`milo-apiserver.project-{name}.svc.cluster.local:6443`	Operators in infrastructure cluster
External	`InternalServiceDiscovery: false`	Via Milo API endpoint	Remote operators

Basic Setup

import (
    "sigs.k8s.io/controller-runtime/pkg/manager"
    mcruntime "sigs.k8s.io/multicluster-runtime/pkg/manager"
    miloprovider "go.datum.net/milo/pkg/multicluster-runtime/milo"
)

func main() {
    // Create the multi-cluster manager
    mgr, err := mcruntime.New(cfg, mcruntime.Options{
        Scheme: scheme,
    })
    if err != nil {
        // handle error
    }

    // Create the Milo provider
    provider, err := miloprovider.New(mgr.GetClient(), miloprovider.Options{
        InternalServiceDiscovery: true,
    })
    if err != nil {
        // handle error
    }

    // Start provider - watches Projects and engages clusters
    go provider.Run(ctx, mgr)

    // Register controllers
    if err := (&MyReconciler{}).SetupWithManager(mgr); err != nil {
        // handle error
    }

    // Start manager
    if err := mgr.Start(ctx); err != nil {
        // handle error
    }
}

Reconciler Pattern

type MyResourceReconciler struct {
    client.Client
    Scheme *runtime.Scheme
}

func (r *MyResourceReconciler) Reconcile(ctx context.Context, req ctrl.Request) (ctrl.Result, error) {
    log := log.FromContext(ctx)

    // Fetch the resource
    var resource myv1.MyResource
    if err := r.Get(ctx, req.NamespacedName, &resource); err != nil {
        return ctrl.Result{}, client.IgnoreNotFound(err)
    }

    // Add finalizer if needed
    if !controllerutil.ContainsFinalizer(&resource, finalizerName) {
        controllerutil.AddFinalizer(&resource, finalizerName)
        if err := r.Update(ctx, &resource); err != nil {
            return ctrl.Result{}, err
        }
    }

    // Handle deletion
    if !resource.DeletionTimestamp.IsZero() {
        return r.reconcileDelete(ctx, &resource)
    }

    // Reconcile desired state
    return r.reconcileNormal(ctx, &resource)
}

func (r *MyResourceReconciler) SetupWithManager(mgr ctrl.Manager) error {
    return ctrl.NewControllerManagedBy(mgr).
        For(&myv1.MyResource{}).
        Owns(&corev1.ConfigMap{}).
        Complete(r)
}

Status Conditions

Use standard Kubernetes condition types:

import "k8s.io/apimachinery/pkg/api/meta"

// Set condition
meta.SetStatusCondition(&resource.Status.Conditions, metav1.Condition{
    Type:               "Ready",
    Status:             metav1.ConditionTrue,
    ObservedGeneration: resource.Generation,
    Reason:             "ReconcileSucceeded",
    Message:            "Resource is ready",
})

// Update status
if err := r.Status().Update(ctx, &resource); err != nil {
    return ctrl.Result{}, err
}

Multi-Cluster Awareness

Controllers receive a cluster-aware context. Access cluster information:

func (r *MyResourceReconciler) Reconcile(ctx context.Context, req ctrl.Request) (ctrl.Result, error) {
    // Get cluster from context (if using multicluster-runtime)
    cluster, ok := mcruntime.ClusterFromContext(ctx)
    if ok {
        log.Info("reconciling in cluster", "cluster", cluster.Name())
    }

    // Use cluster-specific client if needed
    clusterClient := cluster.GetClient()

    // ...
}

When to Use This Pattern

Use Case	Why Controller-Runtime Fits
Infrastructure operators	Reconciliation handles drift naturally
Cross-cluster sync	Multi-cluster runtime discovers clusters
Policy enforcement	React to resource changes
Workflow automation	Watch and act on state transitions
Standard CRUD	No custom API behavior needed

When NOT to Use This Pattern

Use aggregated API servers instead when:

•You need a custom storage backend (database, external system)
•You need fine-grained control over API endpoints
•Synchronous request-response semantics are required

See k8s-apiserver-patterns/architecture-decision.md for the full decision framework.

Related Skills

•k8s-apiserver-patterns — For aggregated API server approach
•go-conventions — Code style and testing
•capability-activity — Emitting events for activity tracking