[subtask] [Subtask 2/4] Implement Kubernetes operator core framework

[github-actions]

Parent Issue: #20

Objective

Build the foundational Kubernetes operator framework that watches CRDs and manages the reconciliation loop for Wassette components.

Context

After the CRD schemas are designed (Subtask 1), we need a Kubernetes operator that:

• Watches for CRD events (create, update, delete)
• Implements reconciliation logic for component lifecycle
• Integrates with existing Wassette LifecycleManager
• Provides operator health and metrics endpoints

This subtask focuses on the operator infrastructure, not the MCP API exposure (which comes in Subtask 3).

Implementation Details

1. Choose Operator Framework

Evaluate and select a Rust Kubernetes operator framework:

• Option A: kube-rs - Most popular Rust k8s client
• Option B: operator-rs - Higher-level operator framework
• Recommendation: Start with kube-rs for flexibility and community support

2. Project Structure

Create new crate or binary:

• Location: crates/wassette-operator/ or src/operator.rs
• Dependencies: kube, k8s-openapi, existing wassette crate
• Binary target: wassette-operator (separate from main wassette binary)

3. Core Components to Implement

a. CRD Controller

• Watch WassetteComponent resources
• Implement reconciliation loop:
  • Fetch CRD spec
  • Download OCI artifact (reuse existing loader::load_resource)
  • Apply policy (integrate with PolicyManager)
  • Update CRD status with component state

b. Component Lifecycle Integration

• Integrate with existing LifecycleManager from crates/wassette/src/lib.rs
• Reuse component loading logic: load_component_optimized()
• Apply policies using existing policy_internal::PolicyManager
• Store component state in operator's in-memory registry

c. Error Handling & Status Updates

• Update CRD status on success/failure
• Implement retry logic with exponential backoff
• Surface errors through Kubernetes events and status conditions

4. Files to Create/Modify

• New: crates/wassette-operator/src/main.rs - Operator entrypoint
• New: crates/wassette-operator/src/controller.rs - Reconciliation logic
• New: crates/wassette-operator/src/crd.rs - CRD type definitions (based on Subtask 1 design)
• Modify: Cargo.toml - Add operator dependencies and workspace member
• New: crates/wassette-operator/Cargo.toml - Operator-specific dependencies

5. Configuration

• Support both in-cluster and out-of-cluster kubeconfig
• Configuration options: namespace to watch, component storage path, concurrency limits
• Environment variables: NAMESPACE, COMPONENT_DIR, RUST_LOG

Acceptance Criteria

• Operator successfully watches WassetteComponent CRDs
• Reconciliation loop loads components from OCI references
• Component status is updated in CRD status field
• Operator handles component creation, updates, and deletion
• Error conditions are properly surfaced through CRD status
• Unit tests for reconciliation logic
• Integration test with local Kubernetes cluster (kind/minikube)

Testing Strategy

Unit Tests

• Mock Kubernetes API client
• Test reconciliation logic with various CRD states
• Verify error handling and retry behavior

Integration Tests

• Use kind or minikube for local Kubernetes cluster
• Deploy test CRDs and verify operator behavior
• Test component loading from OCI registry
• Validate policy application

Example Test Scenario

apiVersion: wassette.microsoft.io/v1
kind: WassetteComponent
metadata:
  name: time-server
spec:
  ociRef: "(redacted)"
  policy:
    network:
      allow: []

Verify:

1. Component is downloaded and loaded
2. Status shows "Ready"
3. Tools are registered in operator's registry

Dependencies

• Requires: Subtask 1 (CRD schema design) must be completed first
• Blocks: Subtask 3 (MCP JSON-RPC API implementation)

References

• Wassette LifecycleManager: crates/wassette/src/lib.rs
• Component loader: crates/wassette/src/loader.rs
• Policy management: crates/wassette/src/policy_internal.rs
• kube-rs documentation: (redacted)
• Kubernetes operator pattern: (redacted)
  Related to Kubernetes Operator for mcp-wasmtime #20

AI generated by Plan for #20

[asw101]

Closing [subtask]s of parent issues.