|
| 1 | +# AGENTS.md |
| 2 | + |
| 3 | +This file provides AI agents with comprehensive context about the Operator Lifecycle Manager (OLM) v0 codebase to enable effective navigation, understanding, and contribution. |
| 4 | + |
| 5 | +## Project Status |
| 6 | + |
| 7 | +**CRITICAL**: This repository is in **maintenance mode**. OLM v0 accepts only critical bug fixes and security updates. For new development, use [operator-controller](https://github.com/operator-framework/operator-controller) (OLM v1). |
| 8 | + |
| 9 | +## Project Overview |
| 10 | + |
| 11 | +Operator Lifecycle Manager (OLM) extends Kubernetes to provide declarative installation, upgrade, and lifecycle management for Kubernetes operators. It's part of the [Operator Framework](https://github.com/operator-framework) ecosystem. |
| 12 | + |
| 13 | +### Core Capabilities |
| 14 | +- **Over-the-Air Updates**: Automatic operator updates via catalog channels |
| 15 | +- **Dependency Resolution**: Automatic resolution and installation of operator dependencies |
| 16 | +- **Multi-tenancy**: Namespace-scoped operator management via OperatorGroups |
| 17 | +- **Discovery**: Catalog-based operator discovery and installation |
| 18 | +- **Stability**: Prevents conflicting operators from owning the same APIs |
| 19 | + |
| 20 | +## Architecture |
| 21 | + |
| 22 | +OLM consists of two main operators working together: |
| 23 | + |
| 24 | +### 1. OLM Operator (`cmd/olm`) |
| 25 | +**Responsibility**: Manages the installation and lifecycle of operators defined by ClusterServiceVersions (CSVs) |
| 26 | + |
| 27 | +**Key Functions**: |
| 28 | +- Creates Deployments, ServiceAccounts, Roles, and RoleBindings from CSV specifications |
| 29 | +- Manages CSV lifecycle states: None → Pending → InstallReady → Installing → Succeeded/Failed |
| 30 | +- Monitors installed operator health and rotates certificates |
| 31 | +- Enforces OperatorGroup namespace scoping |
| 32 | + |
| 33 | +**Primary Controllers**: |
| 34 | +- CSV Controller (pkg/controller/operators/olm) |
| 35 | +- OperatorGroup Controller |
| 36 | + |
| 37 | +### 2. Catalog Operator (`cmd/catalog`) |
| 38 | +**Responsibility**: Manages operator catalogs, subscriptions, and dependency resolution |
| 39 | + |
| 40 | +**Key Functions**: |
| 41 | +- Monitors CatalogSources and builds operator catalogs |
| 42 | +- Processes Subscriptions to track operator updates |
| 43 | +- Generates InstallPlans with resolved dependencies |
| 44 | +- Creates CRDs and CSVs from catalog content |
| 45 | + |
| 46 | +**Primary Controllers**: |
| 47 | +- Subscription Controller |
| 48 | +- InstallPlan Controller |
| 49 | +- CatalogSource Controller |
| 50 | +- Registry Reconciler |
| 51 | + |
| 52 | +## Custom Resource Definitions (CRDs) |
| 53 | + |
| 54 | +| Resource | API Group | Owner | Description | |
| 55 | +|----------|-----------|-------|-------------| |
| 56 | +| **ClusterServiceVersion (CSV)** | operators.coreos.com/v1alpha1 | OLM | Defines operator metadata, installation strategy, permissions, and owned/required CRDs | |
| 57 | +| **Subscription** | operators.coreos.com/v1alpha1 | Catalog | Tracks operator updates from a catalog channel; drives automatic upgrades | |
| 58 | +| **InstallPlan** | operators.coreos.com/v1alpha1 | Catalog | Calculated list of resources to install/upgrade; requires approval (manual or automatic) | |
| 59 | +| **CatalogSource** | operators.coreos.com/v1alpha1 | Catalog | Repository of operators and metadata; served via grpc from operator-registry | |
| 60 | +| **OperatorGroup** | operators.coreos.com/v1 | OLM | Groups namespaces for operator installation scope; enables multi-tenancy | |
| 61 | +| **OperatorCondition** | operators.coreos.com/v2 | OLM | Tracks operator health status and conditions | |
| 62 | + |
| 63 | +## Directory Structure |
| 64 | + |
| 65 | +``` |
| 66 | +operator-lifecycle-manager/ |
| 67 | +├── cmd/ # Entry point binaries |
| 68 | +│ ├── catalog/ # Catalog Operator main |
| 69 | +│ ├── olm/ # OLM Operator main |
| 70 | +│ ├── package-server/ # Package API server |
| 71 | +│ └── copy-content/ # Content copy utility |
| 72 | +│ |
| 73 | +├── pkg/ # Core implementation |
| 74 | +│ ├── api/ # API client and wrappers |
| 75 | +│ │ ├── client/ # Generated Kubernetes clients |
| 76 | +│ │ └── wrappers/ # Client wrapper utilities |
| 77 | +│ │ |
| 78 | +│ ├── controller/ # Main controllers |
| 79 | +│ │ ├── bundle/ # Bundle lifecycle controller |
| 80 | +│ │ ├── install/ # Installation controller |
| 81 | +│ │ ├── operators/ # Operator/CSV controllers (OLM Operator) |
| 82 | +│ │ └── registry/ # Catalog/registry controllers (Catalog Operator) |
| 83 | +│ │ |
| 84 | +│ ├── lib/ # Shared libraries and utilities |
| 85 | +│ │ ├── catalogsource/ # CatalogSource utilities |
| 86 | +│ │ ├── csv/ # CSV manipulation utilities |
| 87 | +│ │ ├── operatorclient/ # Operator client abstractions |
| 88 | +│ │ ├── operatorlister/ # Informer-based listers |
| 89 | +│ │ ├── operatorstatus/ # Status management |
| 90 | +│ │ ├── ownerutil/ # Owner reference utilities |
| 91 | +│ │ ├── queueinformer/ # Queue-based informers |
| 92 | +│ │ ├── scoped/ # Scoped client for multi-tenancy |
| 93 | +│ │ └── [other utilities] |
| 94 | +│ │ |
| 95 | +│ ├── metrics/ # Prometheus metrics |
| 96 | +│ └── package-server/ # Package server implementation |
| 97 | +│ |
| 98 | +├── test/ # Testing infrastructure |
| 99 | +│ ├── e2e/ # End-to-end tests |
| 100 | +│ └── images/ # Test container images |
| 101 | +│ |
| 102 | +├── doc/ # Documentation |
| 103 | +│ ├── design/ # Architecture and design docs |
| 104 | +│ └── contributors/ # Contributor guides |
| 105 | +│ |
| 106 | +└── vendor/ # Vendored dependencies |
| 107 | +``` |
| 108 | + |
| 109 | +## Key Packages and Their Responsibilities |
| 110 | + |
| 111 | +### Controllers (`pkg/controller/`) |
| 112 | + |
| 113 | +#### `pkg/controller/operators/` |
| 114 | +The heart of the OLM Operator. Contains the CSV controller that manages the complete operator lifecycle. |
| 115 | + |
| 116 | +**Key files**: |
| 117 | +- `olm/operator.go` - Main OLM operator reconciler |
| 118 | +- `olm/csv.go` - CSV reconciliation logic |
| 119 | +- `catalog/operator.go` - Catalog operator reconciler |
| 120 | + |
| 121 | +#### `pkg/controller/registry/` |
| 122 | +Registry and catalog management for the Catalog Operator. |
| 123 | + |
| 124 | +**Key files**: |
| 125 | +- `reconciler/reconciler.go` - CatalogSource reconciliation |
| 126 | +- `grpc/source.go` - gRPC catalog source handling |
| 127 | + |
| 128 | +#### `pkg/controller/install/` |
| 129 | +Manages the installation of resources defined in InstallPlans. |
| 130 | + |
| 131 | +### Libraries (`pkg/lib/`) |
| 132 | + |
| 133 | +#### `pkg/lib/operatorclient/` |
| 134 | +Abstraction layer over Kubernetes clients providing OLM-specific operations. |
| 135 | + |
| 136 | +#### `pkg/lib/operatorlister/` |
| 137 | +Informer-based listers for efficient caching and querying of OLM resources. |
| 138 | + |
| 139 | +#### `pkg/lib/queueinformer/` |
| 140 | +Queue-based informer pattern used throughout OLM controllers for event-driven reconciliation. |
| 141 | + |
| 142 | +#### `pkg/lib/ownerutil/` |
| 143 | +Owner reference management ensuring proper garbage collection of OLM-managed resources. |
| 144 | + |
| 145 | +#### `pkg/lib/scoped/` |
| 146 | +Scoped clients that respect OperatorGroup namespace boundaries for multi-tenancy. |
| 147 | + |
| 148 | +## Development Workflow |
| 149 | + |
| 150 | +### Building |
| 151 | +```bash |
| 152 | +make build # Build all binaries |
| 153 | +make image # Build container image |
| 154 | +make local-build # Build with 'local' tag |
| 155 | +``` |
| 156 | + |
| 157 | +### Testing |
| 158 | +```bash |
| 159 | +make unit # Unit tests with setup-envtest |
| 160 | +make e2e # E2E tests (requires cluster) |
| 161 | +make e2e-local # Build + deploy + e2e locally |
| 162 | +make coverage # Unit tests with coverage |
| 163 | +``` |
| 164 | + |
| 165 | +### Code Generation |
| 166 | +```bash |
| 167 | +make gen-all # Generate all code (clients, mocks, manifests) |
| 168 | +make codegen # Generate K8s clients and deep-copy methods |
| 169 | +make mockgen # Generate test mocks |
| 170 | +make manifests # Copy CRD manifests from operator-framework/api |
| 171 | +``` |
| 172 | + |
| 173 | +### Local Development |
| 174 | +```bash |
| 175 | +make run-local # Complete local setup |
| 176 | +# OR step-by-step: |
| 177 | +make kind-create # Create kind cluster (kind-olmv0) |
| 178 | +make cert-manager-install |
| 179 | +make deploy # Deploy OLM to cluster |
| 180 | +``` |
| 181 | + |
| 182 | +## Key Design Patterns |
| 183 | + |
| 184 | +### Control Loop State Machines |
| 185 | + |
| 186 | +**CSV Lifecycle**: |
| 187 | +``` |
| 188 | +None → Pending → InstallReady → Installing → Succeeded |
| 189 | + ↑ ↓ |
| 190 | + ←----------←------←-------Failed |
| 191 | +``` |
| 192 | + |
| 193 | +**InstallPlan Lifecycle**: |
| 194 | +``` |
| 195 | +None → Planning → RequiresApproval → Installing → Complete |
| 196 | + ↓ ↓ |
| 197 | + ←-------←-------Failed |
| 198 | +``` |
| 199 | + |
| 200 | +**Subscription Lifecycle**: |
| 201 | +``` |
| 202 | +None → UpgradeAvailable → UpgradePending → AtLatestKnown |
| 203 | + ↑ ↓ |
| 204 | + ←-----------←-----------←-------------← |
| 205 | +``` |
| 206 | + |
| 207 | +### Dependency Resolution |
| 208 | +- CSVs declare owned CRDs (what they provide) and required CRDs (what they need) |
| 209 | +- Catalog Operator resolves transitive dependencies via graph traversal |
| 210 | +- InstallPlans capture the complete dependency closure |
| 211 | +- Resolution is based on (Group, Version, Kind) - no version pinning |
| 212 | + |
| 213 | +### Catalog and Channel Model |
| 214 | +``` |
| 215 | +Package (e.g., "etcd") |
| 216 | + ├── Channel: "stable" → CSV v0.9.4 → CSV v0.9.3 → CSV v0.9.2 |
| 217 | + ├── Channel: "alpha" → CSV v0.10.0 |
| 218 | + └── Channel: "beta" → CSV v0.9.4 |
| 219 | +``` |
| 220 | + |
| 221 | +Subscriptions track a channel; OLM follows the replacement chain to upgrade operators. |
| 222 | + |
| 223 | +## Testing Strategy |
| 224 | + |
| 225 | +### Unit Tests |
| 226 | +- Use `setup-envtest` for real Kubernetes API behavior |
| 227 | +- Race detection enabled by default (`CGO_ENABLED=1`) |
| 228 | +- Mock generation via `counterfeiter` and `gomock` |
| 229 | + |
| 230 | +### E2E Tests |
| 231 | +- Full cluster testing with Ginkgo/Gomega BDD framework |
| 232 | +- Test images in `test/images/` hosted on quay.io/olmtest |
| 233 | +- Default timeout: 90 minutes (configurable via `E2E_TIMEOUT`) |
| 234 | +- Uses kind cluster named `kind-olmv0` |
| 235 | + |
| 236 | +### Integration Tests |
| 237 | +- Bundle and catalog integration testing |
| 238 | +- Upgrade path validation |
| 239 | +- Multi-tenant scenario testing |
| 240 | + |
| 241 | +## Important Dependencies |
| 242 | + |
| 243 | +| Dependency | Version | Purpose | |
| 244 | +|------------|---------|---------| |
| 245 | +| kubernetes | v0.34.1 | Core K8s libraries | |
| 246 | +| controller-runtime | v0.22.2 | Controller framework | |
| 247 | +| operator-framework/api | v0.35.0 | OLM API definitions | |
| 248 | +| operator-registry | v1.60.0 | Catalog/bundle tooling | |
| 249 | +| ginkgo/gomega | v2.26.0 / v1.38.2 | BDD testing | |
| 250 | + |
| 251 | +## Common Tasks for AI Agents |
| 252 | + |
| 253 | +### Understanding Operator Installation Flow |
| 254 | +1. User creates Subscription pointing to catalog/package/channel |
| 255 | +2. Catalog Operator queries catalog for latest CSV in channel |
| 256 | +3. Catalog Operator creates InstallPlan with resolved dependencies |
| 257 | +4. Upon approval, Catalog Operator creates CRDs and CSV |
| 258 | +5. OLM Operator detects CSV, validates requirements, creates Deployment/RBAC |
| 259 | +6. CSV transitions through: Pending → InstallReady → Installing → Succeeded |
| 260 | + |
| 261 | +### Debugging Installation Issues |
| 262 | +- Check CSV status and conditions: `kubectl get csv -o yaml` |
| 263 | +- Examine InstallPlan: `kubectl get ip -o yaml` |
| 264 | +- Review operator logs: OLM Operator and Catalog Operator pods in `olm` namespace |
| 265 | +- Verify OperatorGroup configuration for namespace scoping |
| 266 | + |
| 267 | +### Adding New Functionality |
| 268 | +**REMINDER**: This repository is in maintenance mode - only critical fixes accepted! |
| 269 | + |
| 270 | +For understanding existing code: |
| 271 | +1. Controllers follow controller-runtime patterns with Reconcile() methods |
| 272 | +2. Use informers and listers from `pkg/lib/operatorlister` |
| 273 | +3. Queue-based event handling via `pkg/lib/queueinformer` |
| 274 | +4. Always respect OperatorGroup namespace scoping |
| 275 | + |
| 276 | +### Code Generation |
| 277 | +Most code is generated - don't hand-edit: |
| 278 | +- Client code: Generated from CRDs using k8s.io/code-generator |
| 279 | +- Deep-copy methods: Auto-generated for all API types |
| 280 | +- Mocks: Generated via counterfeiter/gomock |
| 281 | +- CRD manifests: Copied from operator-framework/api repository |
| 282 | + |
| 283 | +Always run `make gen-all` after API changes. |
| 284 | + |
| 285 | +## Navigation Tips |
| 286 | + |
| 287 | +### Finding Controllers |
| 288 | +- OLM Operator controllers: `pkg/controller/operators/` |
| 289 | +- Catalog Operator controllers: `pkg/controller/registry/`, subscription/installplan logic in `pkg/controller/operators/catalog/` |
| 290 | + |
| 291 | +### Finding API Definitions |
| 292 | +- CRDs are defined in operator-framework/api (external dependency) |
| 293 | +- Clients are in `pkg/api/client/` |
| 294 | +- Listers are in `pkg/lib/operatorlister/` |
| 295 | + |
| 296 | +### Finding Business Logic |
| 297 | +- CSV installation: `pkg/controller/operators/olm/` |
| 298 | +- Dependency resolution: `pkg/controller/registry/resolver/` |
| 299 | +- Catalog management: `pkg/controller/registry/reconciler/` |
| 300 | +- InstallPlan execution: `pkg/controller/install/` |
| 301 | + |
| 302 | +### Finding Utilities |
| 303 | +- Owner references: `pkg/lib/ownerutil/` |
| 304 | +- Scoped clients: `pkg/lib/scoped/` |
| 305 | +- Operator clients: `pkg/lib/operatorclient/` |
| 306 | +- Queue informers: `pkg/lib/queueinformer/` |
| 307 | + |
| 308 | +## Anti-Patterns to Avoid |
| 309 | + |
| 310 | +1. **Don't bypass OperatorGroup scoping** - Always use scoped clients for multi-tenancy |
| 311 | +2. **Don't modify generated code** - Edit source (CRDs, annotations) and regenerate |
| 312 | +3. **Don't skip approval for InstallPlans** - Respect manual approval settings |
| 313 | +4. **Don't create CSVs directly** - Use Subscriptions/Catalogs for proper lifecycle |
| 314 | +5. **Don't ignore owner references** - Critical for garbage collection |
| 315 | + |
| 316 | +## Resources and Links |
| 317 | + |
| 318 | +- [OLM Documentation](https://olm.operatorframework.io/) |
| 319 | +- [Architecture Doc](doc/design/architecture.md) |
| 320 | +- [Philosophy](doc/design/philosophy.md) |
| 321 | +- [Design Proposals](doc/contributors/design-proposals/) |
| 322 | +- [Operator Framework Community](https://github.com/operator-framework/community) |
| 323 | +- [OperatorHub.io](https://operatorhub.io/) - Public operator catalog |
| 324 | + |
| 325 | +## Quick Reference |
| 326 | + |
| 327 | +### Resource Short Names |
| 328 | +```bash |
| 329 | +kubectl get csv # ClusterServiceVersions |
| 330 | +kubectl get sub # Subscriptions |
| 331 | +kubectl get ip # InstallPlans |
| 332 | +kubectl get catsrc # CatalogSources |
| 333 | +kubectl get og # OperatorGroups |
| 334 | +``` |
| 335 | + |
| 336 | +### Common Build Targets |
| 337 | +```bash |
| 338 | +make build build-utils # Build all binaries |
| 339 | +make test unit e2e # Run tests |
| 340 | +make lint verify # Code quality |
| 341 | +make gen-all # Generate everything |
| 342 | +make run-local # Full local dev environment |
| 343 | +``` |
| 344 | + |
| 345 | +### Tool Management |
| 346 | +Tools are managed via **bingo** (`.bingo/Variables.mk`) for reproducible builds. All tools are version-pinned. |
| 347 | + |
| 348 | +## Contributing |
| 349 | + |
| 350 | +See [CONTRIBUTING.md](CONTRIBUTING.md) and [CLAUDE.md](CLAUDE.md) for detailed guidelines. |
| 351 | + |
| 352 | +**Remember**: OLM v0 is in maintenance mode - only critical security fixes and outage issues are accepted! |
0 commit comments