.coderabbit.yaml

# yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json
language: en-US
early_access: false
enable_free_tier: true

reviews:
  profile: chill
  request_changes_workflow: false
  high_level_summary: true
  poem: false
  in_progress_fortune: false
  review_status: true
  collapse_walkthrough: false

  path_filters:
    - "!**/vendor/**"
    - "!vendor/**"
    - "!go.sum"

  auto_review:
    enabled: true
    drafts: true

  # ── Path instructions ────────────────────────────────────────
  path_instructions:

    # ── Go code — general ────────────────────────────────────────
    - path: "**/*.go"
      instructions: |
        Go security and best practices:
        - Never ignore error returns; wrap errors with context using fmt.Errorf("...: %w", err)
        - Use stdlib crypto/* and golang.org/x/crypto; avoid third-party crypto libraries
        - Integer overflow: bounds-check user-supplied sizes
        - context.Context for cancellation and timeouts on all network/gRPC operations
        - Nil pointer checks on CNI args, NetConf, and DelegateNetConf structs
          before accessing fields — CNI runtime data can be incomplete
        - Use structured logging (klog v1/v2) with appropriate verbosity levels
        - Network namespace operations require careful error handling —
          always restore the original namespace on error paths
        - CNI handler functions (cmdAdd, cmdDel, cmdCheck) may return
          contextual or wrapped Go errors; the skel.PluginMainFuncs
          dispatcher handles CNI JSON error serialization

    # ── Core multus logic ────────────────────────────────────────
    - path: "pkg/multus/**/*.go"
      instructions: |
        Core CNI plugin logic review:
        - Verify CNI spec compliance across supported versions (0.2.0, 0.4.0, 1.0.0)
        - Delegate plugin invocation must handle failures gracefully —
          rollback already-added interfaces if a later delegate fails in cmdAdd
        - Network selection from annotations must validate the annotation format
          and handle malformed NetworkSelectionElement entries
        - Network namespace operations: always defer namespace restoration
          to prevent leaving the process in the wrong namespace on error
        - Check that the default network delegate is always invoked first
        - Verify proper handling of prevResult for chained CNI plugins
        - Resource (device) allocation tracking via checkpoint must be atomic

    # ── Thick plugin server ──────────────────────────────────────
    - path: "pkg/server/**/*.go"
      instructions: |
        Thick plugin daemon and gRPC server review:
        - gRPC server must handle concurrent requests safely
        - Socket file cleanup on shutdown: verify the Unix domain socket
          is removed on graceful and ungraceful termination
        - Configuration file watching (fsnotify): verify events are debounced
          and do not cause config reload storms
        - TLS/certificate handling for metrics endpoint: verify proper rotation
        - Server configuration must be validated at startup, not at request time
        - API compatibility between shim and daemon must be maintained —
          check that request/response types match across versions
        - Health check and readiness endpoints should reflect actual daemon state

    # ── Kubernetes client ────────────────────────────────────────
    - path: "pkg/k8sclient/**/*.go"
      instructions: |
        Kubernetes client review:
        - Kubeconfig file handling: verify secure file permissions (0600)
          and no secrets logged
        - Network Attachment Definition (NAD) lookups must handle not-found
          errors distinctly from other API errors
        - Client caching: verify informer/lister patterns are used correctly
          to avoid excessive API server load
        - Namespace scoping: ensure NAD lookups respect the pod's namespace
          and the default network namespace

    # ── Types and configuration ──────────────────────────────────
    - path: "pkg/types/**/*.go"
      instructions: |
        Type definitions and configuration review:
        - NetConf changes must maintain backward compatibility with existing
          CNI configuration files — new fields should have sensible zero values
        - NetworkSelectionElement: verify JSON tag naming follows CNI conventions
        - DelegateNetConf: ensure all delegate plugin configuration fields
          are properly forwarded without loss
        - Validate that type changes are reflected in all serialization paths
          (JSON marshaling/unmarshaling)

    # ── Network utilities ────────────────────────────────────────
    - path: "pkg/netutils/**/*.go"
      instructions: |
        Network utility review:
        - Network namespace operations are privileged — verify proper
          error handling and namespace restoration on all code paths
        - Interface naming must comply with Linux kernel limits (15 chars max)
        - Device operations (link set, address set) via netlink must handle
          EBUSY and other transient errors with appropriate retries
        - Verify no file descriptor leaks when opening network namespaces

    # ── DRA and kubelet client ───────────────────────────────────
    - path: "pkg/{draclient,kubeletclient}/**/*.go"
      instructions: |
        Dynamic Resource Allocation and kubelet client review:
        - gRPC connections to kubelet must use proper timeouts and retry logic
        - PodResources API responses may be incomplete — handle nil/empty
          device lists gracefully
        - DRA resource claims must be validated before use
        - Connection errors should be distinguished from data errors
          for proper error reporting

    # ── Command binaries ─────────────────────────────────────────
    - path: "cmd/**/*.go"
      instructions: |
        Binary entry point review:
        - CNI plugin binaries (multus, multus-shim, passthru-cni) must follow
          the CNI specification for stdin/stdout/stderr handling
        - Daemon binary (multus-daemon) must handle signals (SIGTERM, SIGINT)
          for graceful shutdown including socket cleanup
        - Installer binary must validate target paths and handle permission
          errors when copying binaries to host CNI directories
        - Entry point binaries must not panic — recover and return proper
          error codes/messages

    # ── Kubernetes/OpenShift manifests ────────────────────────────
    - path: "**/*.{yaml,yml}"
      instructions: |
        If this is a Kubernetes/OpenShift manifest:
        - securityContext: runAsNonRoot, readOnlyRootFilesystem,
          allowPrivilegeEscalation: false where applicable
        - Drop ALL capabilities, add only what is required
        - Resource limits (cpu, memory) on every container
        - RBAC: least privilege; no cluster-admin for workloads
        - Liveness + readiness probes defined
        - automountServiceAccountToken: false unless needed

        Multus-specific manifest review:
        - DaemonSet manifests: verify volume mounts for CNI bin directory
          (/opt/cni/bin) and CNI config directory (/etc/cni/net.d) are correct
        - Thick plugin DaemonSet: verify Unix socket mount for daemon communication
        - Host path volumes require proper mount propagation settings
        - Init containers for CNI binary installation must complete before
          the main container starts
        - CRI-O specific manifests: verify socket path differences from containerd

    # ── Deployment manifests ─────────────────────────────────────
    - path: "deployments/**/*.{yaml,yml}"
      instructions: |
        Deployment manifest review:
        - Three deployment modes: thin (multus-daemonset.yml),
          thick (multus-daemonset-thick.yml), CRI-O (multus-daemonset-crio.yml)
        - Verify consistency across all three manifests for shared configuration
        - DaemonSet update strategy should be RollingUpdate with appropriate
          maxUnavailable settings for CNI plugins (node disruption impact)
        - ServiceAccount and RBAC must allow reading NetworkAttachmentDefinitions
          and pods but nothing more
        - Tolerations should include control-plane/master taints for full coverage

    # ── Dockerfiles / container images ───────────────────────────
    - path: "images/{Dockerfile,Dockerfile.*}"
      instructions: |
        Container image security:
        - Multi-stage builds preferred; no build tools in final image
        - USER non-root where possible (note: CNI plugins may need root for
          network namespace operations)
        - COPY specific files, not entire context
        - No secrets in ENV, ARG, or COPY
        - No package manager cache in final layer
        - This project has 4 container images:
          Dockerfile (thin plugin, distroless base),
          Dockerfile.thick (thick plugin, Debian slim),
          Dockerfile.debug (debug, distroless debug base),
          Dockerfile.openshift (OpenShift/OKD build)
        - Verify ENTRYPOINT matches the intended binary for each image variant

    # ── E2E tests ────────────────────────────────────────────────
    - path: "e2e/**/*.sh"
      instructions: |
        E2E test scripts review (shell-based, Kind clusters):
        - Scripts must use 'set -e' for fail-fast behavior
        - Cleanup of Kind clusters and resources on test failure (trap handlers)
        - kubectl commands should use explicit timeouts and --wait flags
        - Network connectivity tests should retry with backoff for pod readiness
        - Verify tests cover both thin and thick plugin deployment modes
        - Template generation (j2cli/sed) must produce valid YAML
        - Check for hardcoded image tags that should be parameterized

    # ── Unit tests ───────────────────────────────────────────────
    - path: "{cmd,pkg}/**/*_test.go"
      instructions: |
        Unit test quality (Ginkgo v2 + Gomega):
        - Test names must be stable and deterministic — no dynamic values
          in It/Describe/Context titles
        - CNI version-specific tests (020, 040, 100): verify coverage
          across all supported CNI spec versions
        - Mock external dependencies (Kubernetes API, network namespace ops)
          rather than requiring real cluster or root access
        - Test error paths, not just happy paths
        - Network namespace tests may require sudo — verify CI compatibility
        - Assertions should include meaningful failure messages
        - Verify test cleanup of any created network namespaces or interfaces

    # ── Hack/build scripts ───────────────────────────────────────
    - path: "hack/**/*.sh"
      instructions: |
        Build and test scripts review:
        - Scripts must be POSIX-compatible or explicitly require bash
        - build-go.sh: verify cross-compilation settings (GOOS, GOARCH)
          produce correct binaries for all supported platforms
          (386, amd64, arm, arm64, ppc64le, s390x)
        - test-go.sh: verify race detector and coverage flags are correct
        - Scripts should exit with appropriate error codes on failure

    # ── Supply chain & dependencies ──────────────────────────────
    - path: "**/go.mod"
      instructions: |
        Supply chain security:
        - New dependencies: justify the need, check license compatibility
        - Pin exact versions; verify no unnecessary indirect dependencies added
        - Flag known CVEs (cross-ref osv.dev)
        - Check for replace directives pointing to forks — ensure they are
          intentional and documented
        - This project uses gopkg.in import path (gopkg.in/k8snetworkplumbingwg/multus-cni.v4)
          — verify import path consistency
        - Key dependencies: containernetworking/cni, containernetworking/plugins,
          k8snetworkplumbingwg/network-attachment-definition-client,
          vishvananda/netlink, k8s.io/api, k8s.io/client-go, k8s.io/kubelet

    # ── CI/CD & GitHub Actions ───────────────────────────────────
    - path: ".github/workflows/**/*"
      instructions: |
        CI/CD security:
        - Pin actions by full SHA, not tag
        - No secrets in logs; mask sensitive outputs
        - Least privilege: minimize GITHUB_TOKEN permissions
        - No pull_request_target with checkout of PR head
        - Verify test.yml includes Revive lint, go fmt, go vet, and unit tests
        - Build workflow must cover all architectures:
          386, amd64, arm, arm64, ppc64le, s390x
        - Image build workflow: verify Trivy scan is included
        - Kind E2E workflow: verify both thin and thick plugin modes are tested

    # ── Cryptography files ───────────────────────────────────────
    - path: "**/*{crypt,cipher,sign,hash,tls,ssl,cert,key,token}*"
      instructions: |
        Cryptographic security:
        - Banned: MD5, SHA1, DES, RC4, 3DES, Blowfish, ECB mode
        - Symmetric: AES-256-GCM or ChaCha20-Poly1305
        - Signing: Ed25519 or ECDSA P-256+
        - Constant-time comparison for all secret/token data
        - No custom crypto; use vetted libraries only
        - Certificate generation and approval (cert-approver, kubeconfig_generator):
          verify minimum TLS version (1.2+) and proper certificate rotation

    # ── GoReleaser configuration ─────────────────────────────────
    - path: ".goreleaser.yml"
      instructions: |
        GoReleaser configuration review:
        - Verify build targets match the supported architectures
        - Check that binary names match the expected CNI plugin names
        - Verify release artifacts include all necessary binaries

  # ── Security scanners ────────────────────────────────────────
  tools:
    gitleaks:
      enabled: true
    semgrep:
      enabled: true
    checkov:
      enabled: true
    hadolint:
      enabled: true
    trivy:
      enabled: true
    osvScanner:
      enabled: true
    actionlint:
      enabled: true
    ast-grep:
      essential_rules: true

# ── Knowledge base ───────────────────────────────────────────
knowledge_base:
  code_guidelines:
    enabled: true
    filePatterns:
      - "**/CONTRIBUTING.md"
  issues:
    scope: "auto"
  pull_requests:
    scope: "auto"

chat:
  auto_reply: true
  art: false