Trigger uninstalls through a new annotation

Previous to this, a finalizer on the Deployment was added so that if the
Deployment was deleted, it would handle immediate clean up. This doesn't
handle the common case where the config-policy-controller ManagedClusterAddOn
is deleted, which causes the ManifestWork to be deleted, which triggers
all Configuration Policy controller deployment artifacts, including the
service account.

A new approach is taken so that a new annotation of
policy.open-cluster-management.io/uninstalling=true is set on the
Deployment to indicate that the Configuration Policy controller should
remove all finalizers because it's getting deleted.

The Policy Addon controller will be updated so that when the
config-policy-controller ManagedClusterAddOn object is deleted, a
finalizer will prevent it and a Pod will run on the managed cluster with
the new `trigger-uninstall` subcommand. This sets the uninstalling
annotation on the Deployment and then waits until all
ConfigurationPolicy finalizers have been removed. Once the command ends,
the Pod exits, and the ManagedClusterAddOn object's finalizer is removed
and the uninstall proceeds.

Relates:
https://issues.redhat.com/browse/ACM-3233
https://issues.redhat.com/browse/ACM-2923

Signed-off-by: mprahl <mprahl@users.noreply.github.com>

.github/workflows/kind.yml

@@ -94,6 +94,11 @@ jobs:
  make build-images
  make kind-deploy-controller-dev
 
+ - name: E2E tests that require the controller running in a cluster
+ run: |
+ export GOPATH=$(go env GOPATH)
+ make e2e-test-running-in-cluster
+
  - name: Debug
  if: ${{ failure() }}
  run: |

Makefile

@@ -306,15 +306,19 @@ e2e-test: e2e-dependencies
  $(GINKGO) -v --fail-fast --slow-spec-threshold=10s $(E2E_TEST_ARGS) test/e2e
 
 .PHONY: e2e-test-coverage
-e2e-test-coverage: E2E_TEST_ARGS = --json-report=report_e2e.json --label-filter="!hosted-mode" --output-dir=.
+e2e-test-coverage: E2E_TEST_ARGS = --json-report=report_e2e.json --label-filter="!hosted-mode && !running-in-cluster" --output-dir=.
 e2e-test-coverage: e2e-run-instrumented e2e-test e2e-stop-instrumented
 
 .PHONY: e2e-test-hosted-mode-coverage
-e2e-test-hosted-mode-coverage: E2E_TEST_ARGS = --json-report=report_e2e_hosted_mode.json --label-filter="hosted-mode" --output-dir=.
+e2e-test-hosted-mode-coverage: E2E_TEST_ARGS = --json-report=report_e2e_hosted_mode.json --label-filter="hosted-mode && !running-in-cluster" --output-dir=.
 e2e-test-hosted-mode-coverage: COVERAGE_E2E_OUT = coverage_e2e_hosted_mode.out
 e2e-test-hosted-mode-coverage: export TARGET_KUBECONFIG_PATH = $(PWD)/kubeconfig_managed2
 e2e-test-hosted-mode-coverage: e2e-run-instrumented e2e-test e2e-stop-instrumented
 
+.PHONY: e2e-test-running-in-cluster
+e2e-test-running-in-cluster: E2E_TEST_ARGS = --label-filter="running-in-cluster"
+e2e-test-running-in-cluster: e2e-test
+
 .PHONY: e2e-build-instrumented
 e2e-build-instrumented:
  go test -covermode=atomic -coverpkg=$(shell cat go.mod | head -1 | cut -d ' ' -f 2)/... -c -tags e2e ./ -o build/_output/bin/$(IMG)-instrumented

build/Dockerfile

@@ -24,7 +24,7 @@ COPY --from=builder ${REPO_PATH}/build/_output/bin/${COMPONENT} ${OPERATOR}
 COPY --from=builder ${REPO_PATH}/build/bin /usr/local/bin
 RUN /usr/local/bin/user_setup
 
-ENTRYPOINT ["/usr/local/bin/entrypoint"]
+ENTRYPOINT ["/usr/local/bin/entrypoint", "controller"]
 
 RUN microdnf update && \
  microdnf clean all

controllers/configurationpolicy_controller.go

@@ -163,20 +163,11 @@ func (r *ConfigurationPolicyReconciler) PeriodicallyExecConfigPolicies(
  }
 
  const waiting = 10 * time.Minute
- var exiting bool
- // Loop twice after exit condition is received to account for race conditions and retries.
- loopsAfterExit := 2
 
- for !exiting || (exiting && loopsAfterExit > 0) {
- start := time.Now()
-
- select {
- case <-ctx.Done():
- exiting = true
- loopsAfterExit--
- default:
- }
+ exiting := false
 
+ for !exiting {
+ start := time.Now()
  policiesList := policyv1.ConfigurationPolicyList{}
 
  var skipLoop bool
@@ -206,28 +197,6 @@ func (r *ConfigurationPolicyReconciler) PeriodicallyExecConfigPolicies(
  skipLoop = true
  }
 
- needDeploymentFinalizer := false
-
- for i := range policiesList.Items {
- plc := policiesList.Items[i]
-
- if objHasFinalizer(&plc, pruneObjectFinalizer) {
- needDeploymentFinalizer = true
-
- break
- }
- }
-
- if err := r.manageDeploymentFinalizer(needDeploymentFinalizer); err != nil {
- if errors.Is(err, common.ErrNoNamespace) || errors.Is(err, common.ErrRunLocal) {
- log.Info("Not managing the controller's deployment finalizer because it is running locally")
- } else {
- log.Error(err, "Failed to manage the controller's deployment finalizer, skipping loop")
-
- skipLoop = true
- }
- }
-
  cleanupImmediately, err := r.cleanupImmediately()
  if err != nil {
  log.Error(err, "Failed to determine if it's time to cleanup immediately")
@@ -281,6 +250,12 @@ func (r *ConfigurationPolicyReconciler) PeriodicallyExecConfigPolicies(
  log.V(2).Info("Sleeping before reprocessing the configuration policies", "seconds", sleepTime)
  time.Sleep(sleepTime)
  }
+
+ select {
+ case <-ctx.Done():
+ exiting = true
+ default:
+ }
  }
 }
 
@@ -618,10 +593,10 @@ func (r *ConfigurationPolicyReconciler) cleanUpChildObjects(plc policyv1.Configu
 
 // cleanupImmediately returns true when the cluster is in a state where configurationpolicies should
 // be removed as soon as possible, ignoring the pruneObjectBehavior of the policies. This is the
-// case when the CRD or the controller's deployment are already being deleted.
+// case when the controller is being uninstalled or the CRD is being deleted.
 func (r *ConfigurationPolicyReconciler) cleanupImmediately() (bool, error) {
- deployDeleting, deployErr := r.deploymentIsDeleting()
- if deployErr == nil && deployDeleting {
+ beingUninstalled, beingUninstalledErr := r.isBeingUninstalled()
+ if beingUninstalledErr == nil && beingUninstalled {
  return true, nil
  }
 
@@ -630,36 +605,16 @@ func (r *ConfigurationPolicyReconciler) cleanupImmediately() (bool, error) {
  return true, nil
  }
 
- if deployErr == nil && defErr == nil {
+ if beingUninstalledErr == nil && defErr == nil {
  // if either was deleting, we would've already returned.
  return false, nil
  }
 
  // At least one had an unexpected error, so the decision can't be made right now
  //nolint:errorlint // we can't choose just one of the errors to "correctly" wrap
- return false, fmt.Errorf("deploymentIsDeleting error: '%v', definitionIsDeleting error: '%v'",
- deployErr, defErr)
-}
-
-func (r *ConfigurationPolicyReconciler) deploymentIsDeleting() (bool, error) {
- key, keyErr := common.GetOperatorNamespacedName()
- if keyErr != nil {
- if errors.Is(keyErr, common.ErrNoNamespace) || errors.Is(keyErr, common.ErrRunLocal) {
- // running locally
- return false, nil
- }
-
- return false, keyErr
- }
-
- deployment := appsv1.Deployment{}
-
- err := r.Get(context.TODO(), key, &deployment)
- if err != nil {
- return false, err
- }
-
- return deployment.DeletionTimestamp != nil, nil
+ return false, fmt.Errorf(
+ "isBeingUninstalled error: '%v', definitionIsDeleting error: '%v'", beingUninstalledErr, defErr,
+ )
 }
 
 func (r *ConfigurationPolicyReconciler) definitionIsDeleting() (bool, error) {
@@ -2808,32 +2763,23 @@ func convertPolicyStatusToString(plc *policyv1.ConfigurationPolicy) (results str
  return result
 }
 
-func (r *ConfigurationPolicyReconciler) manageDeploymentFinalizer(shouldBeSet bool) error {
+func (r *ConfigurationPolicyReconciler) isBeingUninstalled() (bool, error) {
  key, err := common.GetOperatorNamespacedName()
  if err != nil {
- return err
+ // Running locally
+ if errors.Is(err, common.ErrNoNamespace) || errors.Is(err, common.ErrRunLocal) {
+ return false, nil
+ }
+
+ return false, err
  }
 
  deployment := appsv1.Deployment{}
  if err := r.Client.Get(context.TODO(), key, &deployment); err != nil {
- return err
- }
-
- if objHasFinalizer(&deployment, pruneObjectFinalizer) {
- if shouldBeSet {
- return nil
- }
-
- deployment.SetFinalizers(removeObjFinalizer(&deployment, pruneObjectFinalizer))
- } else {
- if !shouldBeSet {
- return nil
- }
-
- deployment.SetFinalizers(addObjFinalizer(&deployment, pruneObjectFinalizer))
+ return false, err
  }
 
- return r.Update(context.TODO(), &deployment)
+ return deployment.Annotations[common.UninstallingAnnotation] == "true", nil
 }
 
 func recoverFlow() {

controllers/configurationpolicy_utils.go

@@ -549,7 +549,6 @@ func addObjFinalizer(obj metav1.Object, finalizer string) []string {
  return append(obj.GetFinalizers(), finalizer)
 }
 
-// nolint: unparam
 func removeObjFinalizer(obj metav1.Object, finalizer string) []string {
  result := []string{}
 

deploy/manager/manager.yaml

@@ -20,6 +20,7 @@ spec:
  command:
  - config-policy-controller
  args:
+ - "controller"
  - "--enable-lease=true"
  - "--log-level=2"
  - "--v=0"

deploy/operator.yaml

@@ -45,6 +45,7 @@ spec:
  spec:
  containers:
  - args:
+ - controller
  - --enable-lease=true
  - --log-level=2
  - --v=0

main.go

@@ -43,6 +43,7 @@ import (
  policyv1 "open-cluster-management.io/config-policy-controller/api/v1"
  "open-cluster-management.io/config-policy-controller/controllers"
  "open-cluster-management.io/config-policy-controller/pkg/common"
+ "open-cluster-management.io/config-policy-controller/pkg/triggeruninstall"
  "open-cluster-management.io/config-policy-controller/version"
 )
 
@@ -74,51 +75,52 @@ func main() {
  }
 
  zflags.Bind(flag.CommandLine)
- pflag.CommandLine.AddGoFlagSet(flag.CommandLine)
+
+ controllerFlagSet := pflag.NewFlagSet("controller", pflag.ExitOnError)
 
  var clusterName, hubConfigPath, targetKubeConfig, metricsAddr, probeAddr string
  var frequency uint
  var decryptionConcurrency, evaluationConcurrency uint8
  var enableLease, enableLeaderElection, legacyLeaderElection, enableMetrics bool
 
- pflag.UintVar(&frequency, "update-frequency", 10,
+ controllerFlagSet.UintVar(&frequency, "update-frequency", 10,
  "The status update frequency (in seconds) of a mutation policy")
- pflag.BoolVar(&enableLease, "enable-lease", false,
+ controllerFlagSet.BoolVar(&enableLease, "enable-lease", false,
  "If enabled, the controller will start the lease controller to report its status")
- pflag.StringVar(&clusterName, "cluster-name", "acm-managed-cluster", "Name of the cluster")
- pflag.StringVar(&hubConfigPath, "hub-kubeconfig-path", "/var/run/klusterlet/kubeconfig",
+ controllerFlagSet.StringVar(&clusterName, "cluster-name", "acm-managed-cluster", "Name of the cluster")
+ controllerFlagSet.StringVar(&hubConfigPath, "hub-kubeconfig-path", "/var/run/klusterlet/kubeconfig",
  "Path to the hub kubeconfig")
- pflag.StringVar(
+ controllerFlagSet.StringVar(
  &targetKubeConfig,
  "target-kubeconfig-path",
  "",
  "A path to an alternative kubeconfig for policy evaluation and enforcement.",
  )
- pflag.StringVar(
+ controllerFlagSet.StringVar(
  &metricsAddr, "metrics-bind-address", "localhost:8383", "The address the metrics endpoint binds to.",
  )
- pflag.StringVar(&probeAddr, "health-probe-bind-address", ":8081", "The address the probe endpoint binds to.")
- pflag.BoolVar(&enableLeaderElection, "leader-elect", true,
+ controllerFlagSet.StringVar(
+ &probeAddr, "health-probe-bind-address", ":8081", "The address the probe endpoint binds to.",
+ )
+ controllerFlagSet.BoolVar(&enableLeaderElection, "leader-elect", true,
  "Enable leader election for controller manager. "+
  "Enabling this will ensure there is only one active controller manager.")
- pflag.BoolVar(&legacyLeaderElection, "legacy-leader-elect", false,
+ controllerFlagSet.BoolVar(&legacyLeaderElection, "legacy-leader-elect", false,
  "Use a legacy leader election method for controller manager instead of the lease API.")
- pflag.Uint8Var(
+ controllerFlagSet.Uint8Var(
  &decryptionConcurrency,
  "decryption-concurrency",
  5,
  "The max number of concurrent policy template decryptions",
  )
- pflag.Uint8Var(
+ controllerFlagSet.Uint8Var(
  &evaluationConcurrency,
  "evaluation-concurrency",
  // Set a low default to not add too much load to the Kubernetes API server in resource constrained deployments.
  2,
  "The max number of concurrent configuration policy evaluations",
  )
- pflag.BoolVar(&enableMetrics, "enable-metrics", true, "Disable custom metrics collection")
-
- pflag.Parse()
+ controllerFlagSet.BoolVar(&enableMetrics, "enable-metrics", true, "Disable custom metrics collection")
 
  ctrlZap, err := zflags.BuildForCtrl()
  if err != nil {
@@ -142,6 +144,24 @@ func main() {
  klog.SetLogger(zapr.NewLogger(klogZap).WithName("klog"))
  }
 
+ subcommand := ""
+ if len(os.Args) >= 2 {
+ subcommand = os.Args[1]
+ }
+
+ switch subcommand {
+ case "controller":
+ controllerFlagSet.AddGoFlagSet(flag.CommandLine)
+ _ = controllerFlagSet.Parse(os.Args[2:])
+ case "trigger-uninstall":
+ handleTriggerUninstall()
+
+ return
+ default:
+ fmt.Fprintln(os.Stderr, "expected 'controller' or 'trigger-uninstall' subcommands")
+ os.Exit(1)
+ }
+
  if evaluationConcurrency < 1 {
  panic("The --evaluation-concurrency option cannot be less than 1")
  }
@@ -372,3 +392,45 @@ func main() {
  os.Exit(1)
  }
 }
+
+func handleTriggerUninstall() {
+ triggerUninstallFlagSet := pflag.NewFlagSet("trigger-uninstall", pflag.ExitOnError)
+
+ var deploymentName, deploymentNamespace, policyNamespace string
+
+ triggerUninstallFlagSet.StringVar(
+ &deploymentName, "deployment-name", "config-policy-controller", "The name of the controller Deployment object",
+ )
+ triggerUninstallFlagSet.StringVar(
+ &deploymentNamespace,
+ "deployment-namespace",
+ "open-cluster-management-agent-addon",
+ "The namespace of the controller Deployment object",
+ )
+ triggerUninstallFlagSet.StringVar(
+ &policyNamespace, "policy-namespace", "", "The namespace of where ConfigurationPolicy objects are stored",
+ )
+ triggerUninstallFlagSet.AddGoFlagSet(flag.CommandLine)
+
+ _ = triggerUninstallFlagSet.Parse(os.Args[2:])
+
+ if deploymentName == "" || deploymentNamespace == "" || policyNamespace == "" {
+ fmt.Fprintln(os.Stderr, "--deployment-name, --deployment-namespace, --policy-namespace must all have values")
+ os.Exit(1)
+ }
+
+ terminatingCtx := ctrl.SetupSignalHandler()
+
+ // Get a config to talk to the apiserver
+ cfg, err := config.GetConfig()
+ if err != nil {
+ log.Error(err, "Failed to get config")
+ os.Exit(1)
+ }
+
+ err = triggeruninstall.TriggerUninstall(terminatingCtx, cfg, deploymentName, deploymentNamespace, policyNamespace)
+ if err != nil {
+ klog.Errorf("Failed to trigger the uninstall due to the error: %s", err)
+ os.Exit(1)
+ }
+}

main_test.go

@@ -15,8 +15,11 @@ import (
 // TestRunMain wraps the main() function in order to build a test binary and collection coverage for
 // E2E/Integration tests. Controller CLI flags are also passed in here.
 func TestRunMain(t *testing.T) {
+ args := append([]string{os.Args[1], "controller"}, os.Args[2:]...)
  os.Args = append(
- os.Args, "--leader-elect=false", fmt.Sprintf("--target-kubeconfig-path=%s", os.Getenv("TARGET_KUBECONFIG_PATH")),
+ args,
+ "--leader-elect=false",
+ fmt.Sprintf("--target-kubeconfig-path=%s", os.Getenv("TARGET_KUBECONFIG_PATH")),
  )
 
  main()

pkg/common/common.go

@@ -11,6 +11,8 @@ import (
  "k8s.io/client-go/tools/record"
 )
 
+const UninstallingAnnotation string = "policy.open-cluster-management.io/uninstalling"
+
 // CreateRecorder return recorder
 func CreateRecorder(kubeClient kubernetes.Interface, componentName string) (record.EventRecorder, error) {
  eventsScheme := runtime.NewScheme()