preserve failed unpack jobs, enforce minimum interval between failing unpack jobs

Signed-off-by: Ankita Thomas <[email protected]>

Author: ankitathomas

pkg/controller/bundle/bundle_unpacker.go

@@ -41,6 +41,13 @@ const (
 	// e.g 1m30s
 	BundleUnpackTimeoutAnnotationKey = "operatorframework.io/bundle-unpack-timeout"
 	BundleUnpackPodLabel             = "job-name"
+
+	// BundleUnpackRetryMinimumIntervalAnnotationKey sets a minimum interval to wait before
+	// attempting to recreate a failed unpack job for a bundle.
+	BundleUnpackRetryMinimumIntervalAnnotationKey = "operatorframework.io/bundle-unpack-min-retry-interval"
+
+	// bundleUnpackRefLabel is used to filter for all unpack jobs for a specific bundle.
+	bundleUnpackRefLabel = "operatorframework.io/bundle-unpack-ref"
 )
 
 type BundleUnpackResult struct {
@@ -247,6 +254,7 @@ func (c *ConfigMapUnpacker) job(cmRef *corev1.ObjectReference, bundlePath string
 	}
 	job.SetNamespace(cmRef.Namespace)
 	job.SetName(cmRef.Name)
+	job.SetLabels(map[string]string{bundleUnpackRefLabel: cmRef.Name})
 	job.SetOwnerReferences([]metav1.OwnerReference{ownerRef(cmRef)})
 	if c.runAsUser > 0 {
 		job.Spec.Template.Spec.SecurityContext.RunAsUser = &c.runAsUser
@@ -287,7 +295,7 @@ func (c *ConfigMapUnpacker) job(cmRef *corev1.ObjectReference, bundlePath string
 //go:generate go run github.com/maxbrunsfeld/counterfeiter/v6 . Unpacker
 
 type Unpacker interface {
-	UnpackBundle(lookup *operatorsv1alpha1.BundleLookup, timeout time.Duration) (result *BundleUnpackResult, err error)
+	UnpackBundle(lookup *operatorsv1alpha1.BundleLookup, timeout, retryInterval time.Duration) (result *BundleUnpackResult, err error)
 }
 
 type ConfigMapUnpacker struct {
@@ -448,7 +456,7 @@ const (
 	NotUnpackedMessage          = "bundle contents have not yet been persisted to installplan status"
 )
 
-func (c *ConfigMapUnpacker) UnpackBundle(lookup *operatorsv1alpha1.BundleLookup, timeout time.Duration) (result *BundleUnpackResult, err error) {
+func (c *ConfigMapUnpacker) UnpackBundle(lookup *operatorsv1alpha1.BundleLookup, timeout, retryInterval time.Duration) (result *BundleUnpackResult, err error) {
 	result = newBundleUnpackResult(lookup)
 
 	// if bundle lookup failed condition already present, then there is nothing more to do
@@ -510,7 +518,7 @@ func (c *ConfigMapUnpacker) UnpackBundle(lookup *operatorsv1alpha1.BundleLookup,
 		secrets = append(secrets, corev1.LocalObjectReference{Name: secretName})
 	}
 	var job *batchv1.Job
-	job, err = c.ensureJob(cmRef, result.Path, secrets, timeout)
+	job, err = c.ensureJob(cmRef, result.Path, secrets, timeout, retryInterval)
 	if err != nil || job == nil {
 		// ensureJob can return nil if the job present does not match the expected job (spec and ownerefs)
 		// The current job is deleted in that case so UnpackBundle needs to be retried
@@ -649,7 +657,7 @@ func (c *ConfigMapUnpacker) ensureConfigmap(csRef *corev1.ObjectReference, name
 	return
 }
 
-func (c *ConfigMapUnpacker) ensureJob(cmRef *corev1.ObjectReference, bundlePath string, secrets []corev1.LocalObjectReference, timeout time.Duration) (job *batchv1.Job, err error) {
+func (c *ConfigMapUnpacker) ensureJob(cmRef *corev1.ObjectReference, bundlePath string, secrets []corev1.LocalObjectReference, timeout time.Duration, unpackRetryInterval time.Duration) (job *batchv1.Job, err error) {
 	fresh := c.job(cmRef, bundlePath, secrets, timeout)
 	job, err = c.jobLister.Jobs(fresh.GetNamespace()).Get(fresh.GetName())
 	if err != nil {
@@ -659,13 +667,40 @@ func (c *ConfigMapUnpacker) ensureJob(cmRef *corev1.ObjectReference, bundlePath
 
 		return
 	}
-	// Cleanup old unpacking job and retry
-	if _, isFailed := getCondition(job, batchv1.JobFailed); isFailed {
-		err = c.client.BatchV1().Jobs(job.GetNamespace()).Delete(context.TODO(), job.GetName(), metav1.DeleteOptions{})
-		if err == nil {
-			job, err = c.client.BatchV1().Jobs(fresh.GetNamespace()).Create(context.TODO(), fresh, metav1.CreateOptions{})
+
+	// only check for retries if an unpackRetryInterval is specified
+	if unpackRetryInterval > 0 {
+		if failedCond, isFailed := getCondition(job, batchv1.JobFailed); isFailed {
+			lastFailureTime := failedCond.LastTransitionTime.Time
+			// Look for other unpack jobs for the same bundle
+			var jobs []*batchv1.Job
+			jobs, err = c.jobLister.Jobs(fresh.GetNamespace()).List(k8slabels.ValidatedSetSelector{bundleUnpackRefLabel: cmRef.Name})
+			if err != nil {
+				return
+			}
+
+			var failed bool
+			var cond *batchv1.JobCondition
+			for _, j := range jobs {
+				cond, failed = getCondition(j, batchv1.JobFailed)
+				if !failed {
+					// found an in-progress unpack attempt
+					job = j
+					break
+				}
+				if cond != nil && lastFailureTime.Before(cond.LastTransitionTime.Time) {
+					lastFailureTime = cond.LastTransitionTime.Time
+				}
+			}
+
+			if failed {
+				if time.Now().After(lastFailureTime.Add(unpackRetryInterval)) {
+					fresh.SetName(fmt.Sprintf("%s-%d", fresh.GetName(), len(jobs)))
+					job, err = c.client.BatchV1().Jobs(fresh.GetNamespace()).Create(context.TODO(), fresh, metav1.CreateOptions{})
+					return
+				}
+			}
 		}
-		return
 	}
 
 	if equality.Semantic.DeepDerivative(fresh.GetOwnerReferences(), job.GetOwnerReferences()) && equality.Semantic.DeepDerivative(fresh.Spec, job.Spec) {
@@ -835,3 +870,28 @@ func OperatorGroupBundleUnpackTimeout(ogLister v1listers.OperatorGroupNamespaceL
 
 	return d, nil
 }
+
+// OperatorGroupBundleUnpackRetryInterval returns bundle unpack retry interval from annotation if specified.
+// If the retry annotation is not set, return retry = 0 which is subsequently ignored. This interval, if > 0,
+// determines the minimum interval between recreating a failed unpack job.
+func OperatorGroupBundleUnpackRetryInterval(ogLister v1listers.OperatorGroupNamespaceLister) (time.Duration, error) {
+	ogs, err := ogLister.List(k8slabels.Everything())
+	if err != nil {
+		return 0, err
+	}
+	if len(ogs) != 1 {
+		return 0, fmt.Errorf("found %d operatorGroups, expected 1", len(ogs))
+	}
+
+	timeoutStr, ok := ogs[0].GetAnnotations()[BundleUnpackRetryMinimumIntervalAnnotationKey]
+	if !ok {
+		return 0, nil
+	}
+
+	d, err := time.ParseDuration(timeoutStr)
+	if err != nil {
+		return 0, fmt.Errorf("failed to parse unpack timeout annotation(%s: %s): %w", BundleUnpackRetryMinimumIntervalAnnotationKey, timeoutStr, err)
+	}
+
+	return d, nil
+}

pkg/controller/bundle/bundle_unpacker_test.go

@@ -1673,7 +1673,7 @@ func TestConfigMapUnpacker(t *testing.T) {
 			)
 			require.NoError(t, err)
 
-			res, err := unpacker.UnpackBundle(tt.args.lookup, tt.args.annotationTimeout)
+			res, err := unpacker.UnpackBundle(tt.args.lookup, tt.args.annotationTimeout, 0)
 			require.Equal(t, tt.expected.err, err)
 
 			if tt.expected.res == nil {

pkg/controller/bundle/bundlefakes/fake_unpacker.go

@@ -1112,6 +1112,11 @@ func (o *Operator) syncResolvingNamespace(obj interface{}) error {
 		return err
 	}
 
+	minUnpackRetryInterval, err := bundle.OperatorGroupBundleUnpackRetryInterval(ogLister)
+	if err != nil {
+		return err
+	}
+
 	// TODO: parallel
 	maxGeneration := 0
 	subscriptionUpdated := false
@@ -1207,7 +1212,7 @@ func (o *Operator) syncResolvingNamespace(obj interface{}) error {
 		logger.Debug("unpacking bundles")
 
 		var unpacked bool
-		unpacked, steps, bundleLookups, err = o.unpackBundles(namespace, steps, bundleLookups, unpackTimeout)
+		unpacked, steps, bundleLookups, err = o.unpackBundles(namespace, steps, bundleLookups, unpackTimeout, minUnpackRetryInterval)
 		if err != nil {
 			// If the error was fatal capture and fail
 			if olmerrors.IsFatal(err) {
@@ -1664,7 +1669,7 @@ type UnpackedBundleReference struct {
 	Properties             string `json:"properties"`
 }
 
-func (o *Operator) unpackBundles(namespace string, installPlanSteps []*v1alpha1.Step, bundleLookups []v1alpha1.BundleLookup, unpackTimeout time.Duration) (bool, []*v1alpha1.Step, []v1alpha1.BundleLookup, error) {
+func (o *Operator) unpackBundles(namespace string, installPlanSteps []*v1alpha1.Step, bundleLookups []v1alpha1.BundleLookup, unpackTimeout, minUnpackRetryInterval time.Duration) (bool, []*v1alpha1.Step, []v1alpha1.BundleLookup, error) {
 	unpacked := true
 
 	outBundleLookups := make([]v1alpha1.BundleLookup, len(bundleLookups))
@@ -1679,7 +1684,7 @@ func (o *Operator) unpackBundles(namespace string, installPlanSteps []*v1alpha1.
 	var errs []error
 	for i := 0; i < len(outBundleLookups); i++ {
 		lookup := outBundleLookups[i]
-		res, err := o.bundleUnpacker.UnpackBundle(&lookup, unpackTimeout)
+		res, err := o.bundleUnpacker.UnpackBundle(&lookup, unpackTimeout, minUnpackRetryInterval)
 		if err != nil {
 			errs = append(errs, err)
 			continue