Introduce UpgradeStrategy field to OperatorGroups (#235)

This commit introduces the upgradeStrategy field to the operatorGroup
type. The upgradeStrategy field defines OLM behavior when upgrading
operators in the namespace. Two upgrade strategies are currently
supported, "Default" and "TechPreviewUnsafeFailForward".

The "Default" strategy mimics existing behavior.

The "TechPreviewUnsafeFailForward" strategy allows users to recover
from failed installPlans or failed clusterServiceVersions, it should
be noted that these upgrades are unsafe and can lead to unpredictable
behavior and potentially dataloss.

Upstream-repository: api
Upstream-commit: 33310d6154f344da5494c788451bd34cdf54711c

manifests/0000_50_olm_00-operatorgroups.crd.yaml

@@ -39,6 +39,8 @@ spec:
             spec:
               description: OperatorGroupSpec is the spec for an OperatorGroup resource.
               type: object
+              default:
+                upgradeStrategy: Default
               properties:
                 selector:
                   description: Selector selects the OperatorGroup's target namespaces.
@@ -82,6 +84,13 @@ spec:
                   items:
                     type: string
                   x-kubernetes-list-type: set
+                upgradeStrategy:
+                  description: "UpgradeStrategy defines the upgrade strategy for operators in the namespace. There are currently two supported upgrade strategies: \n Default: OLM will only allow clusterServiceVersions to move to the replacing phase from the succeeded phase. This effectively means that OLM will not allow operators to move to the next version if an installation or upgrade has failed. \n TechPreviewUnsafeFailForward: OLM will allow clusterServiceVersions to move to the replacing phase from the succeeded phase or from the failed phase. Additionally, OLM will generate new installPlans when a subscription references a failed installPlan and the catalog has been updated with a new upgrade for the existing set of operators. \n WARNING: The TechPreviewUnsafeFailForward upgrade strategy is unsafe and may result in unexpected behavior or unrecoverable data loss unless you have deep understanding of the set of operators being managed in the namespace."
+                  type: string
+                  default: Default
+                  enum:
+                    - Default
+                    - TechPreviewUnsafeFailForward
             status:
               description: OperatorGroupStatus is the status for an OperatorGroupResource.
               type: object

staging/api/crds/operators.coreos.com_operatorgroups.yaml

@@ -37,6 +37,8 @@ spec:
             spec:
               description: OperatorGroupSpec is the spec for an OperatorGroup resource.
               type: object
+              default:
+                upgradeStrategy: Default
               properties:
                 selector:
                   description: Selector selects the OperatorGroup's target namespaces.
@@ -80,6 +82,13 @@ spec:
                   items:
                     type: string
                   x-kubernetes-list-type: set
+                upgradeStrategy:
+                  description: "UpgradeStrategy defines the upgrade strategy for operators in the namespace. There are currently two supported upgrade strategies: \n Default: OLM will only allow clusterServiceVersions to move to the replacing phase from the succeeded phase. This effectively means that OLM will not allow operators to move to the next version if an installation or upgrade has failed. \n TechPreviewUnsafeFailForward: OLM will allow clusterServiceVersions to move to the replacing phase from the succeeded phase or from the failed phase. Additionally, OLM will generate new installPlans when a subscription references a failed installPlan and the catalog has been updated with a new upgrade for the existing set of operators. \n WARNING: The TechPreviewUnsafeFailForward upgrade strategy is unsafe and may result in unexpected behavior or unrecoverable data loss unless you have deep understanding of the set of operators being managed in the namespace."
+                  type: string
+                  default: Default
+                  enum:
+                    - Default
+                    - TechPreviewUnsafeFailForward
             status:
               description: OperatorGroupStatus is the status for an OperatorGroupResource.
               type: object

staging/api/pkg/operators/v1/operatorgroup_test.go

@@ -0,0 +1,70 @@
+package v1
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func TestUpgradeStrategy(t *testing.T) {
+	tests := []struct {
+		description string
+		og          *OperatorGroup
+		expected    UpgradeStrategy
+	}{
+		{
+			description: "NoSpec",
+			og:          &OperatorGroup{},
+			expected:    UpgradeStrategyDefault,
+		},
+		{
+			description: "NoUpgradeStrategy",
+			og: &OperatorGroup{
+				Spec: OperatorGroupSpec{},
+			},
+			expected: UpgradeStrategyDefault,
+		},
+		{
+			description: "NoUpgradeStrategy",
+			og: &OperatorGroup{
+				Spec: OperatorGroupSpec{
+					UpgradeStrategy: "",
+				},
+			},
+			expected: UpgradeStrategyDefault,
+		},
+		{
+			description: "NonSupportedUpgradeStrategy",
+			og: &OperatorGroup{
+				Spec: OperatorGroupSpec{
+					UpgradeStrategy: "foo",
+				},
+			},
+			expected: UpgradeStrategyDefault,
+		},
+		{
+			description: "DefaultUpgradeStrategy",
+			og: &OperatorGroup{
+				Spec: OperatorGroupSpec{
+					UpgradeStrategy: "Default",
+				},
+			},
+			expected: UpgradeStrategyDefault,
+		},
+		{
+			description: "UnsafeFailForwardUpgradeStrategy",
+			og: &OperatorGroup{
+				Spec: OperatorGroupSpec{
+					UpgradeStrategy: "TechPreviewUnsafeFailForward",
+				},
+			},
+			expected: UpgradeStrategyUnsafeFailForward,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.description, func(t *testing.T) {
+			require.EqualValues(t, tt.expected, tt.og.UpgradeStrategy())
+		})
+	}
+}

staging/api/pkg/operators/v1/operatorgroup_types.go

@@ -24,8 +24,29 @@ const (
 	MutlipleOperatorGroupCondition       = "MultipleOperatorGroup"
 	MultipleOperatorGroupsReason         = "MultipleOperatorGroupsFound"
 	OperatorGroupServiceAccountReason    = "ServiceAccountNotFound"
+
+	// UpgradeStrategyDefault configures OLM such that it will only allow
+	// clusterServiceVersions to move to the replacing phase to the succeeded
+	// phase. This effectively means that OLM will not allow operators to move
+	// to the next version if an installation or upgrade has failed.
+	UpgradeStrategyDefault UpgradeStrategy = "Default"
+
+	// UpgradeStrategyUnsafeFailForward configures OLM such that it  will allow
+	// clusterServiceVersions to move to the  replacing phase from the succeeded
+	// phase or from the failed phase. Additionally, OLM will generate new
+	// installPlans when a subscription references a failed installPlan and the
+	// catalog has been updated with a new upgrade for the existing set of
+	// operators.
+	//
+	// WARNING: The UpgradeStrategyUnsafeFailForward upgrade strategy is unsafe
+	// and may result in unexpected behavior or unrecoverable data loss unless
+	// you have deep understanding of the set of operators being managed in the
+	// namespace.
+	UpgradeStrategyUnsafeFailForward UpgradeStrategy = "TechPreviewUnsafeFailForward"
 )
 
+type UpgradeStrategy string
+
 // OperatorGroupSpec is the spec for an OperatorGroup resource.
 type OperatorGroupSpec struct {
 	// Selector selects the OperatorGroup's target namespaces.
@@ -45,6 +66,29 @@ type OperatorGroupSpec struct {
 	// Static tells OLM not to update the OperatorGroup's providedAPIs annotation
 	// +optional
 	StaticProvidedAPIs bool `json:"staticProvidedAPIs,omitempty"`
+
+	// UpgradeStrategy defines the upgrade strategy for operators in the namespace.
+	// There are currently two supported upgrade strategies:
+	//
+	// Default: OLM will only allow clusterServiceVersions to move to the replacing
+	// phase from the succeeded phase. This effectively means that OLM will not
+	// allow operators to move to the next version if an installation or upgrade
+	// has failed.
+	//
+	// TechPreviewUnsafeFailForward: OLM will allow clusterServiceVersions to move to the
+	// replacing phase from the succeeded phase or from the failed phase.
+	// Additionally, OLM will generate new installPlans when a subscription references
+	// a failed installPlan and the catalog has been updated with a new upgrade for
+	// the existing set of operators.
+	//
+	// WARNING: The TechPreviewUnsafeFailForward upgrade strategy is unsafe and may result
+	// in unexpected behavior or unrecoverable data loss unless you have deep
+	// understanding of the set of operators being managed in the namespace.
+	//
+	// +kubebuilder:validation:Enum=Default;TechPreviewUnsafeFailForward
+	// +kubebuilder:default=Default
+	// +optional
+	UpgradeStrategy UpgradeStrategy `json:"upgradeStrategy,omitempty"`
 }
 
 // OperatorGroupStatus is the status for an OperatorGroupResource.
@@ -76,6 +120,7 @@ type OperatorGroup struct {
 	metav1.ObjectMeta `json:"metadata"`
 
 	// +optional
+	// +kubebuilder:default={upgradeStrategy:Default}
 	Spec   OperatorGroupSpec   `json:"spec"`
 	Status OperatorGroupStatus `json:"status,omitempty"`
 }
@@ -98,6 +143,17 @@ func (o *OperatorGroup) BuildTargetNamespaces() string {
 	return strings.Join(ns, ",")
 }
 
+// UpgradeStrategy returns the UpgradeStrategy specified or the default value otherwise.
+func (o *OperatorGroup) UpgradeStrategy() UpgradeStrategy {
+	strategyName := o.Spec.UpgradeStrategy
+	switch {
+	case strategyName == UpgradeStrategyUnsafeFailForward:
+		return strategyName
+	default:
+		return UpgradeStrategyDefault
+	}
+}
+
 // IsServiceAccountSpecified returns true if the spec has a service account name specified.
 func (o *OperatorGroup) IsServiceAccountSpecified() bool {
 	if o.Spec.ServiceAccountName == "" {