fix: client can detect if rollout is in the process of unpausing (argoproj#1798)

Signed-off-by: Jesse Suen <[email protected]>

Author: jessesuen

pkg/apiclient/rollout/rollout.swagger.json

@@ -1300,7 +1300,7 @@
           "items": {
             "$ref": "#/definitions/github.com.argoproj.argo_rollouts.pkg.apis.rollouts.v1alpha1.PauseCondition"
           },
-          "title": "PauseConditions indicates why the rollout is currently paused"
+          "title": "PauseConditions is a list of reasons why rollout became automatically paused (e.g.\nCanaryPauseStep, BlueGreenPause, InconclusiveAnalysis). The items in this list are populated\nby the controller but are cleared by the user (e.g. plugin, argo-cd resume action) when they\nwish to unpause. If pause conditions is empty, but controllerPause is true, it indicates\nthe user manually unpaused the Rollout"
         },
         "controllerPause": {
           "type": "boolean",

pkg/apis/rollouts/v1alpha1/generated.proto

@@ -684,7 +684,11 @@ const (
 type RolloutStatus struct {
 	// Abort cancel the current rollout progression
 	Abort bool `json:"abort,omitempty" protobuf:"varint,1,opt,name=abort"`
-	// PauseConditions indicates why the rollout is currently paused
+	// PauseConditions is a list of reasons why rollout became automatically paused (e.g.
+	// CanaryPauseStep, BlueGreenPause, InconclusiveAnalysis). The items in this list are populated
+	// by the controller but are cleared by the user (e.g. plugin, argo-cd resume action) when they
+	// wish to unpause. If pause conditions is empty, but controllerPause is true, it indicates
+	// the user manually unpaused the Rollout
 	PauseConditions []PauseCondition `json:"pauseConditions,omitempty" protobuf:"bytes,2,rep,name=pauseConditions"`
 	// ControllerPause indicates the controller has paused the rollout. It is set to true when
 	// the controller adds a pause condition. This field helps to discern the scenario where a

pkg/apis/rollouts/v1alpha1/openapi_generated.go

@@ -1,3 +1,4 @@
+//go:build e2e
 // +build e2e
 
 package e2e
@@ -292,10 +293,6 @@ func (s *IstioSuite) TestIstioAbortUpdateDeleteAllCanaryPods() {
 		PromoteRollout().
 		WaitForRolloutStatus("Paused").
 		Then().
-		When().
-		PromoteRollout().
-		WaitForRolloutStatus("Paused").
-		Then().
 		ExpectRevisionPodCount("2", 4).
 		When().
 		AbortRollout().

pkg/apis/rollouts/v1alpha1/types.go

@@ -436,7 +436,7 @@ func GetCurrentCanaryStep(rollout *v1alpha1.Rollout) (*v1alpha1.CanaryStep, *int
 
 // GetCanaryReplicasOrWeight either returns a static set of replicas or a weight percentage
 func GetCanaryReplicasOrWeight(rollout *v1alpha1.Rollout) (*int32, int32) {
-	if rollout.Status.PromoteFull || rollout.Status.CurrentPodHash == rollout.Status.StableRS {
+	if rollout.Status.PromoteFull || rollout.Status.StableRS == "" || rollout.Status.CurrentPodHash == rollout.Status.StableRS {
 		return nil, 100
 	}
 	if scs := UseSetCanaryScale(rollout); scs != nil {

test/e2e/istio_test.go

@@ -23,7 +23,9 @@ func GetRolloutPhase(ro *v1alpha1.Rollout) (v1alpha1.RolloutPhase, string) {
 	if !isGenerationObserved(ro) {
 		return v1alpha1.RolloutPhaseProgressing, "waiting for rollout spec update to be observed"
 	}
-
+	if IsUnpausing(ro) {
+		return v1alpha1.RolloutPhaseProgressing, "waiting for rollout to unpause"
+	}
 	if ro.Spec.TemplateResolvedFromRef && !isWorkloadGenerationObserved(ro) {
 		return v1alpha1.RolloutPhaseProgressing, "waiting for rollout spec update to be observed for the reference workload"
 	}
@@ -51,6 +53,17 @@ func isGenerationObserved(ro *v1alpha1.Rollout) bool {
 	return int64(observedGen) == ro.Generation
 }
 
+// IsUnpausing detects if we are in the process of unpausing a rollout. This is determined by seeing
+// if status.controllerPause is true, but the list of pause conditions (status.pauseConditions)
+// is empty. This implies that a user cleared the pause conditions but controller has not yet
+// observed or reacted to it.
+// NOTE: this function is necessary because unlike metadata.generation & status.observedGeneration
+// status.controllerPause & status.pauseConditions are both status fields and does not benefit from
+// the auto-incrementing behavior of metadata.generation.
+func IsUnpausing(ro *v1alpha1.Rollout) bool {
+	return ro.Status.ControllerPause && len(ro.Status.PauseConditions) == 0
+}
+
 func isWorkloadGenerationObserved(ro *v1alpha1.Rollout) bool {
 	if _, ok := annotations.GetWorkloadGenerationAnnotation(ro); !ok {
 		return true

utils/replicaset/canary.go

@@ -394,3 +394,23 @@ func TestCanaryStepString(t *testing.T) {
 		assert.Equal(t, test.expectedString, CanaryStepString(test.step))
 	}
 }
+
+func TestIsUnpausing(t *testing.T) {
+	ro := newCanaryRollout()
+	ro.Status.Phase = v1alpha1.RolloutPhasePaused
+	ro.Status.Message = "canary pause"
+	ro.Status.PauseConditions = []v1alpha1.PauseCondition{
+		{
+			Reason: v1alpha1.PauseReasonCanaryPauseStep,
+		},
+	}
+	ro.Status.ControllerPause = true
+	status, message := GetRolloutPhase(ro)
+	assert.Equal(t, v1alpha1.RolloutPhasePaused, status)
+	assert.Equal(t, "canary pause", message)
+
+	ro.Status.PauseConditions = nil
+	status, message = GetRolloutPhase(ro)
+	assert.Equal(t, v1alpha1.RolloutPhaseProgressing, status)
+	assert.Equal(t, "waiting for rollout to unpause", message)
+}