Feat: support customizing application revision limit number in policy (#5995)

Signed-off-by: Somefive <yd219913@alibaba-inc.com>

apis/core.oam.dev/v1alpha1/garbagecollect_policy_types.go

@@ -27,6 +27,10 @@ const (
 
 // GarbageCollectPolicySpec defines the spec of configuration drift
 type GarbageCollectPolicySpec struct {
+	// ApplicationRevisionLimit if set, this application will use this number for application revision instead of
+	// the global configuration
+	ApplicationRevisionLimit *int `json:"applicationRevisionLimit,omitempty"`
+
 	// KeepLegacyResource if is set, outdated versioned resourcetracker will not be recycled automatically
 	// outdated resources will be kept until resourcetracker be deleted manually
 	KeepLegacyResource bool `json:"keepLegacyResource,omitempty"`

apis/core.oam.dev/v1alpha1/zz_generated.deepcopy.go

@@ -329,6 +329,11 @@ func (in *GarbageCollectPolicyRule) DeepCopy() *GarbageCollectPolicyRule {
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in *GarbageCollectPolicySpec) DeepCopyInto(out *GarbageCollectPolicySpec) {
 	*out = *in
+	if in.ApplicationRevisionLimit != nil {
+		in, out := &in.ApplicationRevisionLimit, &out.ApplicationRevisionLimit
+		*out = new(int)
+		**out = **in
+	}
 	if in.Rules != nil {
 		in, out := &in.Rules, &out.Rules
 		*out = make([]GarbageCollectPolicyRule, len(*in))

charts/vela-core/templates/defwithtemplate/garbage-collect.yaml

@@ -34,6 +34,8 @@ spec:
         }
 
         parameter: {
+        	// +usage=If set, it will override the default revision limit number and customize this number for the current application
+        	applicationRevisionLimit?: int
         	// +usage=If is set, outdated versioned resourcetracker will not be recycled automatically, outdated resources will be kept until resourcetracker be deleted manually
         	keepLegacyResource: *false | bool
         	// +usage=If is set, continue to execute gc when the workflow fails, by default gc will be executed only after the workflow succeeds

pkg/controller/core.oam.dev/v1alpha2/application/revision.go

@@ -901,6 +901,18 @@ func (h *AppHandler) UpdateAppLatestRevisionStatus(ctx context.Context) error {
 	return nil
 }
 
+func getApplicationRevisionLimitForApp(app *v1beta1.Application, fallback int) int {
+	for _, p := range app.Spec.Policies {
+		if p.Type == v1alpha1.GarbageCollectPolicyType && p.Properties != nil && p.Properties.Raw != nil {
+			prop := &v1alpha1.GarbageCollectPolicySpec{}
+			if err := json.Unmarshal(p.Properties.Raw, prop); err == nil && prop.ApplicationRevisionLimit != nil && *prop.ApplicationRevisionLimit >= 0 {
+				return *prop.ApplicationRevisionLimit
+			}
+		}
+	}
+	return fallback
+}
+
 // cleanUpApplicationRevision check all appRevisions of the application, remove them if the number of them exceed the limit
 func cleanUpApplicationRevision(ctx context.Context, h *AppHandler) error {
 	if DisableAllApplicationRevision {
@@ -915,11 +927,12 @@ func cleanUpApplicationRevision(ctx context.Context, h *AppHandler) error {
 		return err
 	}
 	appRevisionInUse := gatherUsingAppRevision(h)
-	needKill := len(sortedRevision) - h.r.appRevisionLimit - len(appRevisionInUse)
+	appRevisionLimit := getApplicationRevisionLimitForApp(h.app, h.r.appRevisionLimit)
+	needKill := len(sortedRevision) - appRevisionLimit - len(appRevisionInUse)
 	if needKill <= 0 {
 		return nil
 	}
-	klog.InfoS("Going to garbage collect app revisions", "limit", h.r.appRevisionLimit,
+	klog.InfoS("Going to garbage collect app revisions", "limit", appRevisionLimit,
 		"total", len(sortedRevision), "using", len(appRevisionInUse), "kill", needKill)
 
 	for _, rev := range sortedRevision {

test/e2e-multicluster-test/multicluster_test.go

@@ -1111,5 +1111,60 @@ var _ = Describe("Test multicluster scenario", func() {
 			Expect(k8sClient.Get(workerCtx, appKey, _deploy)).Should(Succeed())
 			Expect(int(*_deploy.Spec.Replicas)).Should(Equal(0))
 		})
+
+		It("Test application with customized application revision limit", func() {
+			ctx := context.Background()
+			app := &v1beta1.Application{}
+			bs, err := os.ReadFile("./testdata/app/app-lite.yaml")
+			Expect(err).Should(Succeed())
+			Expect(yaml.Unmarshal(bs, app)).Should(Succeed())
+			app.SetNamespace(namespace)
+			Eventually(func(g Gomega) {
+				g.Expect(k8sClient.Create(ctx, app)).Should(Succeed())
+			}).WithPolling(2 * time.Second).WithTimeout(5 * time.Second).Should(Succeed())
+			appKey := client.ObjectKeyFromObject(app)
+			Eventually(func(g Gomega) {
+				_app := &v1beta1.Application{}
+				g.Expect(k8sClient.Get(ctx, appKey, _app)).Should(Succeed())
+				g.Expect(_app.Status.Phase).Should(Equal(common.ApplicationRunning))
+			}).WithPolling(2 * time.Second).WithTimeout(20 * time.Second).Should(Succeed())
+
+			By("update app and should have two revisions")
+			Eventually(func(g Gomega) {
+				_app := &v1beta1.Application{}
+				g.Expect(k8sClient.Get(ctx, appKey, _app)).Should(Succeed())
+				_app.Spec.Components[0].Name = "dw"
+				g.Expect(k8sClient.Update(ctx, _app)).Should(Succeed())
+			}).WithPolling(2 * time.Second).WithTimeout(20 * time.Second).Should(Succeed())
+			Eventually(func(g Gomega) {
+				_app := &v1beta1.Application{}
+				g.Expect(k8sClient.Get(ctx, appKey, _app)).Should(Succeed())
+				g.Expect(_app.Status.Phase).Should(Equal(common.ApplicationRunning))
+				_revs := &v1beta1.ApplicationRevisionList{}
+				g.Expect(k8sClient.List(ctx, _revs, client.InNamespace(namespace))).Should(Succeed())
+				g.Expect(len(_revs.Items)).Should(Equal(2))
+			}).WithPolling(2 * time.Second).WithTimeout(20 * time.Second).Should(Succeed())
+
+			By("update app with gc policy and should have one revision")
+			Eventually(func(g Gomega) {
+				_app := &v1beta1.Application{}
+				g.Expect(k8sClient.Get(ctx, appKey, _app)).Should(Succeed())
+				_app.Spec.Components[0].Name = "dw2"
+				_app.Spec.Policies = []v1beta1.AppPolicy{{
+					Type:       "garbage-collect",
+					Name:       "gc",
+					Properties: &runtime.RawExtension{Raw: []byte(`{"applicationRevisionLimit":0}`)},
+				}}
+				g.Expect(k8sClient.Update(ctx, _app)).Should(Succeed())
+			}).WithPolling(2 * time.Second).WithTimeout(20 * time.Second).Should(Succeed())
+			Eventually(func(g Gomega) {
+				_app := &v1beta1.Application{}
+				g.Expect(k8sClient.Get(ctx, appKey, _app)).Should(Succeed())
+				g.Expect(_app.Status.Phase).Should(Equal(common.ApplicationRunning))
+				_revs := &v1beta1.ApplicationRevisionList{}
+				g.Expect(k8sClient.List(ctx, _revs, client.InNamespace(namespace))).Should(Succeed())
+				g.Expect(len(_revs.Items)).Should(Equal(1))
+			}).WithPolling(2 * time.Second).WithTimeout(20 * time.Second).Should(Succeed())
+		})
 	})
 })

test/e2e-multicluster-test/testdata/app/app-lite.yaml

@@ -0,0 +1,13 @@
+apiVersion: core.oam.dev/v1beta1
+kind: Application
+metadata:
+  name: example-lite-app
+spec:
+  components:
+    - name: data-worker
+      type: worker
+      properties:
+        image: busybox
+        cmd:
+          - sleep
+          - '1000000'

vela-templates/definitions/internal/policy/garbage-collect.cue

@@ -30,6 +30,8 @@ template: {
 	}
 
 	parameter: {
+		// +usage=If set, it will override the default revision limit number and customize this number for the current application
+		applicationRevisionLimit?: int
 		// +usage=If is set, outdated versioned resourcetracker will not be recycled automatically, outdated resources will be kept until resourcetracker be deleted manually
 		keepLegacyResource: *false | bool
 		// +usage=If is set, continue to execute gc when the workflow fails, by default gc will be executed only after the workflow succeeds