openkruise · FillZpp · Sep 27, 2022 · Sep 16, 2022
diff --git a/.github/workflows/ci.yaml b/.github/workflows/ci.yaml
@@ -48,6 +48,7 @@ jobs:
           version: ${{ env.GOLANGCI_VERSION }}
           args: --verbose
           skip-pkg-cache: true
+          mod: readonly
 
   markdownlint-misspell-shellcheck:
     runs-on: ubuntu-18.04

diff --git a/.github/workflows/e2e-1.16.yaml b/.github/workflows/e2e-1.16.yaml
@@ -10,7 +10,7 @@ on:
 
 env:
   # Common versions
-  GO_VERSION: '1.17'
+  GO_VERSION: '1.18'
   KIND_VERSION: 'v0.14.0'
   KIND_IMAGE: 'kindest/node:v1.16.15'
   KIND_CLUSTER_NAME: 'ci-testing'
@@ -85,6 +85,18 @@ jobs:
               kubectl get pod -n kruise-system --no-headers -l control-plane=controller-manager | awk '{print $1}' | xargs kubectl logs -p -n kruise-system
               exit 1
           fi
+          kubectl get pods -n kruise-system -l control-plane=daemon -o=jsonpath="{range .items[*]}{.metadata.namespace}{\"\t\"}{.metadata.name}{\"\n\"}{end}" | while read ns name;
+          do
+            restartCount=$(kubectl get pod -n ${ns} ${name} --no-headers | awk '{print $4}')
+            if [ "${restartCount}" -eq "0" ];then
+                echo "Kruise-daemon has not restarted"
+            else
+                kubectl get pods -n ${ns} -l control-plane=daemon --no-headers
+                echo "Kruise-daemon has restarted, abort!!!"
+                kubectl logs -p -n ${ns} ${name}
+                exit 1
+            fi
+          done
           exit $retVal
 
   pullimages-containerrecreate:
@@ -154,6 +166,18 @@ jobs:
               kubectl get pod -n kruise-system --no-headers -l control-plane=controller-manager | awk '{print $1}' | xargs kubectl logs -p -n kruise-system
               exit 1
           fi
+          kubectl get pods -n kruise-system -l control-plane=daemon -o=jsonpath="{range .items[*]}{.metadata.namespace}{\"\t\"}{.metadata.name}{\"\n\"}{end}" | while read ns name;
+          do
+            restartCount=$(kubectl get pod -n ${ns} ${name} --no-headers | awk '{print $4}')
+            if [ "${restartCount}" -eq "0" ];then
+                echo "Kruise-daemon has not restarted"
+            else
+                kubectl get pods -n ${ns} -l control-plane=daemon --no-headers
+                echo "Kruise-daemon has restarted, abort!!!"
+                kubectl logs -p -n ${ns} ${name}
+                exit 1
+            fi
+          done
           exit $retVal
 
   advanced-daemonset:
@@ -223,6 +247,18 @@ jobs:
               kubectl get pod -n kruise-system --no-headers -l control-plane=controller-manager | awk '{print $1}' | xargs kubectl logs -p -n kruise-system
               exit 1
           fi
+          kubectl get pods -n kruise-system -l control-plane=daemon -o=jsonpath="{range .items[*]}{.metadata.namespace}{\"\t\"}{.metadata.name}{\"\n\"}{end}" | while read ns name;
+          do
+            restartCount=$(kubectl get pod -n ${ns} ${name} --no-headers | awk '{print $4}')
+            if [ "${restartCount}" -eq "0" ];then
+                echo "Kruise-daemon has not restarted"
+            else
+                kubectl get pods -n ${ns} -l control-plane=daemon --no-headers
+                echo "Kruise-daemon has restarted, abort!!!"
+                kubectl logs -p -n ${ns} ${name}
+                exit 1
+            fi
+          done
           exit $retVal
 
   sidecarset:
@@ -292,6 +328,18 @@ jobs:
               kubectl get pod -n kruise-system --no-headers -l control-plane=controller-manager | awk '{print $1}' | xargs kubectl logs -p -n kruise-system
               exit 1
           fi
+          kubectl get pods -n kruise-system -l control-plane=daemon -o=jsonpath="{range .items[*]}{.metadata.namespace}{\"\t\"}{.metadata.name}{\"\n\"}{end}" | while read ns name;
+          do
+            restartCount=$(kubectl get pod -n ${ns} ${name} --no-headers | awk '{print $4}')
+            if [ "${restartCount}" -eq "0" ];then
+                echo "Kruise-daemon has not restarted"
+            else
+                kubectl get pods -n ${ns} -l control-plane=daemon --no-headers
+                echo "Kruise-daemon has restarted, abort!!!"
+                kubectl logs -p -n ${ns} ${name}
+                exit 1
+            fi
+          done
           exit $retVal
 
   podUnavailableBudget:
@@ -419,4 +467,16 @@ jobs:
               kubectl get pod -n kruise-system --no-headers -l control-plane=controller-manager | awk '{print $1}' | xargs kubectl logs -p -n kruise-system
               exit 1
           fi
+          kubectl get pods -n kruise-system -l control-plane=daemon -o=jsonpath="{range .items[*]}{.metadata.namespace}{\"\t\"}{.metadata.name}{\"\n\"}{end}" | while read ns name;
+          do
+            restartCount=$(kubectl get pod -n ${ns} ${name} --no-headers | awk '{print $4}')
+            if [ "${restartCount}" -eq "0" ];then
+                echo "Kruise-daemon has not restarted"
+            else
+                kubectl get pods -n ${ns} -l control-plane=daemon --no-headers
+                echo "Kruise-daemon has restarted, abort!!!"
+                kubectl logs -p -n ${ns} ${name}
+                exit 1
+            fi
+          done
           exit $retVal
diff --git a/.github/workflows/e2e-1.24.yaml b/.github/workflows/e2e-1.24.yaml
@@ -154,6 +154,18 @@ jobs:
               kubectl get pod -n kruise-system --no-headers -l control-plane=controller-manager | awk '{print $1}' | xargs kubectl logs -p -n kruise-system
               exit 1
           fi
+          kubectl get pods -n kruise-system -l control-plane=daemon -o=jsonpath="{range .items[*]}{.metadata.namespace}{\"\t\"}{.metadata.name}{\"\n\"}{end}" | while read ns name;
+          do
+            restartCount=$(kubectl get pod -n ${ns} ${name} --no-headers | awk '{print $4}')
+            if [ "${restartCount}" -eq "0" ];then
+                echo "Kruise-daemon has not restarted"
+            else
+                kubectl get pods -n ${ns} -l control-plane=daemon --no-headers
+                echo "Kruise-daemon has restarted, abort!!!"
+                kubectl logs -p -n ${ns} ${name}
+                exit 1
+            fi
+          done
           exit $retVal
 
   advanced-daemonset:
@@ -223,6 +235,18 @@ jobs:
               kubectl get pod -n kruise-system --no-headers -l control-plane=controller-manager | awk '{print $1}' | xargs kubectl logs -p -n kruise-system
               exit 1
           fi
+          kubectl get pods -n kruise-system -l control-plane=daemon -o=jsonpath="{range .items[*]}{.metadata.namespace}{\"\t\"}{.metadata.name}{\"\n\"}{end}" | while read ns name;
+          do
+            restartCount=$(kubectl get pod -n ${ns} ${name} --no-headers | awk '{print $4}')
+            if [ "${restartCount}" -eq "0" ];then
+                echo "Kruise-daemon has not restarted"
+            else
+                kubectl get pods -n ${ns} -l control-plane=daemon --no-headers
+                echo "Kruise-daemon has restarted, abort!!!"
+                kubectl logs -p -n ${ns} ${name}
+                exit 1
+            fi
+          done
           exit $retVal
 
   sidecarset:
@@ -292,6 +316,18 @@ jobs:
               kubectl get pod -n kruise-system --no-headers -l control-plane=controller-manager | awk '{print $1}' | xargs kubectl logs -p -n kruise-system
               exit 1
           fi
+          kubectl get pods -n kruise-system -l control-plane=daemon -o=jsonpath="{range .items[*]}{.metadata.namespace}{\"\t\"}{.metadata.name}{\"\n\"}{end}" | while read ns name;
+          do
+            restartCount=$(kubectl get pod -n ${ns} ${name} --no-headers | awk '{print $4}')
+            if [ "${restartCount}" -eq "0" ];then
+                echo "Kruise-daemon has not restarted"
+            else
+                kubectl get pods -n ${ns} -l control-plane=daemon --no-headers
+                echo "Kruise-daemon has restarted, abort!!!"
+                kubectl logs -p -n ${ns} ${name}
+                exit 1
+            fi
+          done
           exit $retVal
 
   ephemeraljob:
@@ -477,4 +513,16 @@ jobs:
               kubectl get pod -n kruise-system --no-headers -l control-plane=controller-manager | awk '{print $1}' | xargs kubectl logs -p -n kruise-system
               exit 1
           fi
+          kubectl get pods -n kruise-system -l control-plane=daemon -o=jsonpath="{range .items[*]}{.metadata.namespace}{\"\t\"}{.metadata.name}{\"\n\"}{end}" | while read ns name;
+          do
+            restartCount=$(kubectl get pod -n ${ns} ${name} --no-headers | awk '{print $4}')
+            if [ "${restartCount}" -eq "0" ];then
+                echo "Kruise-daemon has not restarted"
+            else
+                kubectl get pods -n ${ns} -l control-plane=daemon --no-headers
+                echo "Kruise-daemon has restarted, abort!!!"
+                kubectl logs -p -n ${ns} ${name}
+                exit 1
+            fi
+          done
           exit $retVal
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
@@ -88,7 +88,7 @@ We encourage contributors to follow the [PR template](./.github/PULL_REQUEST_TEM
 As a contributor, if you want to make any contribution to Kruise project, we should reach an agreement on the version of tools used in the development environment.
 Here are some dependents with specific version:
 
-- Golang : v1.17+
+- Golang : v1.18+
 - Kubernetes: v1.16+
 
 ### Developing guide

diff --git a/Dockerfile b/Dockerfile
@@ -1,5 +1,5 @@
 # Build the manager and daemon binaries
-FROM golang:1.17 as builder
+FROM golang:1.18 as builder
 
 WORKDIR /workspace
 # Copy the Go Modules manifests

diff --git a/Dockerfile_multiarch b/Dockerfile_multiarch
@@ -1,5 +1,5 @@
 # Build the manager and daemon binaries
-FROM --platform=$BUILDPLATFORM golang:1.17 as builder
+FROM --platform=$BUILDPLATFORM golang:1.18 as builder
 
 WORKDIR /workspace
 # Copy the Go Modules manifests

diff --git a/Makefile b/Makefile
@@ -22,7 +22,7 @@ all: build
 ##@ Development
 
 go_check:
-	@scripts/check_go_version "1.17.0"
+	@scripts/check_go_version "1.18.0"
 
 generate: controller-gen ## Generate code containing DeepCopy, DeepCopyInto, and DeepCopyObject method implementations.
 	@scripts/generate_client.sh

diff --git a/config/rbac/daemon_role.yaml b/config/rbac/daemon_role.yaml
@@ -72,3 +72,19 @@ rules:
   - get
   - list
   - watch
+- apiGroups:
+  - apps.kruise.io
+  resources:
+  - nodepodprobes
+  verbs:
+  - get
+  - list
+  - watch
+- apiGroups:
+  - apps.kruise.io
+  resources:
+  - nodepodprobes/status
+  verbs:
+  - get
+  - patch
+  - update
diff --git a/go.mod b/go.mod
@@ -1,6 +1,6 @@
 module github.com/openkruise/kruise
 
-go 1.17
+go 1.18
 
 require (
 	github.com/alibaba/pouch v0.0.0-20190328125340-37051654f368

diff --git a/pkg/controller/nodepodprobe/node_pod_probe_controller.go b/pkg/controller/nodepodprobe/node_pod_probe_controller.go
@@ -23,18 +23,19 @@ import (
 	"strings"
 	"time"
 
-	"k8s.io/apimachinery/pkg/util/sets"
-
 	appsv1alpha1 "github.com/openkruise/kruise/apis/apps/v1alpha1"
+	"github.com/openkruise/kruise/pkg/features"
 	"github.com/openkruise/kruise/pkg/util"
 	utilclient "github.com/openkruise/kruise/pkg/util/client"
 	"github.com/openkruise/kruise/pkg/util/controllerfinder"
 	utildiscovery "github.com/openkruise/kruise/pkg/util/discovery"
+	utilfeature "github.com/openkruise/kruise/pkg/util/feature"
 	"github.com/openkruise/kruise/pkg/util/ratelimiter"
 	corev1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/api/errors"
 	"k8s.io/apimachinery/pkg/runtime"
 	"k8s.io/apimachinery/pkg/types"
+	"k8s.io/apimachinery/pkg/util/sets"
 	"k8s.io/client-go/util/retry"
 	"k8s.io/klog/v2"
 	kubecontroller "k8s.io/kubernetes/pkg/controller"
@@ -67,7 +68,7 @@ var (
 // Add creates a new NodePodProbe Controller and adds it to the Manager with default RBAC. The Manager will set fields on the Controller
 // and Start it when the Manager is Started.
 func Add(mgr manager.Manager) error {
-	if !utildiscovery.DiscoverGVK(controllerKind) {
+	if !utildiscovery.DiscoverGVK(controllerKind) || !utilfeature.DefaultFeatureGate.Enabled(features.PodProbeMarkerGate) {
 		return nil
 	}
 	return add(mgr, newReconciler(mgr))
@@ -220,9 +221,10 @@ func (r *ReconcileNodePodProbe) syncPodFromNodePodProbe(npp *appsv1alpha1.NodePo
 
 func (r *ReconcileNodePodProbe) updatePodProbeStatus(pod *corev1.Pod, status appsv1alpha1.PodProbeStatus) error {
 	// map[probe.name]->probeState
-	currentConditions := make(map[string]appsv1alpha1.ProbeState)
-	for _, condition := range pod.Status.Conditions {
-		currentConditions[string(condition.Type)] = appsv1alpha1.ProbeState(condition.Status)
+	currentConditions := make(map[string]*corev1.PodCondition)
+	for i := range pod.Status.Conditions {
+		condition := &pod.Status.Conditions[i]
+		currentConditions[string(condition.Type)] = condition
 	}
 	type metadata struct {
 		Labels      map[string]interface{} `json:"labels,omitempty"`
@@ -239,11 +241,9 @@ func (r *ReconcileNodePodProbe) updatePodProbeStatus(pod *corev1.Pod, status app
 	validConditionTypes := sets.NewString()
 	for i := range status.ProbeStates {
 		probeState := status.ProbeStates[i]
-		// ignore the probe state
-		if probeState.State == "" || probeState.State == currentConditions[probeState.Name] {
+		if probeState.State == "" {
 			continue
 		}
-
 		// fetch podProbeMarker
 		ppmName, probeName := strings.Split(probeState.Name, "#")[0], strings.Split(probeState.Name, "#")[1]
 		ppm := &appsv1alpha1.PodProbeMarker{}
@@ -267,7 +267,6 @@ func (r *ReconcileNodePodProbe) updatePodProbeStatus(pod *corev1.Pod, status app
 				break
 			}
 		}
-
 		if conditionType != "" && validConditionTypes.Has(conditionType) {
 			klog.Warningf("NodePodProbe(%s) pod(%s/%s) condition(%s) is conflict", ppmName, pod.Namespace, pod.Name, conditionType)
 			// patch pod condition
@@ -287,11 +286,9 @@ func (r *ReconcileNodePodProbe) updatePodProbeStatus(pod *corev1.Pod, status app
 				Message:            probeState.Message,
 			})
 		}
-
 		if len(policy) == 0 {
 			continue
 		}
-
 		// matchedPolicy is when policy.state is equal to probeState.State, otherwise oppositePolicy
 		// 1. If policy[0].state = Succeeded, policy[1].state = Failed. probeState.State = Succeeded.
 		// So policy[0] is matchedPolicy, policy[1] is oppositePolicy
@@ -328,15 +325,14 @@ func (r *ReconcileNodePodProbe) updatePodProbeStatus(pod *corev1.Pod, status app
 	if len(probeConditions) == 0 && len(probeMetadata.Labels) == 0 && len(probeMetadata.Annotations) == 0 {
 		return nil
 	}
-
 	//update pod metadata and status condition
 	podClone := pod.DeepCopy()
 	if err = retry.RetryOnConflict(retry.DefaultBackoff, func() error {
 		if err = r.Client.Get(context.TODO(), types.NamespacedName{Namespace: pod.Namespace, Name: pod.Name}, podClone); err != nil {
 			klog.Errorf("error getting updated pod(%s/%s) from client", pod.Namespace, pod.Name)
 			return err
 		}
-		oldStatus := podClone.DeepCopy()
+		oldStatus := podClone.Status.DeepCopy()
 		for i := range probeConditions {
 			condition := probeConditions[i]
 			util.SetPodCondition(podClone, condition)
@@ -363,7 +359,7 @@ func (r *ReconcileNodePodProbe) updatePodProbeStatus(pod *corev1.Pod, status app
 				podClone.Annotations[k] = v.(string)
 			}
 		}
-		if reflect.DeepEqual(oldStatus, podClone.Status) && reflect.DeepEqual(oldMetadata.Labels, podClone.Labels) &&
+		if reflect.DeepEqual(oldStatus.Conditions, podClone.Status.Conditions) && reflect.DeepEqual(oldMetadata.Labels, podClone.Labels) &&
 			reflect.DeepEqual(oldMetadata.Annotations, podClone.Annotations) {
 			return nil
 		}
@@ -372,6 +368,7 @@ func (r *ReconcileNodePodProbe) updatePodProbeStatus(pod *corev1.Pod, status app
 		klog.Errorf("NodePodProbe patch pod(%s/%s) status failed: %s", podClone.Namespace, podClone.Name, err.Error())
 		return err
 	}
-	klog.V(3).Infof("NodePodProbe update pod(%s/%s) status conditions(%s) success", podClone.Namespace, podClone.Name, util.DumpJSON(probeConditions))
+	klog.V(3).Infof("NodePodProbe update pod(%s/%s) metadata(%s) conditions(%s) success", podClone.Namespace, podClone.Name,
+		util.DumpJSON(probeMetadata), util.DumpJSON(probeConditions))
 	return nil
 }