fix(edas): add retry and k8s API fallback for service selector resolu…

internal/tools/orchestrator/scheduler/executor/plugins/edas/edas_app_flow.go

@@ -115,7 +115,7 @@
 	} else {
 		//查询最新一次的发布单，如果存在运行中则终止
 		orderList, _ := e.wrapEDASClient.ListRecentChangeOrderInfo(appID)
-		if len(orderList.ChangeOrder) > 0 && orderList.ChangeOrder[0].Status == 1 {
+		if orderList != nil && len(orderList.ChangeOrder) > 0 && orderList.ChangeOrder[0].Status == 1 {
 			e.wrapEDASClient.AbortChangeOrder(orderList.ChangeOrder[0].ChangeOrderId)
 		}
 
@@ -140,12 +140,37 @@
 }
 
 func (e *EDAS) resolveServiceSelector(appName, sgID, serviceName string) map[string]string {
-	var deployment *appsv1.Deployment
-	if currentDeployment, err := e.wrapEDASClient.GetAppDeployment(appName); err == nil {
-		deployment = currentDeployment
+	l := e.l.WithField("func", "resolveServiceSelector")
+
+	for i := 0; i < 3; i++ {
+		if dep, err := e.wrapEDASClient.GetAppDeployment(appName); err == nil && dep != nil {
+			selector := resolveServiceSelectorFromDeployment(dep, sgID, serviceName)
+			l.Infof("resolved selector from edas api, appName: %s, selector: %v", appName, selector)
+			return selector
+		} else if err != nil {
+			l.Warnf("failed to get deployment from edas api, appName: %s, attempt: %d, err: %v", appName, i+1, err)
+		}
+
+		if i < 2 {
+			time.Sleep(5 * time.Second)
+		}
+	}
+
+	// EDAS API 全部失败，通过 k8s API 最终兜底
+	l.Warnf("edas api all retries exhausted, falling back to k8s api, appName: %s", appName)
+	if dep, err := e.wrapClientSet.GetDeployment(context.Background(), appName); err == nil && dep != nil {
+		selector := resolveServiceSelectorFromDeployment(dep, sgID, serviceName)
+		l.Infof("resolved selector from k8s api fallback, appName: %s, selector: %v", appName, selector)
+		return selector
+	} else if err != nil {
+		l.Errorf("failed to get deployment from k8s api, appName: %s, err: %v", appName, err)
 	}
 
-	return resolveServiceSelectorFromDeployment(deployment, sgID, serviceName)
+	l.Errorf("failed to resolve selector from all sources, appName: %s, using default selector", appName)
+	return map[string]string{
+		types.LabelServiceName:    serviceName,
+		types.LabelServiceGroupID: sgID,
+	}
 }
 
 func resolveServiceSelectorFromDeployment(deployment *appsv1.Deployment, sgID, serviceName string) map[string]string {

internal/tools/orchestrator/scheduler/executor/plugins/edas/wrapclient/kubernetes/deployment.go

@@ -20,6 +20,7 @@
 	"strings"
 
 	"github.com/pkg/errors"
+	appsv1 "k8s.io/api/apps/v1"
 	apiv1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 
@@ -90,3 +91,18 @@
 	l.Debugf("old service list : %+v", services)
 	return nil
 }
+
+// GetDeployment get deployment by name via k8s API.
+// EDAS may alter the deployment name in k8s, so use list + contains match.
+func (e *wrapKubernetes) GetDeployment(ctx context.Context, name string) (*appsv1.Deployment, error) {
+	deployList, err := e.cs.AppsV1().Deployments(e.namespace).List(ctx, metav1.ListOptions{})
+	if err != nil {
+		return nil, err
+	}
+	for i := range deployList.Items {
+		if strings.Contains(deployList.Items[i].Name, name) {
+			return &deployList.Items[i], nil
+		}
+	}
+	return nil, errors.Errorf("deployment contains %q not found", name)
+}

internal/tools/orchestrator/scheduler/executor/plugins/edas/wrapclient/kubernetes/provider.go

@@ -18,6 +18,7 @@ import (
 	"context"
 
 	"github.com/sirupsen/logrus"
+	appsv1 "k8s.io/api/apps/v1"
 	corev1 "k8s.io/api/core/v1"
 	"k8s.io/client-go/kubernetes"
 
@@ -30,6 +31,7 @@ import (
 type Interface interface {
 	GetK8sService(name string) (*corev1.Service, error)
 	GetK8sDeployList(group string, services *[]apistructs.Service) error
+	GetDeployment(ctx context.Context, name string) (*appsv1.Deployment, error)
 	CreateK8sService(appName string, selectors map[string]string, ports []int) error
 	CreateOrUpdateK8sService(ctx context.Context, appName string, selectors map[string]string, ports []int) error
 	DeleteK8sService(appName string) error