[NPM][Fix] Adding redundant check to ignore endpoint not found while …

network/hnswrapper/hnsv2wrapperfake.go

@@ -203,7 +203,7 @@ func (f Hnsv2wrapperFake) GetEndpointByID(endpointID string) (*hcn.HostComputeEn
 	if ep, ok := f.Cache.endpoints[endpointID]; ok {
 		return ep.GetHCNObj(), nil
 	}
-	return &hcn.HostComputeEndpoint{}, nil
+	return &hcn.HostComputeEndpoint{}, hcn.EndpointNotFoundError{EndpointID: endpointID}
 }
 
 func (f Hnsv2wrapperFake) CreateEndpoint(endpoint *hcn.HostComputeEndpoint) (*hcn.HostComputeEndpoint, error) {

npm/pkg/dataplane/policies/policymanager_windows.go

@@ -174,8 +174,9 @@ func (pMgr *PolicyManager) removePolicy(policy *NPMNetworkPolicy, endpointList m
 func (pMgr *PolicyManager) removePolicyByEndpointID(ruleID, epID string, noOfRulesToRemove int, resetAllACL shouldResetAllACLs) error {
 	epObj, err := pMgr.ioShim.Hns.GetEndpointByID(epID)
 	if err != nil {
-		if isNotFoundErr(err) {
-			klog.Infof("[PolicyManagerWindows] ignoring remove policy on endpoint since the endpoint wasn't found. the corresponding pod was most likely deleted. policy: %s, endpoint: %s", ruleID, epID)
+		// IsNotFound check is being skipped at times. So adding a redundant check here.
+		if isNotFoundErr(err) || strings.Contains(err.Error(), "endpoint was not found") {
+			klog.Infof("[PolicyManagerWindows] ignoring remove policy since the endpoint wasn't found. the corresponding pod might be deleted. policy: %s, endpoint: %s, err: %s", ruleID, epID, err.Error())
 			return nil
 		}
 		return fmt.Errorf("[PolicyManagerWindows] failed to remove policy while getting the endpoint. policy: %s, endpoint: %s, err: %w", ruleID, epID, err)
@@ -221,9 +222,10 @@ func (pMgr *PolicyManager) removePolicyByEndpointID(ruleID, epID string, noOfRul
 func (pMgr *PolicyManager) applyPoliciesToEndpointID(epID string, policies hcn.PolicyEndpointRequest) error {
 	epObj, err := pMgr.ioShim.Hns.GetEndpointByID(epID)
 	if err != nil {
-		if isNotFoundErr(err) {
+		// IsNotFound check is being skipped at times. So adding a redundant check here.
+		if isNotFoundErr(err) || strings.Contains(err.Error(), "endpoint was not found") {
 			// unlikely scenario where an endpoint is deleted right after we refresh HNS endpoints, or an unlikely scenario where an endpoint is deleted right after we refresh HNS endpoints
-			metrics.SendErrorLogAndMetric(util.IptmID, "[PolicyManagerWindows] ignoring apply policies to endpoint since the endpoint wasn't found. endpoint: %s", epID)
+			metrics.SendErrorLogAndMetric(util.IptmID, "[PolicyManagerWindows] ignoring apply policies to endpoint since the endpoint wasn't found. endpoint: %s, err: %s", epID, err.Error())
 			return nil
 		}
 		return fmt.Errorf("[PolicyManagerWindows] to apply policies while getting the endpoint. endpoint: %s, err: %w", epID, err)

npm/pkg/dataplane/policies/policymanager_windows_test.go

@@ -129,6 +129,32 @@ func TestRemovePolicies(t *testing.T) {
 	verifyACLCacheIsCleaned(t, hns, len(endPointIDList))
 }
 
+func TestApplyPoliciesEndpointNotFound(t *testing.T) {
+	pMgr, _ := getPMgr(t)
+	testendPointIDList := map[string]string{
+		"10.0.0.5": "test10",
+	}
+	err := pMgr.AddPolicy(TestNetworkPolicies[0], testendPointIDList)
+	require.NoError(t, err)
+}
+
+func TestRemovePoliciesEndpointNotFound(t *testing.T) {
+	pMgr, hns := getPMgr(t)
+	err := pMgr.AddPolicy(TestNetworkPolicies[0], endPointIDList)
+	require.NoError(t, err)
+
+	aclID := TestNetworkPolicies[0].ACLPolicyID
+
+	_, err = hns.Cache.ACLPolicies(endPointIDList, aclID)
+	require.NoError(t, err)
+	testendPointIDList := map[string]string{
+		"10.0.0.5": "test10",
+	}
+	err = pMgr.RemovePolicy(TestNetworkPolicies[0].PolicyKey, testendPointIDList)
+	require.NoError(t, err, err)
+	verifyACLCacheIsCleaned(t, hns, len(endPointIDList))
+}
+
 // Helper functions for UTS
 
 func getPMgr(t *testing.T) (*PolicyManager, *hnswrapper.Hnsv2wrapperFake) {