enhance terminted pod handle

AGENTS.md

@@ -0,0 +1,31 @@
+# Terway Project Development Guide
+
+## Dev environment tips
+
+- Use `make help` to see all available make targets and their descriptions
+- Run `make generate` to generate code containing DeepCopy, DeepCopyInto, and DeepCopyObject method implementations
+- Use `make manifests` to generate CustomResourceDefinition objects
+- Run `make fmt` to format Go code with `go fmt`
+- Use `make vet` to run `go vet` against code with build tags
+
+## Testing instructions
+
+- Find the CI plan in the `.github/workflows` folder (check.yml, build.yml)
+- Run `make quick-test` to run all tests
+- Run `make lint` to run golangci-lint with the configuration in `.golangci.yml`
+- Run `make lint-fix` to run golangci-lint and perform automatic fixes
+- Fix any test or lint errors until the whole suite passes
+- After moving files or changing imports, run `make fmt` and `make vet` to ensure code quality
+- Add or update tests for the code you change, even if nobody asked
+
+## PR instructions
+
+- Title format: `[terway] <Title>` or `[component] <Title>`
+- Always run `make lint` and `make test` before committing
+- Ensure `go mod tidy` and `go mod vendor` are run before submitting (checked in CI)
+- Make sure all build tags are properly set (privileged, default_build)
+- Verify that your changes pass the GitHub Actions workflows:
+  - Code formatting and linting (golangci-lint)
+  - Unit tests with coverage
+  - Module vendoring checks
+  - Super-linter for markdown and bash files

Makefile

@@ -53,6 +53,10 @@ vet: ## Run go vet against code.
 test: manifests generate fmt vet envtest datapath-test## Run tests.
 	KUBEBUILDER_ASSETS="$(shell $(ENVTEST) use $(ENVTEST_K8S_VERSION) --bin-dir $(LOCALBIN) -p path)" go test -gcflags=all=-l -race --tags "$(GO_BUILD_TAGS)" $$(go list ./... | grep -Ev '/e2e|/mocks|/generated|/apis|/examples|/tests|/rpc|/windows|/internal/testutil') -coverprofile coverage.txt
 
+.PHONY: test-quick
+test-quick: manifests generate fmt vet envtest ## Run tests.
+	KUBEBUILDER_ASSETS="$(shell $(ENVTEST) use $(ENVTEST_K8S_VERSION) --bin-dir $(LOCALBIN) -p path)" go test -gcflags=all=-l -race --tags "$(GO_BUILD_TAGS)" $$(go list ./... | grep -Ev '/e2e|/mocks|/generated|/apis|/examples|/tests|/rpc|/windows|/internal/testutil') -coverprofile coverage.txt
+
 .PHONY: lint
 lint: golangci-lint ## Run golangci-lint linter & yamllint
 	$(GOLANGCI_LINT) run

cmd/terway-cli/cmd_test.go

@@ -6,8 +6,8 @@ import (
 	"testing"
 
 	"github.com/agiledragon/gomonkey/v2"
-	"github.com/stretchr/testify/assert"
 	"github.com/spf13/cobra"
+	"github.com/stretchr/testify/assert"
 
 	"github.com/AliyunContainerService/terway/pkg/aliyun/metadata"
 )
@@ -42,12 +42,12 @@ func TestPrintKV(t *testing.T) {
 func TestArgumentValidation(t *testing.T) {
 	t.Run("runList argument validation", func(t *testing.T) {
 		cmd := &cobra.Command{}
-		
+
 		// Test with exactly 2 arguments (should fail)
 		err := runList(cmd, []string{"arg1", "arg2"})
 		assert.Error(t, err)
 		assert.Contains(t, err.Error(), "too many arguments")
-		
+
 		// Test with more than 2 arguments (should fail)
 		err = runList(cmd, []string{"arg1", "arg2", "arg3"})
 		assert.Error(t, err)
@@ -56,17 +56,17 @@ func TestArgumentValidation(t *testing.T) {
 
 	t.Run("runShow argument validation", func(t *testing.T) {
 		cmd := &cobra.Command{}
-		
+
 		// Test with no arguments
 		err := runShow(cmd, []string{})
 		assert.Error(t, err)
 		assert.Contains(t, err.Error(), "no arguments")
-		
+
 		// Test with 2 or more arguments
 		err = runShow(cmd, []string{"type", "name"})
 		assert.Error(t, err)
 		assert.Contains(t, err.Error(), "too many arguments")
-		
+
 		// Test with 3 arguments
 		err = runShow(cmd, []string{"type", "name", "extra"})
 		assert.Error(t, err)
@@ -75,28 +75,28 @@ func TestArgumentValidation(t *testing.T) {
 
 	t.Run("runExecute argument validation", func(t *testing.T) {
 		cmd := &cobra.Command{}
-		
+
 		// Test with no arguments
 		err := runExecute(cmd, []string{})
 		assert.Error(t, err)
 		assert.Contains(t, err.Error(), "too few arguments")
-		
+
 		// Test with 1 argument
 		err = runExecute(cmd, []string{"type"})
 		assert.Error(t, err)
 		assert.Contains(t, err.Error(), "too few arguments")
-		
+
 		// Test with 2 arguments
 		err = runExecute(cmd, []string{"type", "name"})
 		assert.Error(t, err)
 		assert.Contains(t, err.Error(), "too few arguments")
-		
+
 		// Test with exactly 3 arguments should pass validation
 		// We don't call runExecute here since it will access nil client
 		// Instead, we verify the argument parsing logic
 		args := []string{"type", "name", "command"}
 		assert.True(t, len(args) >= 3, "Should have at least 3 arguments")
-		
+
 		// Verify argument assignment logic
 		typ, name, command := args[0], args[1], args[2]
 		remainingArgs := args[3:]
@@ -556,4 +556,4 @@ func TestRunMetadata_ErrorCases(t *testing.T) {
 		err := runMetadata(cmd, []string{})
 		assert.NoError(t, err)
 	})
-}
+}

daemon/builder.go

@@ -465,7 +465,7 @@ func (b *NetworkServiceBuilder) PostInitForCRDV2() *NetworkServiceBuilder {
 	if b.err != nil {
 		return b
 	}
-	crdv2 := eni.NewCRDV2(b.service.k8s.NodeName(), b.namespace)
+	crdv2 := eni.NewCRDV2(b.service.k8s.GetRestConfig(), b.service.k8s.NodeName(), b.namespace)
 	mgr := eni.NewManager(nil, 0, []eni.NetworkInterface{crdv2}, daemon.EniSelectionPolicy(b.config.EniSelectionPolicy), nil)
 
 	svc := b.RunENIMgr(b.ctx, mgr)

daemon/daemon.go

@@ -86,8 +86,8 @@ var serviceLog = logf.Log.WithName("server")
 var _ rpc.TerwayBackendServer = (*networkService)(nil)
 
 // return resource relation in db, or return nil.
-func (n *networkService) getPodResource(info *daemon.PodInfo) (daemon.PodResources, error) {
-	obj, err := n.resourceDB.Get(utils.PodInfoKey(info.Namespace, info.Name))
+func (n *networkService) getPodResource(podID string) (daemon.PodResources, error) {
+	obj, err := n.resourceDB.Get(podID)
 	if err == nil {
 		return obj.(daemon.PodResources), nil
 	}
@@ -98,9 +98,8 @@ func (n *networkService) getPodResource(info *daemon.PodInfo) (daemon.PodResourc
 	return daemon.PodResources{}, err
 }
 
-func (n *networkService) deletePodResource(info *daemon.PodInfo) error {
-	key := utils.PodInfoKey(info.Namespace, info.Name)
-	return n.resourceDB.Delete(key)
+func (n *networkService) deletePodResource(podID string) error {
+	return n.resourceDB.Delete(podID)
 }
 
 func (n *networkService) AllocIP(ctx context.Context, r *rpc.AllocIPRequest) (*rpc.AllocIPReply, error) {
@@ -163,7 +162,7 @@ func (n *networkService) AllocIP(ctx context.Context, r *rpc.AllocIPRequest) (*r
 	}
 
 	// 2. Find old resource info
-	oldRes, err := n.getPodResource(pod)
+	oldRes, err := n.getPodResource(podID)
 	if err != nil {
 		return nil, &types.Error{
 			Code: types.ErrInternalError,
@@ -331,30 +330,31 @@ func (n *networkService) ReleaseIP(ctx context.Context, r *rpc.ReleaseIPRequest)
 	// 0. Get pod Info
 	pod, err := n.k8s.GetPod(ctx, r.K8SPodNamespace, r.K8SPodName, true)
 	if err != nil {
-		if k8sErr.IsNotFound(err) {
-			return reply, nil
+		if !k8sErr.IsNotFound(err) {
+			l.Error(err, "get pod failed")
+			// ignore error, do not block delete
 		}
-		return nil, err
 	}
 
 	cni := &daemon.CNI{
 		PodName:      r.K8SPodName,
 		PodNamespace: r.K8SPodNamespace,
 		PodID:        podID,
-		PodUID:       pod.PodUID,
+	}
+
+	var podIpStickTime time.Duration
+	if pod != nil {
+		cni.PodUID = pod.PodUID
+		podIpStickTime = pod.IPStickTime
 	}
 
 	// 1. Init Context
 
-	oldRes, err := n.getPodResource(pod)
+	oldRes, err := n.getPodResource(podID)
 	if err != nil {
 		return nil, err
 	}
 
-	if !n.verifyPodNetworkType(pod.PodNetworkType) {
-		return nil, fmt.Errorf("unexpect pod network type allocate, maybe daemon mode changed: %+v", pod.PodNetworkType)
-	}
-
 	if oldRes.ContainerID != nil {
 		if r.K8SPodInfraContainerId != *oldRes.ContainerID {
 			l.Info("cni request not match stored resource, ignored", "old", *oldRes.ContainerID)
@@ -365,7 +365,7 @@ func (n *networkService) ReleaseIP(ctx context.Context, r *rpc.ReleaseIPRequest)
 		cni.PodUID = oldRes.PodInfo.PodUID
 	}
 
-	if n.ipamType == types.IPAMTypeCRD || pod.IPStickTime == 0 {
+	if n.ipamType == types.IPAMTypeCRD || podIpStickTime <= 0 {
 		for _, resource := range oldRes.Resources {
 			res := parseNetworkResource(resource)
 			if res == nil {
@@ -379,7 +379,7 @@ func (n *networkService) ReleaseIP(ctx context.Context, r *rpc.ReleaseIPRequest)
 				return nil, err
 			}
 		}
-		err = n.deletePodResource(pod)
+		err = n.deletePodResource(podID)
 		if err != nil {
 			return nil, fmt.Errorf("error delete pod resource: %w", err)
 		}
@@ -388,6 +388,7 @@ func (n *networkService) ReleaseIP(ctx context.Context, r *rpc.ReleaseIPRequest)
 	return reply, nil
 }
 
+// GetIPInfo return cached alloc ip info
 func (n *networkService) GetIPInfo(ctx context.Context, r *rpc.GetInfoRequest) (*rpc.GetInfoReply, error) {
 	podID := utils.PodInfoKey(r.K8SPodNamespace, r.K8SPodName)
 	log := logf.FromContext(ctx)
@@ -409,27 +410,17 @@ func (n *networkService) GetIPInfo(ctx context.Context, r *rpc.GetInfoRequest) (
 
 	var err error
 
-	// 0. Get pod Info
-	pod, err := n.k8s.GetPod(ctx, r.K8SPodNamespace, r.K8SPodName, true)
-	if err != nil {
-		return nil, &types.Error{
-			Code: types.ErrInvalidArgsErrCode,
-			Msg:  err.Error(),
-			R:    err,
-		}
-	}
-
 	// 1. Init Context
 	reply := &rpc.GetInfoReply{
 		Success: true,
 		IPv4:    n.enableIPv4,
 		IPv6:    n.enableIPv6,
 	}
 
-	switch pod.PodNetworkType {
-	case daemon.PodNetworkTypeENIMultiIP:
+	switch n.daemonMode {
+	case daemon.ModeENIMultiIP:
 		reply.IPType = rpc.IPType_TypeENIMultiIP
-	case daemon.PodNetworkTypeVPCENI:
+	case daemon.ModeENIOnly:
 		reply.IPType = rpc.IPType_TypeVPCENI
 	default:
 		return nil, &types.Error{
@@ -439,7 +430,7 @@ func (n *networkService) GetIPInfo(ctx context.Context, r *rpc.GetInfoRequest) (
 	}
 
 	// 2. Find old resource info
-	oldRes, err := n.getPodResource(pod)
+	oldRes, err := n.getPodResource(podID)
 	if err != nil {
 		return nil, &types.Error{
 			Code: types.ErrInternalError,
@@ -448,12 +439,6 @@ func (n *networkService) GetIPInfo(ctx context.Context, r *rpc.GetInfoRequest) (
 		}
 	}
 
-	if !n.verifyPodNetworkType(pod.PodNetworkType) {
-		return nil, &types.Error{
-			Code: types.ErrInvalidArgsErrCode,
-			Msg:  "Unexpected network type, maybe daemon mode changed",
-		}
-	}
 	if oldRes.ContainerID != nil {
 		if r.K8SPodInfraContainerId != *oldRes.ContainerID {
 			log.Info("cni request not match stored resource, ignored", "old", *oldRes.ContainerID)
@@ -640,7 +625,7 @@ func (n *networkService) gcPods(ctx context.Context) error {
 			}
 		}
 
-		err = n.deletePodResource(podRes.PodInfo)
+		err = n.deletePodResource(podID)
 		if err != nil {
 			return err
 		}