[increment][refactor] refactored to use podname/namespace instead of claim UID. I think I'm missing default networks for DRA items

cmd/dra-multus-driver/example.yaml

@@ -6,6 +6,7 @@ metadata:
   namespace: default
 spec:
   config: '{
+    "name":  "bridge-net",
     "cniVersion": "0.4.0",
     "type": "bridge",
     "bridge": "cni0",

cmd/dra-multus-driver/state.go

@@ -18,6 +18,7 @@ import (
 	configapi "gopkg.in/k8snetworkplumbingwg/multus-cni.v4/pkg/dra/api/multus-cni.io/resource/net/v1alpha1"
 	multusk8sutils "gopkg.in/k8snetworkplumbingwg/multus-cni.v4/pkg/k8sclient"
 	"gopkg.in/k8snetworkplumbingwg/multus-cni.v4/pkg/types"
+	multustypes "gopkg.in/k8snetworkplumbingwg/multus-cni.v4/pkg/types"
 
 	cdiapi "tags.cncf.io/container-device-interface/pkg/cdi"
 	cdispec "tags.cncf.io/container-device-interface/specs-go"
@@ -202,7 +203,21 @@ func (s *DeviceState) prepareDevices(claim *resourceapi.ResourceClaim) (Prepared
 		results = append(results, &claim.Status.Allocation.Devices.Results[i])
 	}
 
-	perDeviceCDIContainerEdits, err := s.applyConfig(netConfig, results, claim.Namespace, string(claim.UID))
+	var podName, podNamespace string
+
+	for _, owner := range claim.OwnerReferences {
+		if owner.Kind == "Pod" && owner.Name != "" {
+			podName = owner.Name
+			podNamespace = claim.Namespace
+			break
+		}
+	}
+
+	if podName == "" {
+		return nil, fmt.Errorf("could not determine owning pod from claim metadata")
+	}
+
+	perDeviceCDIContainerEdits, err := s.applyConfig(netConfig, results, podName, podNamespace, string(claim.UID))
 
 	if err != nil {
 		return nil, fmt.Errorf("failed to apply CDI container edits: %w", err)
@@ -228,6 +243,7 @@ func (s *DeviceState) prepareDevices(claim *resourceapi.ResourceClaim) (Prepared
 func (s *DeviceState) applyConfig(
 	config *configapi.NetConfig,
 	results []*resourceapi.DeviceRequestAllocationResult,
+	podName string,
 	podNamespace string,
 	claimUID string,
 ) (PerDeviceCDIContainerEdits, error) {
@@ -248,11 +264,10 @@ func (s *DeviceState) applyConfig(
 
 		klog.Infof("!bang: Whole net-attach-def: %+v", nad)
 
-		delegate := &types.DelegateNetConf{
-			Name: net.Name,
-		}
-		if err := json.Unmarshal([]byte(nad.Spec.Config), &delegate.Conf); err != nil {
-			return nil, fmt.Errorf("failed to unmarshal NAD config into Conf: %w", err)
+		delegate, err := multustypes.LoadDelegateNetConf([]byte(nad.Spec.Config), net, "", "")
+		if err != nil {
+			// Handle error loading delegate
+			return nil, fmt.Errorf("failed to load delegate netconf from NAD %s/%s: %w", net.Namespace, net.Name, err)
 		}
 
 		// Preserve ifname from network selection
@@ -264,7 +279,7 @@ func (s *DeviceState) applyConfig(
 	}
 
 	// Save delegates to a file
-	delegatesPath := filepath.Join("/run/k8s.cni.cncf.io/dra", claimUID+".json")
+	delegatesPath := filepath.Join("/run/k8s.cni.cncf.io/dra", fmt.Sprintf("%s_%s.json", podNamespace, podName))
 	if err := os.MkdirAll(filepath.Dir(delegatesPath), 0755); err != nil {
 		return nil, fmt.Errorf("failed to ensure delegate output dir: %w", err)
 	}
@@ -281,6 +296,8 @@ func (s *DeviceState) applyConfig(
 		envs := []string{
 			fmt.Sprintf("MULTUS_DRA_DEVICE_NAME=%s", result.Device),
 			fmt.Sprintf("MULTUS_DRA_NETWORKS=%s", config.Networks),
+			fmt.Sprintf("MULTUS_DRA_POD_NAMESPACE=%s", podNamespace),
+			fmt.Sprintf("MULTUS_DRA_POD_NAME=%s", podName),
 			fmt.Sprintf("MULTUS_DRA_CLAIM_UID=%s", claimUID),
 		}
 

pkg/drahelpers/drahelpers.go

@@ -0,0 +1,27 @@
+package drahelpers
+
+import (
+	"encoding/json"
+	"fmt"
+	"os"
+
+	"gopkg.in/k8snetworkplumbingwg/multus-cni.v4/pkg/types"
+)
+
+const DRADelegateDir = "/run/k8s.cni.cncf.io/dra"
+
+func LoadDelegatesFromDRAFile(podName string, podNamespace string) ([]*types.DelegateNetConf, error) {
+
+	delegatePath := fmt.Sprintf("%s/%s_%s.json", DRADelegateDir, podNamespace, podName)
+	data, err := os.ReadFile(delegatePath)
+	if err != nil {
+		return nil, fmt.Errorf("failed to read delegate file %s: %w", delegatePath, err)
+	}
+
+	var delegates []*types.DelegateNetConf
+	if err := json.Unmarshal(data, &delegates); err != nil {
+		return nil, fmt.Errorf("failed to unmarshal delegate JSON: %w", err)
+	}
+
+	return delegates, nil
+}

pkg/multus/multus.go

@@ -39,6 +39,7 @@ import (
 	k8snet "k8s.io/apimachinery/pkg/util/net"
 	"k8s.io/apimachinery/pkg/util/wait"
 
+	"gopkg.in/k8snetworkplumbingwg/multus-cni.v4/pkg/drahelpers"
 	k8s "gopkg.in/k8snetworkplumbingwg/multus-cni.v4/pkg/k8sclient"
 	"gopkg.in/k8snetworkplumbingwg/multus-cni.v4/pkg/logging"
 	"gopkg.in/k8snetworkplumbingwg/multus-cni.v4/pkg/netutils"
@@ -647,34 +648,61 @@ func CmdAdd(args *skel.CmdArgs, exec invoke.Exec, kubeClient *k8s.ClientInfo) (c
 		}
 	}
 
-	pod, err := GetPod(kubeClient, k8sArgs, false)
-	if err != nil {
-		if err == errPodNotFound {
-			logging.Verbosef("CmdAdd: Warning: pod [%s/%s] not found, exiting with empty CNI result", k8sArgs.K8S_POD_NAMESPACE, k8sArgs.K8S_POD_NAME)
-			return &cni100.Result{
-				CNIVersion: n.CNIVersion,
-			}, nil
-		}
-		return nil, err
-	}
+	// !bang. This is where we're going to enter the netns and get the env variables.
+
+	var pod *v1.Pod
 
 	// resourceMap holds Pod device allocation information; only initizized if CRD contains 'resourceName' annotation.
 	// This will only be initialized once and all delegate objects can reference this to look up device info.
 	var resourceMap map[string]*types.ResourceInfo
 
-	if n.ClusterNetwork != "" {
-		resourceMap, err = k8s.GetDefaultNetworks(pod, n, kubeClient, resourceMap)
+	// !bang here we're probably going to have to get the delegates another way if we have the env vars.
+	// then we conditionally do this.
+
+	// Load from env-based file
+	var kc *k8s.ClientInfo
+	useddra := false
+	delegates, err := drahelpers.LoadDelegatesFromDRAFile(string(k8sArgs.K8S_POD_NAME), string(k8sArgs.K8S_POD_NAMESPACE))
+	if err != nil {
+		// Fall back to usual kube-based logic
+		logging.Errorf("CmdAdd: No DRA delegates from env in netns: %v", err)
+
+		pod, err = GetPod(kubeClient, k8sArgs, false)
 		if err != nil {
-			return nil, cmdErr(k8sArgs, "failed to get clusterNetwork/defaultNetworks: %v", err)
+			if err == errPodNotFound {
+				logging.Verbosef("CmdAdd: Warning: pod [%s/%s] not found, exiting with empty CNI result", k8sArgs.K8S_POD_NAMESPACE, k8sArgs.K8S_POD_NAME)
+				return &cni100.Result{
+					CNIVersion: n.CNIVersion,
+				}, nil
+			}
+			return nil, err
 		}
-		// First delegate is always the master plugin
-		n.Delegates[0].MasterPlugin = true
+
+		if n.ClusterNetwork != "" {
+			resourceMap, err = k8s.GetDefaultNetworks(pod, n, kubeClient, resourceMap)
+			if err != nil {
+				return nil, cmdErr(k8sArgs, "failed to get clusterNetwork/defaultNetworks: %v", err)
+			}
+			// First delegate is always the master plugin
+			n.Delegates[0].MasterPlugin = true
+		}
+
+		_, kc, err = k8s.TryLoadPodDelegates(pod, n, kubeClient, resourceMap)
+		if err != nil {
+			return nil, cmdErr(k8sArgs, "error loading k8s delegates k8s args: %v", err)
+		}
+
+	} else if len(delegates) > 0 {
+		// Successfully loaded delegates from env, use them
+		logging.Verbosef("CmdAdd: Successfully loaded %d DRA delegates from env in netns for pod [%s/%s]", len(delegates), k8sArgs.K8S_POD_NAMESPACE, k8sArgs.K8S_POD_NAME)
+		n.Delegates = delegates
+		useddra = true
+	} else {
+		// I think this is just like there's no additional delegates?
 	}
 
-	_, kc, err := k8s.TryLoadPodDelegates(pod, n, kubeClient, resourceMap)
-	if err != nil {
-		return nil, cmdErr(k8sArgs, "error loading k8s delegates k8s args: %v", err)
-	}
+	// if we have the  env var -- then here we'd load the delegates from file.
+	// then lets put them in n.Delegates so we can use the following logic.
 
 	// cache the multus config
 	if err := saveDelegates(args.ContainerID, n.CNIDir, n.Delegates); err != nil {
@@ -792,7 +820,8 @@ func CmdAdd(args *skel.CmdArgs, exec invoke.Exec, kubeClient *k8s.ClientInfo) (c
 		}
 
 		// Create the network statuses, only in case Multus has kubeconfig
-		if kubeClient != nil && kc != nil {
+		// !bang TODO: importantly, we're skipping this when using DRA for now, because we just want to "not use k8s at CNI time" with this pattern.
+		if kubeClient != nil && kc != nil && !useddra {
 			if !types.CheckSystemNamespaces(string(k8sArgs.K8S_POD_NAME), n.SystemNamespaces) {
 				delegateNetStatuses, err := nadutils.CreateNetworkStatuses(tmpResult, delegate.Name, delegate.MasterPlugin, devinfo)
 				if err != nil {