change main to v1.2.2

.github/workflows/build_push_concheck.yaml

@@ -4,12 +4,12 @@ on:
   push:
     branches:
       - main
-      - v1.2.1
+      - v1.2.2
     paths:
       - connection-check/**
 
 env:
-  IMAGE_VERSION: '1.2.1'
+  IMAGE_VERSION: '1.2.2'
   IMAGE_REGISTRY: ghcr.io/${{ github.repository_owner }}
 
 jobs:

.github/workflows/build_push_controller.yaml

@@ -4,7 +4,7 @@ on:
   push:
     branches:
       - main
-      - v1.2.1
+      - v1.2.2
     paths:
       - controllers/**
       - compute/**
@@ -17,7 +17,7 @@ on:
       - ./Makefile
 
 env:
-  VERSION: '1.2.1'
+  VERSION: '1.2.2'
   IMAGE_REGISTRY: ghcr.io/${{ github.repository_owner }}
   DAEMON_REGISTRY: ghcr.io/${{ github.repository_owner }}
 

.github/workflows/build_push_daemon.yaml

@@ -4,14 +4,14 @@ on:
   push:
     branches:
       - main
-      - v1.2.1
+      - v1.2.2
     paths:
       - daemon/**
       - cni/**
       - Makefile
 
 env:
-  IMAGE_VERSION: '1.2.1'
+  IMAGE_VERSION: '1.2.2'
   DAEMON_REGISTRY: ghcr.io/${{ github.repository_owner }}
 
 jobs:

Makefile

@@ -17,7 +17,7 @@ endif
 # - use the VERSION as arg of the bundle target (e.g make bundle VERSION=0.0.2)
 # - use environment variables to overwrite this value (e.g export VERSION=0.0.2)
 # VERSION ?= 0.0.1
-VERSION ?= 1.2.1
+VERSION ?= 1.2.2
 export CHANNELS = "alpha"
 
 # CHANNELS define the bundle channels used in the bundle.

README.md

@@ -119,7 +119,7 @@ Recommended to deploy in the same default namespace for [health check service](.
   ```
 ##### by bundle with operator-sdk
   ```bash
-  operator-sdk run bundle ghcr.io/foundation-model-stack/multi-nic-cni-bundle:v1.2.1 -n multi-nic-cni-operator
+  operator-sdk run bundle ghcr.io/foundation-model-stack/multi-nic-cni-bundle:v1.2.2 -n multi-nic-cni-operator
   ```
 #### Deploy MultiNicNetwork resource
 1. Prepare `network.yaml` as shown in the [example](#multinicnetwork)

config/manager/kustomization.yaml

@@ -13,4 +13,4 @@ kind: Kustomization
 images:
 - name: controller
   newName: ghcr.io/foundation-model-stack/multi-nic-cni-controller
-  newTag: v1.2.1
+  newTag: v1.2.2

config/samples/config.yaml

@@ -11,7 +11,7 @@ spec:
       value: "11000"
     - name: RT_TABLE_PATH
       value: /opt/rt_tables
-    image: ghcr.io/foundation-model-stack/multi-nic-cni-daemon:v1.2.1
+    image: ghcr.io/foundation-model-stack/multi-nic-cni-daemon:v1.2.2
     imagePullPolicy: Always
     mounts:
     - hostpath: /var/lib/cni/bin

config/samples/kustomization.yaml

@@ -15,4 +15,4 @@ kind: Kustomization
 images:
 - name: multi-nic-cni-daemon
   newName: ghcr.io/foundation-model-stack/multi-nic-cni-daemon
-  newTag: v1.2.1
+  newTag: v1.2.2

connection-check/concheck.yaml

@@ -71,7 +71,7 @@ spec:
       serviceAccountName: multi-nic-concheck-account
       containers:
       - name: concheck
-        image: ghcr.io/foundation-model-stack/multi-nic-cni-concheck:v1.2.1
+        image: ghcr.io/foundation-model-stack/multi-nic-cni-concheck:v1.2.2
         imagePullPolicy: Always
         securityContext:
           privileged: true

controllers/vars/vars.go

@@ -39,7 +39,7 @@ const (
 	DefaultOperatorNamespace                  = "multi-nic-cni-operator"
 	DefaultCNIType                            = "multi-nic"
 	DefaultIPAMType                           = "multi-nic-ipam"
-	DefaultDaemonImage                        = "ghcr.io/foundation-model-stack/multi-nic-cni-daemon:v1.2.1"
+	DefaultDaemonImage                        = "ghcr.io/foundation-model-stack/multi-nic-cni-daemon:v1.2.2"
 	DefaultJoinPath                           = "/join"
 	DefaultInterfacePath                      = "/interface"
 	DefaultAddRoutePath                       = "/addl3"

daemon/Makefile

@@ -6,7 +6,7 @@ export DAEMON_REGISTRY ?= ghcr.io/foundation-model-stack
 
 # DAEMON_IMG defines the image:tag used for daemon
 IMAGE_TAG_BASE = $(DAEMON_REGISTRY)/multi-nic-cni
-IMAGE_VERSION ?= 1.2.1
+IMAGE_VERSION ?= 1.2.2
 DAEMON_IMG ?= $(IMAGE_TAG_BASE)-daemon:v$(IMAGE_VERSION)
 
 

daemon/src/iface/iface.go

@@ -164,3 +164,28 @@ func GetInterfaces() []backend.InterfaceInfoType {
 	}
 	return interfaces
 }
+
+func getNetAddressFromDevice(devName string) (string, error) {
+	devLink, err := netlink.LinkByName(devName)
+	if err != nil {
+		log.Printf("cannot find link %s: %v", devName, err)
+		return "", err
+	}
+	addrs, err := netlink.AddrList(devLink, netlink.FAMILY_V4)
+	if err != nil || len(addrs) == 0 {
+		log.Printf("cannot list address on %s: %v", devName, err)
+		return "", err
+	}
+	addr := addrs[0].IPNet
+	if addr == nil {
+		log.Printf("no address set on %s", devName)
+		return "", err
+	}
+	if devLink.Attrs().Flags&net.FlagUp == 0 {
+		// interface down
+		log.Printf("%s down", devName)
+		return "", err
+	}
+	netAddress := getNetAddress(addr)
+	return netAddress, nil
+}

daemon/src/iface/pci.go

@@ -26,6 +26,8 @@ const (
 
 var CheckPointfile string = "/var/lib/kubelet/device-plugins/kubelet_internal_checkpoint"
 
+var deviceMapCache = InitSafeCache()
+
 // modify from https://github.com/k8snetworkplumbingwg/multus-cni/blob/9b45d4b211728aa0db44a1624aac8e61843390cf/pkg/checkpoint/checkpoint.go#L72
 // DeviceIDs can map[string]string or []string
 type PodDevicesEntry struct {
@@ -121,10 +123,32 @@ func GetDeviceMap(resourceMap map[string][]string, resourceName string) map[stri
 
 	if deviceIDs, exist := resourceMap[resourceName]; exist {
 		for _, deviceID := range deviceIDs {
-			masterName, err := GetPfName(deviceID)
-			if err == nil {
-				if netAddress, exist := nameNetMap[masterName]; exist {
+			var masterName string
+			deviceNameInterface := deviceMapCache.GetCache(deviceID)
+			if deviceNameInterface != nil {
+				masterName = deviceNameInterface.(string)
+			} else {
+				var err error
+				masterName, err = GetPfName(deviceID)
+				if err != nil {
+					log.Printf("cannot get physical device %s: %v\n", deviceID, err)
+				} else {
+					log.Printf("set deviceMapCache %s=%s\n", deviceID, masterName)
+					deviceMapCache.SetCache(deviceID, masterName)
+				}
+			}
+			if netAddress, exist := nameNetMap[masterName]; exist {
+				deviceMap[netAddress] = deviceID
+			} else {
+				netAddress, err := getNetAddressFromDevice(masterName)
+				if err != nil {
+					log.Printf("cannot get network address of device %s: %v\n", masterName, err)
+				} else {
 					deviceMap[netAddress] = deviceID
+					// found new device, update interfaces
+					GetInterfaces()
+					nameNetMap = GetNameNetMap()
+					log.Printf("updated nameNetMap map: %v\n", nameNetMap)
 				}
 			}
 		}

daemon/src/selector/selector.go

@@ -71,8 +71,6 @@ func getDefaultResponse(req NICSelectRequest, masterNameMap map[string]string, n
 }
 
 func Select(req NICSelectRequest) NICSelectResponse {
-	masterNameMap := iface.GetInterfaceNameMap()
-	nameNetMap := iface.GetNameNetMap()
 	deviceMap := make(map[string]string)
 	resourceMap := make(map[string][]string)
 
@@ -93,6 +91,8 @@ func Select(req NICSelectRequest) NICSelectResponse {
 		log.Printf("Cannot get pod: %v\n", err)
 	}
 
+	masterNameMap := iface.GetInterfaceNameMap()
+	nameNetMap := iface.GetNameNetMap()
 	netSpec, err := MultinicnetHandler.Get(req.NetAttachDefName, req.PodNamespace)
 	if err != nil {
 		return getDefaultResponse(req, masterNameMap, nameNetMap, deviceMap, resourceMap)