Use podIP rather than hostNetwork

README.md

@@ -50,6 +50,7 @@ metadata:
   name: torrent-sample
 spec:
   hub:
+    name: Huggingface
     repoID: Qwen/Qwen2.5-0.5B-Instruct
 ```
 
@@ -62,6 +63,7 @@ metadata:
   name: torrent-sample
 spec:
   hub:
+    name: Huggingface
     repoID: Qwen/Qwen2.5-0.5B-Instruct
   nodeSelector:
     zone: zone-a
@@ -78,6 +80,7 @@ metadata:
   name: torrent-sample
 spec:
   hub:
+    name: Huggingface
     repoID: Qwen/Qwen2.5-0.5B-Instruct
   reclaimPolicy: Delete
 ```

agent/config/base/clusterrole.yaml

@@ -47,3 +47,11 @@ rules:
   - get
   - list
   - watch
+- apiGroups:
+  - ""
+  resources:
+  - pods
+  verbs:
+  - get
+  - list
+  - watch

agent/config/manager/daemonset.yaml

@@ -4,6 +4,7 @@ metadata:
   name: manta-agent
   namespace: manta-system
   labels:
+    # This is required for manta when fetching peer IP.
     app: manta-agent
 spec:
   selector:
@@ -14,7 +15,6 @@ spec:
       labels:
         app: manta-agent
     spec:
-      hostNetwork: true
       serviceAccountName: manta-agent
       initContainers:
       - name: init-permissions

agent/pkg/controller/replication_controller.go

@@ -20,6 +20,7 @@ import (
 	"context"
 	"os"
 
+	corev1 "k8s.io/api/core/v1"
 	apimeta "k8s.io/apimachinery/pkg/api/meta"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/runtime"
@@ -92,6 +93,7 @@ func (r *ReplicationReconciler) Reconcile(ctx context.Context, req ctrl.Request)
 	}
 
 	// This may take a long time, the concurrency is controlled by the MaxConcurrentReconciles.
+	// TODO: should we create a Job to handle this? See discussion: https://github.com/InftyAI/Manta/issues/25
 	if err := handler.HandleReplication(ctx, r.Client, replication); err != nil {
 		logger.Error(err, "error to handle replication", "Replication", klog.KObj(replication))
 		return ctrl.Result{}, err
@@ -144,6 +146,13 @@ func (r *ReplicationReconciler) updateNodeTracker(ctx context.Context, replicati
 
 // SetupWithManager sets up the controller with the Manager.
 func (r *ReplicationReconciler) SetupWithManager(mgr ctrl.Manager) error {
+	if err := mgr.GetFieldIndexer().IndexField(context.TODO(), &corev1.Pod{}, "spec.nodeName", func(rawObj client.Object) []string {
+		pod := rawObj.(*corev1.Pod)
+		return []string{pod.Spec.NodeName}
+	}); err != nil {
+		return err
+	}
+
 	return ctrl.NewControllerManagedBy(mgr).
 		For(&api.Replication{}).
 		WithOptions(controller.Options{MaxConcurrentReconciles: 5}).

agent/pkg/handler/chunk_handler.go

@@ -72,8 +72,8 @@ func SendChunk(w http.ResponseWriter, r *http.Request) {
 	}
 }
 
-func recvChunk(blobPath, snapshotPath, ipAddr string) error {
-	url := fmt.Sprintf("http://%s:%s/sync?path=%s", ipAddr, api.HttpPort, blobPath)
+func recvChunk(blobPath, snapshotPath, addr string) error {
+	url := fmt.Sprintf("http://%s:%s/sync?path=%s", addr, api.HttpPort, blobPath)
 
 	resp, err := http.Get(url)
 	if err != nil {

agent/pkg/handler/replication_handler.go → agent/pkg/handler/handler.go

@@ -24,7 +24,8 @@ import (
 	"strings"
 
 	corev1 "k8s.io/api/core/v1"
-	"k8s.io/apimachinery/pkg/types"
+	"k8s.io/apimachinery/pkg/fields"
+	"k8s.io/apimachinery/pkg/labels"
 	"k8s.io/klog/v2"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 	"sigs.k8s.io/controller-runtime/pkg/log"
@@ -104,15 +105,15 @@ func syncChunk(ctx context.Context, client client.Client, replication *api.Repli
 	sourceSplits := strings.Split(*replication.Spec.Source.URI, "://")
 	addresses := strings.Split(sourceSplits[1], "@")
 	nodeName, blobPath := addresses[0], addresses[1]
-	nodeIP, err := nodeIP(ctx, client, nodeName)
+	addr, err := peerAddr(ctx, client, nodeName)
 	if err != nil {
 		return err
 	}
 
 	// The destination URI looks like localhost://<path-to-your-file>
 	destSplits := strings.Split(*replication.Spec.Destination.URI, "://")
 
-	if err := recvChunk(blobPath, destSplits[1], nodeIP); err != nil {
+	if err := recvChunk(blobPath, destSplits[1], addr); err != nil {
 		logger.Error(err, "failed to sync chunk")
 		return err
 	}
@@ -196,15 +197,22 @@ func parseURI(uri string) (host string, address string) {
 	return splits[0], splits[1]
 }
 
-func nodeIP(ctx context.Context, client client.Client, nodeName string) (string, error) {
-	node := corev1.Node{}
-	if err := client.Get(ctx, types.NamespacedName{Name: nodeName}, &node); err != nil {
+func peerAddr(ctx context.Context, c client.Client, nodeName string) (string, error) {
+	fieldSelector := fields.OneTermEqualSelector("spec.nodeName", nodeName)
+	listOptions := &client.ListOptions{
+		FieldSelector: fieldSelector,
+		// HACK: the label is hacked, we must set it explicitly in the daemonSet.
+		LabelSelector: labels.SelectorFromSet(map[string]string{"app": "manta-agent"}),
+	}
+
+	pods := corev1.PodList{}
+	if err := c.List(ctx, &pods, listOptions); err != nil {
 		return "", err
 	}
-	for _, address := range node.Status.Addresses {
-		if address.Type == "InternalIP" {
-			return address.Address, nil
-		}
+
+	if len(pods.Items) != 1 {
+		return "", fmt.Errorf("got more than one pod per node for daemonSet")
 	}
-	return "", fmt.Errorf("can't get node internal IP")
+
+	return pods.Items[0].Status.PodIP, nil
 }

agent/pkg/task/task.go

@@ -23,7 +23,6 @@ import (
 	"path/filepath"
 	"time"
 
-	"github.com/go-logr/logr"
 	corev1 "k8s.io/api/core/v1"
 	apierrors "k8s.io/apimachinery/pkg/api/errors"
 	v1 "k8s.io/apimachinery/pkg/apis/meta/v1"
@@ -43,25 +42,20 @@ const (
 	workspace = cons.DefaultWorkspace
 )
 
-var (
-	logger logr.Logger
-)
-
 func BackgroundTasks(ctx context.Context, c client.Client) {
-	ctrl.SetLogger(zap.New(zap.UseDevMode(true)))
-	logger = ctrl.Log.WithName("Background")
-
 	// Sync the disk chunk infos to the nodeTracker.
 	go syncChunks(ctx, c)
 }
 
 func syncChunks(ctx context.Context, c client.Client) {
+	ctrl.SetLogger(zap.New(zap.UseDevMode(true)))
+	logger := ctrl.Log.WithName("Background tasks")
+
 	forFunc := func(ctx context.Context) error {
 		attempts := 0
 		for {
 			attempts += 1
 			if err := findOrCreateNodeTracker(ctx, c); err != nil {
-				// fmt.Printf("failed to create nodeTracker: %v, retry.", err)
 				logger.Error(err, "Failed to create nodeTracker, retry...")
 
 				if attempts > 10 {

cmd/main.go

@@ -39,7 +39,6 @@ import (
 	"github.com/inftyai/manta/pkg/dispatcher"
 	"github.com/inftyai/manta/pkg/dispatcher/framework"
 	"github.com/inftyai/manta/pkg/dispatcher/plugins/diskaware"
-	"github.com/inftyai/manta/pkg/dispatcher/plugins/gnumber"
 	"github.com/inftyai/manta/pkg/dispatcher/plugins/nodeselector"
 	"github.com/inftyai/manta/pkg/webhook"
 	//+kubebuilder:scaffold:imports
@@ -133,7 +132,7 @@ func setupControllers(mgr ctrl.Manager, certsReady chan struct{}) {
 	<-certsReady
 	setupLog.Info("certs ready")
 
-	dispatcher, err := dispatcher.NewDispatcher([]framework.RegisterFunc{nodeselector.New, diskaware.New, gnumber.New})
+	dispatcher, err := dispatcher.NewDispatcher([]framework.RegisterFunc{nodeselector.New, diskaware.New})
 	if err != nil {
 		setupLog.Error(err, "unable to create dispatcher")
 		os.Exit(1)

pkg/controller/nodetracker_controller.go

@@ -23,7 +23,6 @@ import (
 	corev1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/runtime"
 	"k8s.io/apimachinery/pkg/types"
-	"k8s.io/klog/v2"
 	ctrl "sigs.k8s.io/controller-runtime"
 	"sigs.k8s.io/controller-runtime/pkg/builder"
 	"sigs.k8s.io/controller-runtime/pkg/client"
@@ -91,8 +90,6 @@ func (r *NodeTrackerReconciler) Create(e event.CreateEvent) bool {
 		return false
 	}
 
-	logger := log.FromContext(context.Background()).WithValues("NodeTracker", klog.KObj(nodeTracker))
-	logger.Info("NodeTracker create event")
 	r.dispatcher.AddNodeTracker(nodeTracker)
 	return true
 }
@@ -110,7 +107,11 @@ func (r *NodeTrackerReconciler) Update(e event.UpdateEvent) bool {
 }
 
 func (r *NodeTrackerReconciler) Delete(e event.DeleteEvent) bool {
-	obj := e.Object.(*api.NodeTracker)
+	obj, match := e.Object.(*api.NodeTracker)
+	if !match {
+		return false
+	}
+
 	r.dispatcher.DeleteNodeTracker(obj)
 	return true
 }

test/e2e/suit_test.go

@@ -26,7 +26,6 @@ import (
 	admissionv1 "k8s.io/api/admission/v1"
 	corev1 "k8s.io/api/core/v1"
 	"k8s.io/apiextensions-apiserver/pkg/client/clientset/clientset/scheme"
-	"k8s.io/apimachinery/pkg/types"
 	"k8s.io/client-go/rest"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 	"sigs.k8s.io/controller-runtime/pkg/client/config"
@@ -35,7 +34,6 @@ import (
 
 	api "github.com/inftyai/manta/api/v1alpha1"
 	"github.com/inftyai/manta/test/util"
-	"github.com/inftyai/manta/test/util/wrapper"
 )
 
 var cfg *rest.Config
@@ -78,28 +76,6 @@ var _ = AfterSuite(func() {
 })
 
 func readyForTesting(client client.Client) {
-	By("waiting for webhooks to ready")
-
-	// To verify that webhooks are ready, let's create a simple Replication.
-	replication := wrapper.MakeReplication("sample-replication").
-		NodeName("unknown-node").
-		ChunkName("chunk1").
-		SizeBytes(1024).
-		SourceOfURI("localhost:///workspace/models/modelA").
-		Obj()
-
-	// Once the creation succeeds, that means the webhooks are ready
-	// and we can begin testing.
-	Eventually(func() error {
-		return client.Create(ctx, replication)
-	}, util.Timeout, util.Interval).Should(Succeed())
-
-	// Delete this replication before beginning tests.
-	Expect(client.Delete(ctx, replication)).To(Succeed())
-	Eventually(func() error {
-		return client.Get(ctx, types.NamespacedName{Name: replication.Name}, &api.Replication{})
-	}).ShouldNot(Succeed())
-
 	By("waiting for nodeTrackers to ready")
 	Eventually(func() error {
 		nodeTrackers := &api.NodeTrackerList{}
@@ -110,5 +86,5 @@ func readyForTesting(client client.Client) {
 			return fmt.Errorf("no nodeTrackers")
 		}
 		return nil
-	}, util.Timeout, util.Interval).Should(Succeed())
+	}, util.Timeout*3, util.Interval).Should(Succeed())
 }

test/integration/controller/suite_test.go

@@ -42,7 +42,6 @@ import (
 	"github.com/inftyai/manta/pkg/dispatcher"
 	"github.com/inftyai/manta/pkg/dispatcher/framework"
 	"github.com/inftyai/manta/pkg/dispatcher/plugins/diskaware"
-	"github.com/inftyai/manta/pkg/dispatcher/plugins/gnumber"
 	"github.com/inftyai/manta/pkg/dispatcher/plugins/nodeselector"
 )
 
@@ -106,7 +105,7 @@ var _ = BeforeSuite(func() {
 	})
 	Expect(err).ToNot(HaveOccurred())
 
-	dispatcher, err := dispatcher.NewDispatcher([]framework.RegisterFunc{nodeselector.New, diskaware.New, gnumber.New})
+	dispatcher, err := dispatcher.NewDispatcher([]framework.RegisterFunc{nodeselector.New, diskaware.New})
 	Expect(err).ToNot(HaveOccurred())
 
 	torrentController := controller.NewTorrentReconciler(mgr.GetClient(), mgr.GetScheme(), dispatcher)