feat(scheduler): make node lock timeout configurable

cmd/scheduler/main.go

@@ -20,6 +20,7 @@ import (
 	"fmt"
 	"net/http"
 	"net/http/pprof"
+	"time"
 
 	"github.com/julienschmidt/httprouter"
 	"github.com/spf13/cobra"
@@ -32,6 +33,7 @@ import (
 	"github.com/Project-HAMi/HAMi/pkg/util"
 	"github.com/Project-HAMi/HAMi/pkg/util/client"
 	"github.com/Project-HAMi/HAMi/pkg/util/flag"
+	"github.com/Project-HAMi/HAMi/pkg/util/nodelock"
 	"github.com/Project-HAMi/HAMi/pkg/version"
 )
 
@@ -72,6 +74,7 @@ func init() {
 	rootCmd.Flags().IntVar(&config.Burst, "kube-burst", client.DefaultBurst, "Burst to use while talking with kube-apiserver.")
 	rootCmd.Flags().IntVar(&config.Timeout, "kube-timeout", client.DefaultTimeout, "Timeout to use while talking with kube-apiserver.")
 	rootCmd.Flags().BoolVar(&enableProfiling, "profiling", false, "Enable pprof profiling via HTTP server")
+	rootCmd.Flags().DurationVar(&config.NodeLockTimeout, "node-lock-timeout", time.Minute*5, "timeout for node locks")
 
 	rootCmd.PersistentFlags().AddGoFlagSet(device.GlobalFlagSet())
 	rootCmd.AddCommand(version.VersionCmd)
@@ -99,6 +102,9 @@ func injectProfilingRoute(router *httprouter.Router) {
 }
 
 func start() error {
+	// Initialize node lock timeout from config
+	nodelock.NodeLockTimeout = config.NodeLockTimeout
+	klog.InfoS("Set node lock timeout", "timeout", nodelock.NodeLockTimeout)
 	client.InitGlobalClient(
 		client.WithBurst(config.Burst),
 		client.WithQPS(config.QPS),

pkg/scheduler/config/config.go

@@ -16,7 +16,11 @@ limitations under the License.
 
 package config
 
-import "github.com/Project-HAMi/HAMi/pkg/util"
+import (
+	"time"
+
+	"github.com/Project-HAMi/HAMi/pkg/util"
+)
 
 var (
 	QPS                float32
@@ -37,4 +41,7 @@ var (
 
 	// NodeLabelSelector is scheduler filter node by node label.
 	NodeLabelSelector map[string]string
+
+	// NodeLockTimeout is the timeout for node locks.
+	NodeLockTimeout time.Duration
 )

pkg/util/nodelock/nodelock.go

@@ -36,7 +36,11 @@ const (
 	NodeLockSep  = ","
 )
 
-var lock sync.Mutex
+var (
+	lock sync.Mutex
+	// NodeLockTimeout is the global timeout for node locks.
+	NodeLockTimeout time.Duration = time.Minute * 5
+)
 
 func SetNodeLock(nodeName string, lockname string, pods *corev1.Pod) error {
 	lock.Lock()
@@ -122,16 +126,16 @@ func LockNode(nodeName string, lockname string, pods *corev1.Pod) error {
 	if err != nil {
 		return err
 	}
-	if time.Since(lockTime) > time.Minute*5 {
-		klog.InfoS("Node lock expired", "node", nodeName, "lockTime", lockTime)
+	if time.Since(lockTime) > NodeLockTimeout {
+		klog.InfoS("Node lock expired", "node", nodeName, "lockTime", lockTime, "timeout", NodeLockTimeout)
 		err = ReleaseNodeLock(nodeName, lockname, pods, true)
 		if err != nil {
 			klog.ErrorS(err, "Failed to release node lock", "node", nodeName)
 			return err
 		}
 		return SetNodeLock(nodeName, lockname, pods)
 	}
-	return fmt.Errorf("node %s has been locked within 5 minutes", nodeName)
+	return fmt.Errorf("node %s has been locked within %v", nodeName, NodeLockTimeout)
 }
 
 func ParseNodeLock(value string) (lockTime time.Time, ns, name string, err error) {

pkg/util/nodelock/nodelock_test.go

@@ -18,7 +18,9 @@ package nodelock
 
 import (
 	"context"
+	"strings"
 	"testing"
+	"time"
 
 	corev1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
@@ -133,6 +135,53 @@ func Test_LockNode(t *testing.T) {
 	}
 }
 
+// TestLockNodeWithTimeout tests the specific case where line 130 is executed
+func TestLockNodeWithTimeout(t *testing.T) {
+	client.KubeClient = fake.NewSimpleClientset()
+
+	// Set a custom timeout for testing
+	originalTimeout := NodeLockTimeout
+	NodeLockTimeout = time.Minute * 2
+	defer func() {
+		NodeLockTimeout = originalTimeout
+	}()
+
+	nodeName := "test-node-timeout"
+
+	// Create a node with a fresh lock (should not be expired)
+	freshLockTime := time.Now().Format(time.RFC3339)
+	lockValue := freshLockTime + NodeLockSep + "test-ns" + NodeLockSep + "test-pod"
+
+	client.KubeClient.CoreV1().Nodes().Create(context.TODO(), &corev1.Node{
+		ObjectMeta: metav1.ObjectMeta{
+			Name: nodeName,
+			Annotations: map[string]string{
+				NodeLockKey: lockValue,
+			},
+		},
+	}, metav1.CreateOptions{})
+
+	pod := &corev1.Pod{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      "new-pod",
+			Namespace: "new-ns",
+		},
+	}
+
+	// Try to lock the node again - this should trigger line 130
+	err := LockNode(nodeName, "", pod)
+
+	// Verify the error contains the NodeLockTimeout value
+	if err == nil {
+		t.Fatal("Expected error but got nil")
+	}
+
+	expectedError := "has been locked within 2m0s"
+	if !strings.Contains(err.Error(), expectedError) {
+		t.Errorf("Expected error to contain '%s', but got: %v", expectedError, err)
+	}
+}
+
 func TestReleaseNodeLock(t *testing.T) {
 	client.KubeClient = fake.NewSimpleClientset()
 	type args struct {