diff --git a/device-plugin b/device-plugin
new file mode 100644
index 0000000000000000000000000000000000000000..9b06f89a134c926de01818455a7072c14a2dd9d9
Binary files /dev/null and b/device-plugin differ
diff --git a/pkg/server/manager.go b/pkg/server/manager.go
index c9ffabcd2d64f59ae78c08221eb46c2cf443b861..e7ab82b2b033ef6187fc66b1970c117d9709e491 100644
--- a/pkg/server/manager.go
+++ b/pkg/server/manager.go
@@ -16,9 +16,12 @@
 package server
 
 import (
+	"bufio"
 	"context"
 	"encoding/json"
 	"fmt"
+	"io"
+	"k8s.io/client-go/kubernetes"
 	"os"
 	"path/filepath"
 	"strconv"
@@ -300,6 +303,7 @@ func (hdm *HwDevManager) ListenDevice(ctx context.Context) {
 	// cache to prevent manually separate npu IDs in cache from been lost
 	hdm.separateNPUIDFromDeviceInfoIntoCache()
 	go hdm.pollFaultCodeCM(ctx)
+	go hdm.getErrCodeFromLog()
 	go hdm.Serve(ctx)
 	initTime := time.Now()
 	for {
@@ -824,6 +828,85 @@ func (hdm *HwDevManager) isSupportGraceTolerance() bool {
 	return true
 }
 
+func readLog(reader *bufio.Reader) {
+	for {
+		// 读取一行日志
+		line, err := reader.ReadString('\n')
+		if err == io.EOF {
+			// 如果到达文件末尾，等待一段时间后重试
+			time.Sleep(1 * time.Second)
+			continue
+		} else if err != nil {
+			hwlog.RunLog.Errorf("read err %v", err)
+		}
+		hwlog.RunLog.Warnf("get: %v", line)
+		// 检查是否包含关键字"Err"
+		if strings.Contains(line, "Err") {
+			// 输出包含关键字"Err"的日志行
+			hwlog.RunLog.Infof("got target: %v", line)
+		}
+	}
+}
+
+func readLog2(clientset kubernetes.Interface, namespace, podName string, recordMap map[string]string) {
+	tailLine := int64(2)
+	// 获取 Pod 的日志
+	req := clientset.CoreV1().Pods(namespace).GetLogs(podName, &v1.PodLogOptions{
+		Follow:    true,
+		TailLines: &tailLine,
+	})
+	readCloser, err := req.Stream(context.Background())
+	if err != nil {
+		panic(err)
+	}
+	defer readCloser.Close()
+
+	// 读取 Pod 的日志并输出
+	buf := make([]byte, 2000)
+	for {
+		n, err := readCloser.Read(buf)
+		if err != nil && err != io.EOF {
+			hwlog.RunLog.Errorf("read panic err: %v", err)
+		}
+		if n > 0 {
+			hwlog.RunLog.Infof("get log: %s", string(buf[:n]))
+		}
+		if err == io.EOF {
+			hwlog.RunLog.Infof("log finished")
+			delete(recordMap, namespace+podName)
+			break
+		}
+		time.Sleep(1 * time.Second)
+	}
+}
+
+func (hdm *HwDevManager) getErrCodeFromLog() {
+	recordMap := make(map[string]string)
+
+	for {
+		podList := hdm.manager.GetKubeClient().GetAllPodListCache()
+		for _, pod := range podList {
+			for k, _ := range pod.Annotations {
+				if _, ok := recordMap[pod.Namespace+pod.Name]; ok {
+					continue
+				}
+
+				if strings.Contains(k, "huawei") {
+					if pod.Status.Phase == v1.PodRunning || pod.Status.Phase == v1.PodPending ||  pod.Status.Phase == v1.PodSucceeded {
+						continue
+					}
+					recordMap[pod.Namespace+pod.Name] = "1"
+					hwlog.RunLog.Warnf("get pod %v, status is: %v", pod.Name, pod.Status.Phase)
+					go readLog2(hdm.manager.GetKubeClient().Clientset, pod.Namespace, pod.Name, recordMap)
+				}
+			}
+
+		}
+
+		time.Sleep(5 * time.Second)
+	}
+}
+
 func (hdm *HwDevManager) pollFaultCodeCM(ctx context.Context) {
 	var resourceVersion = ""
 	var interval = common.PollFaultCodeCMInterval