fix(redis): 修复dbmon中一些备份bug TencentBlueKing#7026

lukemakeit · Sep 23, 2024 · 90009e5 · 90009e5
1 parent 73276af
commit 90009e5
Show file tree

Hide file tree

Showing 3 changed files with 11 additions and 5 deletions.
diff --git a/dbm-services/redis/db-tools/dbmon/pkg/redisbinlogbackup/job.go b/dbm-services/redis/db-tools/dbmon/pkg/redisbinlogbackup/job.go
@@ -233,10 +233,13 @@ func (job *Job) CheckOldBinlogBackupStatus(port int) {
 			// taskStatus==4,上传成功;
 			// taskStatus<4,上传中;
 			if taskStatus > 4 {
-				if row.Status != consts.BackupStatusToBakSystemFailed { // 失败状态不重复上报
+				if row.Status != consts.BackupStatusToBakSystemFailed &&
+					time.Now().Local().Sub(row.StartTime) > 1*time.Hour {
+					// 如果上传失败,且距离文件生成时间超过1小时,则上报
 					row.Status = consts.BackupStatusToBakSystemFailed
-					row.Message = fmt.Sprintf("上传失败,err:%s", statusMsg)
+					row.Message = fmt.Sprintf("上传失败,statusCode:%d,err:%s", taskStatus, statusMsg)
 					row.BackupRecordReport(job.Reporter)
+					mylog.Logger.Error(fmt.Sprintf("%s %s", row.BackupFile, row.Message))
 				}
 			} else if taskStatus < 4 {
 				// 上传中,下次继续探测

diff --git a/dbm-services/redis/db-tools/dbmon/pkg/redisbinlogbackup/task.go b/dbm-services/redis/db-tools/dbmon/pkg/redisbinlogbackup/task.go
@@ -391,7 +391,7 @@ func (task *Task) mvBinlogToBackupDir() {
 	filename := filepath.Base(task.BackupFile)
 	var mvCmd string
 	var targetName, targetFullPath string
-	if strings.Contains(filename, strconv.Itoa(task.ServerPort)) {
+	if strings.Contains(filename, "-"+strconv.Itoa(task.ServerPort)+"-") {
 		// binlog-30012-0007515-20221110084710.log => binlog-1.1.1.1-30012-0007515-20221110084710.log
 		targetName = strings.Replace(filename, "binlog-", "binlog-"+task.ServerIP+"-", -1)
 		targetFullPath = filepath.Join(task.BackupDir, targetName)

diff --git a/dbm-services/redis/db-tools/dbmon/pkg/redisfullbackup/backupjob.go b/dbm-services/redis/db-tools/dbmon/pkg/redisfullbackup/backupjob.go
@@ -231,10 +231,13 @@ func (job *Job) CheckOldFullbackupStatus(port int) {
 			// taskStatus==4,上传成功;
 			// taskStatus<4,上传中;
 			if taskStatus > 4 {
-				if row.Status != consts.BackupStatusToBakSystemFailed { // 失败状态不重复上报
+				if row.Status != consts.BackupStatusToBakSystemFailed &&
+					time.Now().Local().Sub(row.StartTime) > 1*time.Hour {
+					// 如果上传失败,且距离文件生成时间超过1小时,则上报
 					row.Status = consts.BackupStatusToBakSystemFailed
-					row.Message = fmt.Sprintf("上传失败,err:%s", statusMsg)
+					row.Message = fmt.Sprintf("上传失败,statusCode:%d,err:%s", taskStatus, statusMsg)
 					row.BackupRecordReport(job.Reporter)
+					mylog.Logger.Error(fmt.Sprintf("%s %s", row.BackupFile, row.Message))
 				}
 			} else if taskStatus < 4 {
 				// 上传中,下次继续探测