Merge pull request 'fix-2251' (#3344) from fix-2251 into V20221214

Reviewed-on: https://openi.pcl.ac.cn/OpenI/aiforge/pulls/3344
Reviewed-by: zouap <zouap@pcl.ac.cn>

models/cloudbrain.go

@@ -204,6 +204,7 @@ type Cloudbrain struct {
 	BenchmarkTypeRankLink string      `xorm:"-"`
 	StartTime             timeutil.TimeStamp
 	EndTime               timeutil.TimeStamp
+	Cleared               bool   `xorm:"DEFAULT false"`
 	Spec                  *Specification `xorm:"-"`
 }
 
@@ -1905,6 +1906,12 @@ func GetCloudbrainByID(id string) (*Cloudbrain, error) {
 	return getRepoCloudBrain(cb)
 }
 
+func IsCloudbrainExistByJobName(jobName string)(bool,error){
+	return x.Unscoped().Exist(&Cloudbrain{
+		JobName: jobName,
+	})
+}
+
 func GetCloudbrainByIDWithDeleted(id string) (*Cloudbrain, error) {
 	idInt64, _ := strconv.ParseInt(id, 10, 64)
 	cb := &Cloudbrain{ID: idInt64}
@@ -2050,6 +2057,83 @@ func GetCloudBrainUnStoppedJob() ([]*Cloudbrain, error) {
 		Find(&cloudbrains)
 }
 
+func GetCloudBrainOneStoppedNotDebugJobDaysAgo(days int, limit int) ([]*Cloudbrain, error) {
+	cloudbrains := make([]*Cloudbrain, 0, 10)
+	endTimeBefore := time.Now().Unix() - int64(days)*24*3600
+	missEndTimeBefore := endTimeBefore - 24*3600
+	return cloudbrains, x.Unscoped().Cols("id,job_name,job_id").
+		In("status",
+			JobStopped, JobSucceeded, JobFailed, ModelArtsCreateFailed, ModelArtsStartFailed, ModelArtsUnavailable, ModelArtsResizFailed, ModelArtsDeleted,
+			ModelArtsStopped, ModelArtsTrainJobCanceled, ModelArtsTrainJobCheckFailed, ModelArtsTrainJobCompleted, ModelArtsTrainJobDeleteFailed, ModelArtsTrainJobDeployServiceFailed,
+			ModelArtsTrainJobFailed, ModelArtsTrainJobImageFailed, ModelArtsTrainJobKilled, ModelArtsTrainJobLost, ModelArtsTrainJobSubmitFailed, ModelArtsTrainJobSubmitModelFailed).
+		Where("(((end_time is null or end_time=0) and updated_unix<? and updated_unix != 0 ) or (end_time<? and end_time != 0)) and cleared=false and type=0 and job_type != 'DEBUG'", missEndTimeBefore, endTimeBefore).
+		Limit(limit).
+		Find(&cloudbrains)
+}
+/**
+  本方法考虑了再次调试的情况，多次调试取最后一次的任务的结束时间
+ */
+func GetCloudBrainOneStoppedDebugJobDaysAgo(days int, limit int) ([]*Cloudbrain, error) {
+	cloudbrains := make([]*Cloudbrain, 0, 10)
+	endTimeBefore := time.Now().Unix() - int64(days)*24*3600
+	missEndTimeBefore := endTimeBefore - 24*3600
+    sql:=`SELECT id,job_name,job_id from (SELECT DISTINCT ON (job_name)
+        id, job_name, job_id,status,end_time,updated_unix,cleared
+        FROM   cloudbrain
+        where type=0 and job_type='DEBUG'
+        ORDER  BY job_name, updated_unix DESC) a
+        where status in ('STOPPED','SUCCEEDED','FAILED') and (((end_time is null or end_time=0) and updated_unix<? and updated_unix != 0 ) or (end_time<? and end_time != 0)) and cleared=false`
+
+	return cloudbrains, x.Unscoped().SQL(sql,missEndTimeBefore, endTimeBefore).Limit(limit).Find(&cloudbrains)
+
+}
+
+
+func UpdateCloudBrainRecordsCleared(ids []int64) error {
+	pageSize := 150
+	n := len(ids) / pageSize
+
+	var err error
+
+	for i := 1; i <= n+1; i++ {
+		tempIds := getPageIds(ids, i, pageSize)
+		if len(tempIds) > 0 {
+			idsIn := ""
+			for i, id := range tempIds {
+				if i == 0 {
+					idsIn += strconv.FormatInt(id, 10)
+				} else {
+					idsIn += "," + strconv.FormatInt(id, 10)
+				}
+			}
+
+			_, errTemp := x.Unscoped().Exec("update cloudbrain set cleared=true where id in (" + idsIn + ")")
+			if errTemp != nil {
+				err = errTemp
+			}
+
+		}
+
+	}
+	return err
+
+}
+
+func getPageIds(ids []int64, page int, pagesize int) []int64 {
+	begin := (page - 1) * pagesize
+	end := (page) * pagesize
+
+	if begin > len(ids)-1 {
+		return []int64{}
+	}
+	if end > len(ids)-1 {
+		return ids[begin:]
+	} else {
+		return ids[begin:end]
+	}
+
+}
+
 func GetStoppedJobWithNoDurationJob() ([]*Cloudbrain, error) {
 	cloudbrains := make([]*Cloudbrain, 0)
 	return cloudbrains, x.

modules/cron/tasks_basic.go

@@ -5,10 +5,13 @@
 package cron
 
 import (
-	"code.gitea.io/gitea/modules/urfs_client/urchin"
+	"code.gitea.io/gitea/modules/setting"
 	"context"
 	"time"
 
+	"code.gitea.io/gitea/modules/urfs_client/urchin"
+	cloudbrainService "code.gitea.io/gitea/services/cloudbrain"
+
 	"code.gitea.io/gitea/modules/modelarts"
 	"code.gitea.io/gitea/services/cloudbrain/resource"
 	"code.gitea.io/gitea/services/reward"
@@ -190,6 +193,17 @@ func registerHandleRepoAndUserStatistic() {
 	})
 }
 
+func registerHandleClearCloudbrainResult() {
+	RegisterTaskFatal("handle_cloudbrain_one_result_clear", &BaseConfig{
+		Enabled:    true,
+		RunAtStart: setting.ClearStrategy.RunAtStart,
+		Schedule:   setting.ClearStrategy.Cron,
+	}, func(ctx context.Context, _ *models.User, _ Config) error {
+		cloudbrainService.ClearCloudbrainResultSpace()
+		return nil
+	})
+}
+
 func registerHandleSummaryStatistic() {
 	RegisterTaskFatal("handle_summary_statistic", &BaseConfig{
 		Enabled:    true,
@@ -306,6 +320,7 @@ func initBasicTasks() {
 
 	registerHandleRepoAndUserStatistic()
 	registerHandleSummaryStatistic()
+	registerHandleClearCloudbrainResult()
 
 	registerSyncCloudbrainStatus()
 	registerHandleOrgStatistic()
@@ -317,6 +332,6 @@ func initBasicTasks() {
 
 	registerHandleModelSafetyTask()
 
-	registerHandleScheduleRecord()	
+	registerHandleScheduleRecord()
 	registerHandleCloudbrainDurationStatistic()
 }

modules/setting/setting.go

@@ -519,6 +519,7 @@ var (
 	CullIdleTimeout        string
 	CullInterval           string
 
+
 	//benchmark config
 	IsBenchmarkEnabled     bool
 	BenchmarkOwner         string
@@ -613,6 +614,16 @@ var (
 		UsageRateBeginTime      string
 	}{}
 
+	ClearStrategy= struct {
+		Enabled bool
+		ResultSaveDays         int
+		BatchSize int
+		DebugJobSize int
+		TrashSaveDays int
+		Cron string
+		RunAtStart bool
+	}{}
+
 	C2NetInfos                 *C2NetSqInfos
 	CenterInfos                *AiCenterInfos
 	C2NetMapInfo               map[string]*C2NetSequenceInfo
@@ -1619,6 +1630,7 @@ func NewContext() {
 	getModelConvertConfig()
 	getModelSafetyConfig()
 	getModelAppConfig()
+	getClearStrategy()
 }
 
 func getModelSafetyConfig() {
@@ -1679,6 +1691,18 @@ func getModelartsCDConfig() {
 	getNotebookFlavorInfos()
 }
 
+func getClearStrategy(){
+
+	sec := Cfg.Section("clear_strategy")
+    ClearStrategy.Enabled=sec.Key("ENABLED").MustBool(false)
+	ClearStrategy.ResultSaveDays=sec.Key("RESULT_SAVE_DAYS").MustInt(30)
+	ClearStrategy.BatchSize=sec.Key("BATCH_SIZE").MustInt(500)
+	ClearStrategy.DebugJobSize=sec.Key("DEBUG_BATCH_SIZE").MustInt(100)
+	ClearStrategy.TrashSaveDays=sec.Key("TRASH_SAVE_DAYS").MustInt(90)
+	ClearStrategy.Cron=sec.Key("CRON").MustString("* 0,30 2-8 * * ?")
+	ClearStrategy.RunAtStart=sec.Key("RUN_AT_START").MustBool(false)
+}
+
 func getGrampusConfig() {
 	sec := Cfg.Section("grampus")
 

options/locale/locale_en-US.ini

@@ -3246,6 +3246,7 @@ specification = specification
 select_specification = select specification
 description = description
 wrong_specification=You cannot use this specification, please choose another item.
+result_cleared=The files of the task have been cleared, can not restart any more, please create a new debug task instead.
 resource_use=Resource Occupancy
 
 job_name_rule = Please enter letters, numbers, _ and - up to 64 characters and cannot end with a dash (-).

options/locale/locale_zh-CN.ini

@@ -3266,6 +3266,8 @@ card_duration = 运行卡时
 card_type = 卡类型
 wrong_specification=您目前不能使用这个资源规格，请选择其他资源规格。
 
+result_cleared=本任务的文件已被清理，无法再次调试，请新建调试任务。
+
 job_name_rule = 请输入字母、数字、_和-，最长64个字符，且不能以中划线（-）结尾。
 train_dataset_path_rule = 数据集位置存储在运行参数 <strong style="color:#010101">data_url</strong> 中，预训练模型存放在运行参数 <strong style="color:#010101">ckpt_url</strong> 中，训练输出路径存储在运行参数 <strong style="color:#010101">train_url</strong> 中。
 infer_dataset_path_rule = 数据集位置存储在运行参数 <strong style="color:#010101">data_url</strong> 中，推理输出路径存储在运行参数 <strong style="color:#010101">result_url</strong> 中。

routers/repo/cloudbrain.go

@@ -670,6 +670,13 @@ func CloudBrainRestart(ctx *context.Context) {
 			break
 		}
 
+		if _, err := os.Stat(getOldJobPath(task)); err != nil {
+			log.Error("Can not find job minio path", err)
+			resultCode = "-1"
+			errorMsg = ctx.Tr("cloudbrain.result_cleared")
+			break
+		}
+
 		count, err := cloudbrainTask.GetNotFinalStatusTaskCount(ctx.User.ID, models.TypeCloudBrainOne, string(models.JobTypeDebug))
 		if err != nil {
 			log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])
@@ -704,6 +711,11 @@ func CloudBrainRestart(ctx *context.Context) {
 	})
 }
 
+
+func getOldJobPath(task *models.Cloudbrain) string {
+	return setting.Attachment.Minio.RealPath + setting.Attachment.Minio.Bucket + "/" + setting.CBCodePathPrefix + task.JobName
+}
+
 func CloudBrainBenchMarkShow(ctx *context.Context) {
 	cloudBrainShow(ctx, tplCloudBrainBenchmarkShow, models.JobTypeBenchmark)
 }

services/cloudbrain/clear.go

@@ -0,0 +1,151 @@
+package cloudbrain
+
+import (
+	"io/ioutil"
+	"os"
+	"sort"
+	"time"
+
+	"code.gitea.io/gitea/models"
+	"code.gitea.io/gitea/modules/log"
+	"code.gitea.io/gitea/modules/setting"
+	"code.gitea.io/gitea/modules/storage"
+)
+
+func ClearCloudbrainResultSpace() {
+	log.Info("clear cloudbrain one result space begin.")
+	if !setting.ClearStrategy.Enabled{
+		return
+	}
+
+	tasks, err := models.GetCloudBrainOneStoppedNotDebugJobDaysAgo(setting.ClearStrategy.ResultSaveDays, setting.ClearStrategy.BatchSize)
+	if err != nil {
+		log.Warn("Failed to get cloudbrain, clear result failed.", err)
+		return
+	}
+	debugTasks, err := models.GetCloudBrainOneStoppedDebugJobDaysAgo(setting.ClearStrategy.ResultSaveDays, setting.ClearStrategy.DebugJobSize)
+	if err != nil {
+		log.Warn("Failed to get debug cloudbrain.", err)
+
+	}
+	tasks=append(tasks,debugTasks...)
+
+	if err != nil {
+		log.Warn("Failed to get cloudbrain, clear result failed.", err)
+		return
+	}
+	var ids []int64
+	for _, task := range tasks {
+		err := DeleteCloudbrainOneJobStorage(task.JobName)
+		if err == nil {
+			log.Info("clear job in cloudbrain table:"+task.JobName)
+			ids = append(ids, task.ID)
+		}
+	}
+
+	err = models.UpdateCloudBrainRecordsCleared(ids)
+	if err != nil {
+		log.Warn("Failed to set cloudbrain cleared status", err)
+	}
+	//如果云脑表处理完了，通过遍历minio对象处理历史垃圾数据，如果存在的话
+	if len(tasks) < setting.ClearStrategy.BatchSize+setting.ClearStrategy.DebugJobSize {
+		clearLocalHistoryTrashFile()
+		clearMinioHistoryTrashFile()
+
+	}
+	log.Info("clear cloudbrain one result space end.")
+
+}
+
+func clearMinioHistoryTrashFile() {
+	JobRealPrefix := setting.Attachment.Minio.RealPath + setting.Attachment.Minio.Bucket + "/" + setting.CBCodePathPrefix
+
+	miniofiles, err := ioutil.ReadDir(JobRealPrefix)
+
+	processCount := 0
+	if err != nil {
+		log.Warn("Can not browser minio job path.")
+	} else {
+		SortModTimeAscend(miniofiles)
+		for _, file := range miniofiles {
+
+			if file.Name()!="" && file.ModTime().Before(time.Now().AddDate(0, 0, -setting.ClearStrategy.TrashSaveDays)) {
+
+				has,err:=models.IsCloudbrainExistByJobName(file.Name())
+				if err==nil && !has {
+					dirPath := setting.CBCodePathPrefix + file.Name() + "/"
+					log.Info("clear job in minio trash:" + file.Name())
+					storage.Attachments.DeleteDir(dirPath)
+					processCount++
+				}
+				if processCount == setting.ClearStrategy.BatchSize {
+					break
+				}
+			} else {
+				break
+			}
+
+		}
+
+	}
+}
+
+func clearLocalHistoryTrashFile()  {
+	files, err := ioutil.ReadDir(setting.JobPath)
+	processCount := 0
+	if err != nil {
+		log.Warn("Can not browser local job path.")
+	} else {
+		SortModTimeAscend(files)
+		for _, file := range files {
+			//清理n天前的历史垃圾数据，清理job目录
+			if file.Name()!="" && file.ModTime().Before(time.Now().AddDate(0, 0, -setting.ClearStrategy.TrashSaveDays)) {
+				has,err:=models.IsCloudbrainExistByJobName(file.Name())
+				if err==nil && !has{
+					os.RemoveAll(setting.JobPath + file.Name())
+					log.Info("clear job in local trash:"+file.Name())
+					processCount++
+				}
+				if processCount == setting.ClearStrategy.BatchSize {
+					break
+				}
+			} else {
+				break
+			}
+
+		}
+
+	}
+
+}
+
+func SortModTimeAscend(files []os.FileInfo) {
+	sort.Slice(files, func(i, j int) bool {
+		return files[i].ModTime().Before(files[j].ModTime())
+	})
+}
+
+func DeleteCloudbrainOneJobStorage(jobName string) error {
+
+	if jobName==""{
+		return nil
+	}
+	//delete local
+	localJobPath := setting.JobPath + jobName
+	err := os.RemoveAll(localJobPath)
+	if err != nil {
+		log.Error("RemoveAll(%s) failed:%v", localJobPath, err)
+	}
+
+	dirPath := setting.CBCodePathPrefix + jobName + "/"
+	err1 := storage.Attachments.DeleteDir(dirPath)
+
+	if err1 != nil {
+		log.Error("DeleteDir(%s) failed:%v", localJobPath, err)
+	}
+	if err == nil {
+		err = err1
+	}
+
+	return err
+}