aiforge/routers/api/v1/repo/modelarts.go

// Copyright 2016 The Gogs Authors. All rights reserved.
// Copyright 2018 The Gitea Authors. All rights reserved.
// Use of this source code is governed by a MIT-style
// license that can be found in the LICENSE file.

package repo

import (
	"net/http"
	"os"
	"strconv"
	"strings"

	"code.gitea.io/gitea/models"
	"code.gitea.io/gitea/modules/context"
	"code.gitea.io/gitea/modules/log"
	"code.gitea.io/gitea/modules/modelarts"
	"code.gitea.io/gitea/modules/setting"
	"code.gitea.io/gitea/modules/storage"
)

func GetModelArtsNotebook(ctx *context.APIContext) {
	var (
		err error
	)

	jobID := ctx.Params(":jobid")
	repoID := ctx.Repo.Repository.ID
	job, err := models.GetRepoCloudBrainByJobID(repoID, jobID)
	if err != nil {
		ctx.NotFound(err)
		return
	}
	result, err := modelarts.GetJob(jobID)
	if err != nil {
		ctx.NotFound(err)
		return
	}

	job.Status = result.Status
	err = models.UpdateJob(job)
	if err != nil {
		log.Error("UpdateJob failed:", err)
	}

	ctx.JSON(http.StatusOK, map[string]interface{}{
		"JobID":     jobID,
		"JobStatus": result.Status,
	})

}

func GetModelArtsTrainJob(ctx *context.APIContext) {
	var (
		err error
	)

	jobID := ctx.Params(":jobid")
	repoID := ctx.Repo.Repository.ID
	job, err := models.GetRepoCloudBrainByJobID(repoID, jobID)
	if err != nil {
		ctx.NotFound(err)
		return
	}
	result, err := modelarts.GetTrainJob(jobID, strconv.FormatInt(job.VersionID, 10))
	if err != nil {
		ctx.NotFound(err)
		return
	}

	job.Status = modelarts.TransTrainJobStatus(result.IntStatus)
	job.Duration = result.Duration
	job.TrainJobDuration = result.TrainJobDuration
	err = models.UpdateJob(job)
	if err != nil {
		log.Error("UpdateJob failed:", err)
	}

	ctx.JSON(http.StatusOK, map[string]interface{}{
		"JobID":       jobID,
		"JobStatus":   job.Status,
		"JobDuration": job.Duration,
	})

}

func GetModelArtsTrainJobVersion(ctx *context.APIContext) {
	var (
		err error
	)

	jobID := ctx.Params(":jobid")
	versionName := ctx.Query("version_name")
	job, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
	if err != nil {
		ctx.NotFound(err)
		return
	}
	result, err := modelarts.GetTrainJob(jobID, strconv.FormatInt(job.VersionID, 10))
	if err != nil {
		ctx.NotFound(err)
		return
	}

	job.Status = modelarts.TransTrainJobStatus(result.IntStatus)
	job.Duration = result.Duration
	job.TrainJobDuration = result.TrainJobDuration

	if result.Duration != 0 {
		job.TrainJobDuration = addZero(result.Duration/3600000) + ":" + addZero(result.Duration%3600000/60000) + ":" + addZero(result.Duration%60000/1000)

	} else {
		job.TrainJobDuration = "00:00:00"
	}

	err = models.UpdateTrainJobVersion(job)
	if err != nil {
		log.Error("UpdateJob failed:", err)
	}

	ctx.JSON(http.StatusOK, map[string]interface{}{
		"JobID":       jobID,
		"JobStatus":   job.Status,
		"JobDuration": job.TrainJobDuration,
	})

}

func addZero(t int64) (m string) {
	if t < 10 {
		m = "0" + strconv.FormatInt(t, 10)
		return m
	} else {
		return strconv.FormatInt(t, 10)
	}
}

func TrainJobGetLog(ctx *context.APIContext) {
	var (
		err error
	)

	var jobID = ctx.Params(":jobid")
	var versionName = ctx.Query("version_name")
	// var logFileName = ctx.Query("file_name")
	var baseLine = ctx.Query("base_line")
	var order = ctx.Query("order")
	var lines = ctx.Query("lines")
	lines_int, err := strconv.Atoi(lines)
	if err != nil {
		log.Error("change lines(%d) string to int failed", lines_int)
	}

	if order != modelarts.OrderDesc && order != modelarts.OrderAsc {
		log.Error("order(%s) check failed", order)
		ctx.JSON(http.StatusBadRequest, map[string]interface{}{
			"err_msg": "order check failed",
		})
		return
	}

	resultLogFile, result, err := trainJobGetLogContent(jobID, versionName, baseLine, order, lines_int)
	if err != nil {
		log.Error("trainJobGetLog(%s) failed:%v", jobID, err.Error())
		// ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
		return
	}

	ctx.Data["log_file_name"] = resultLogFile.LogFileList[0]

	ctx.JSON(http.StatusOK, map[string]interface{}{
		"JobID":       jobID,
		"LogFileName": resultLogFile.LogFileList[0],
		"StartLine":   result.StartLine,
		"EndLine":     result.EndLine,
		"Content":     result.Content,
		"Lines":       result.Lines,
	})
}

func trainJobGetLogContent(jobID string, versionName string, baseLine string, order string, lines int) (*models.GetTrainJobLogFileNamesResult, *models.GetTrainJobLogResult, error) {
	task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
	if err != nil {
		log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
		return nil, nil, err
	}

	resultLogFile, err := modelarts.GetTrainJobLogFileNames(jobID, strconv.FormatInt(task.VersionID, 10))
	if err != nil {
		log.Error("GetTrainJobLogFileNames(%s) failed:%v", jobID, err.Error())
		return nil, nil, err
	}

	result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), baseLine, resultLogFile.LogFileList[0], order, lines)
	if err != nil {
		log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
		return nil, nil, err
	}

	return resultLogFile, result, err
}

func DelTrainJobVersion(ctx *context.APIContext) {
	var (
		err error
	)

	var jobID = ctx.Params(":jobid")
	var versionName = ctx.Query("version_name")
	task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
	if err != nil {
		log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
		ctx.NotFound(err)
		return
	}

	//删除modelarts上的记录
	_, err = modelarts.DelTrainJobVersion(jobID, strconv.FormatInt(task.VersionID, 10))
	if err != nil {
		log.Error("DelTrainJobVersion(%s) failed:%v", task.JobName, err.Error())
		ctx.NotFound(err)
		return
	}

	//删除数据库记录
	err = models.DeleteJob(task)
	if err != nil {
		ctx.ServerError("DeleteJob failed", err)
		ctx.NotFound(err)
		return
	}

	//获取删除后的版本数量
	repo := ctx.Repo.Repository
	VersionTaskList, VersionListCount, err := models.CloudbrainsVersionList(&models.CloudbrainsOptions{
		RepoID:  repo.ID,
		Type:    models.TypeCloudBrainTwo,
		JobType: string(models.JobTypeTrain),
		JobID:   jobID,
	})
	if err != nil {
		ctx.ServerError("get VersionListCount failed", err)
		return
	}
	if VersionListCount > 0 {
		// 判断当前删掉的任务是否是最新版本，若是，将排序后的TotalVersionCount置为删掉的最新版本的TotalVersionCount，若不是，按时间排序后的版本列表的第一个版本设置为最新版本，TotalVersionCount不变
		if task.IsLatestVersion == modelarts.IsLatestVersion {
			err = models.SetVersionCountAndLatestVersion(jobID, VersionTaskList[0].Cloudbrain.VersionName, VersionListCount, modelarts.IsLatestVersion, task.TotalVersionCount)
			if err != nil {
				ctx.ServerError("UpdateJobVersionCount failed", err)
				return
			}
		} else {
			err = models.SetVersionCountAndLatestVersion(jobID, VersionTaskList[0].VersionName, VersionListCount, modelarts.IsLatestVersion, VersionTaskList[0].Cloudbrain.TotalVersionCount)
			if err != nil {
				ctx.ServerError("UpdateJobVersionCount failed", err)
				return
			}
		}
	} else { //已删除该任务下的所有版本
		deleteJobStorage(task.JobName)
	}

	ctx.JSON(http.StatusOK, map[string]interface{}{
		"JobID":            jobID,
		"VersionName":      versionName,
		"StatusOK":         0,
		"VersionListCount": VersionListCount,
	})
}

func StopTrainJobVersion(ctx *context.APIContext) {
	var (
		err error
	)
	var jobID = ctx.Params(":jobid")
	var versionName = ctx.Query("version_name")
	task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
	if err != nil {
		log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
		return
	}

	_, err = modelarts.StopTrainJob(jobID, strconv.FormatInt(task.VersionID, 10))
	if err != nil {
		log.Error("StopTrainJob(%s) failed:%v", task.JobName, err.Error())
		return
	}

	ctx.JSON(http.StatusOK, map[string]interface{}{
		"JobID":       jobID,
		"VersionName": versionName,
		"StatusOK":    0,
	})
}

func ModelList(ctx *context.APIContext) {
	var (
		err error
	)

	var jobID = ctx.Params(":jobid")
	var versionName = ctx.Query("version_name")
	parentDir := ctx.Query("parentDir")
	dirArray := strings.Split(parentDir, "/")
	task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
	if err != nil {
		log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
		return
	}
	models, err := storage.GetObsListObject(task.JobName, parentDir, versionName)
	if err != nil {
		log.Info("get TrainJobListModel failed:", err)
		ctx.ServerError("GetObsListObject:", err)
		return
	}

	ctx.JSON(http.StatusOK, map[string]interface{}{
		"JobID":            jobID,
		"VersionName":      versionName,
		"StatusOK":         0,
		"Path":             dirArray,
		"Dirs":             models,
		"task":             task,
		"PageIsCloudBrain": true,
	})
}

func deleteJobStorage(jobName string) error {
	//delete local
	localJobPath := setting.JobPath + jobName
	err := os.RemoveAll(localJobPath)
	if err != nil {
		log.Error("RemoveAll(%s) failed:%v", localJobPath, err)
	}

	//delete oss
	dirPath := setting.CodePathPrefix + jobName + "/"
	err = storage.ObsRemoveObject(setting.Bucket, dirPath)
	if err != nil {
		log.Error("ObsRemoveObject(%s) failed:%v", localJobPath, err)
	}

	return nil
}

func GetModelArtsInferenceJob(ctx *context.APIContext) {
	var (
		err error
	)

	jobID := ctx.Params(":jobid")
	job, err := models.GetCloudbrainByJobID(jobID)
	if err != nil {
		ctx.NotFound(err)
		return
	}
	result, err := modelarts.GetTrainJob(jobID, strconv.FormatInt(job.VersionID, 10))
	if err != nil {
		ctx.NotFound(err)
		return
	}

	job.Status = modelarts.TransTrainJobStatus(result.IntStatus)
	job.Duration = result.Duration
	job.TrainJobDuration = result.TrainJobDuration

	if result.Duration != 0 {
		job.TrainJobDuration = addZero(result.Duration/3600000) + ":" + addZero(result.Duration%3600000/60000) + ":" + addZero(result.Duration%60000/1000)

	} else {
		job.TrainJobDuration = "00:00:00"
	}

	err = models.UpdateInferenceJob(job)
	if err != nil {
		log.Error("UpdateJob failed:", err)
	}

	ctx.JSON(http.StatusOK, map[string]interface{}{
		"JobID":       jobID,
		"JobStatus":   job.Status,
		"JobDuration": job.TrainJobDuration,
	})

}