train-job log

models/cloudbrain.go

@@ -649,6 +649,23 @@ type GetTrainJobResult struct {
 	NasShareAddr		string			`json:"nas_share_addr"`
 }
 
+type GetTrainJobLogResult struct {
+	ErrorCode    		string                 	`json:"error_code"`
+	ErrorMsg    		string                 	`json:"error_msg"`
+	IsSuccess     		bool                 	`json:"is_success"`
+	Content     		string                 	`json:"content"`
+	Lines     			int                 	`json:"lines"`
+	StartLine    		string                 	`json:"start_line"`
+	EndLine    			string                 	`json:"end_line"`
+}
+
+type GetTrainJobLogFileNamesResult struct {
+	ErrorCode    		string                 	`json:"error_code"`
+	ErrorMsg    		string                 	`json:"error_msg"`
+	IsSuccess     		bool                 	`json:"is_success"`
+	LogFileList     	[]string                `json:"log_file_list"`
+}
+
 func Cloudbrains(opts *CloudbrainsOptions) ([]*Cloudbrain, int64, error) {
 	sess := x.NewSession()
 	defer sess.Close()

modules/modelarts/modelarts.go

@@ -43,6 +43,8 @@ const (
 	OutputPath = "/output/"
 	LogPath = "/log/"
 	JobPath = "/job/"
+	OrderDesc = "desc"
+	OrderAsc = "asc"
 )
 
 type GenerateTrainJobReq struct {

modules/modelarts/resty.go

@@ -1,15 +1,15 @@
 package modelarts
 
 import (
+	"code.gitea.io/gitea/models"
 	"code.gitea.io/gitea/modules/log"
+	"code.gitea.io/gitea/modules/setting"
 	"crypto/tls"
 	"encoding/json"
 	"fmt"
-	"net/http"
-
-	"code.gitea.io/gitea/models"
-	"code.gitea.io/gitea/modules/setting"
 	"github.com/go-resty/resty/v2"
+	"net/http"
+	"strconv"
 )
 
 var (
@@ -509,3 +509,91 @@ sendjob:
 
 	return &result, nil
 }
+
+func GetTrainJobLog(jobID, versionID, baseLine, logFile, order string, lines int) (*models.GetTrainJobLogResult, error) {
+	checkSetting()
+	client := getRestyClient()
+	var result models.GetTrainJobLogResult
+
+	retry := 0
+
+sendjob:
+	res, err := client.R().
+		SetQueryParams(map[string]string{
+			"base_line": 	baseLine,
+			"lines": 		strconv.Itoa(lines),
+			"log_file":		logFile,
+			"order": 		order,
+		}).
+		SetAuthToken(TOKEN).
+		SetResult(&result).
+		Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/aom-log")
+
+	if err != nil {
+		return nil, fmt.Errorf("resty GetTrainJobLog: %v", err)
+	}
+
+	if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
+		retry++
+		_ = getToken()
+		goto sendjob
+	}
+
+	if res.StatusCode() != http.StatusOK {
+		var temp models.ErrorResult
+		if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
+			log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
+			return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
+		}
+		log.Error("GetTrainJobLog failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
+		return &result, fmt.Errorf("GetTrainJobLog failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
+	}
+
+	if !result.IsSuccess {
+		log.Error("GetTrainJobLog(%s) failed", jobID)
+		return &result, fmt.Errorf("获取作业日志失败:%s", result.ErrorMsg)
+	}
+
+	return &result, nil
+}
+
+func GetTrainJobLogFileNames(jobID, versionID string) (*models.GetTrainJobLogFileNamesResult, error) {
+	checkSetting()
+	client := getRestyClient()
+	var result models.GetTrainJobLogFileNamesResult
+
+	retry := 0
+
+sendjob:
+	res, err := client.R().
+		SetAuthToken(TOKEN).
+		SetResult(&result).
+		Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/log/file-names")
+
+	if err != nil {
+		return nil, fmt.Errorf("resty GetTrainJobLog: %v", err)
+	}
+
+	if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
+		retry++
+		_ = getToken()
+		goto sendjob
+	}
+
+	if res.StatusCode() != http.StatusOK {
+		var temp models.ErrorResult
+		if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
+			log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
+			return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
+		}
+		log.Error("GetTrainJobLog failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
+		return &result, fmt.Errorf("GetTrainJobLog failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
+	}
+
+	if !result.IsSuccess {
+		log.Error("GetTrainJobLog(%s) failed", jobID)
+		return &result, fmt.Errorf("获取作业日志失败:%s", result.ErrorMsg)
+	}
+
+	return &result, nil
+}

routers/repo/modelarts.go

@@ -554,3 +554,32 @@ func TrainJobShow(ctx *context.Context) {
 	ctx.Data["result"] = result
 	ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
 }
+
+func TrainJobGetLog(ctx *context.Context) {
+	ctx.Data["PageIsCloudBrain"] = true
+
+	var jobID = ctx.Params(":jobid")
+	task, err := models.GetCloudbrainByJobID(jobID)
+	if err != nil {
+		log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
+		ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
+		return
+	}
+
+	resultLogFile, err := modelarts.GetTrainJobLogFileNames(jobID, strconv.FormatInt(task.VersionID, 10))
+	if err != nil {
+		log.Error("GetTrainJobLogFileNames(%s) failed:%v", jobID, err.Error())
+		ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
+		return
+	}
+
+	result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), "", resultLogFile.LogFileList[0], modelarts.OrderDesc, 20)
+	if err != nil {
+		log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
+		ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
+		return
+	}
+
+	ctx.Data["log"] = result
+	ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
+}

routers/routes/routes.go

@@ -936,6 +936,7 @@ func RegisterRoutes(m *macaron.Macaron) {
 					m.Get("/debug", reqRepoCloudBrainReader, repo.NotebookDebug)
 					m.Post("/stop", reqRepoCloudBrainWriter, repo.NotebookStop)
 					m.Post("/del", reqRepoCloudBrainWriter, repo.NotebookDel)
+					m.Get("/log", reqRepoCloudBrainReader, repo.TrainJobGetLog)
 				})
 				m.Get("/create", reqRepoCloudBrainWriter, repo.TrainJobNew)
 				m.Post("/create", reqRepoCloudBrainWriter, bindIgnErr(auth.CreateModelArtsTrainJobForm{}), repo.TrainJobCreate)