You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

grampus.go 19 kB

3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602
  1. package repo
  2. import (
  3. "code.gitea.io/gitea/modules/auth"
  4. "code.gitea.io/gitea/modules/git"
  5. "code.gitea.io/gitea/modules/grampus"
  6. "code.gitea.io/gitea/modules/modelarts"
  7. "code.gitea.io/gitea/modules/timeutil"
  8. "code.gitea.io/gitea/modules/util"
  9. "encoding/json"
  10. "errors"
  11. "fmt"
  12. "io/ioutil"
  13. "net/http"
  14. "os"
  15. "path"
  16. "strconv"
  17. "strings"
  18. "time"
  19. "code.gitea.io/gitea/models"
  20. "code.gitea.io/gitea/modules/base"
  21. "code.gitea.io/gitea/modules/cloudbrain"
  22. "code.gitea.io/gitea/modules/context"
  23. "code.gitea.io/gitea/modules/log"
  24. "code.gitea.io/gitea/modules/setting"
  25. )
  26. const (
  27. tplGrampusTrainJobShow base.TplName = "repo/grampus/trainjob/show"
  28. //GPU
  29. tplGrampusTrainJobGPUNew base.TplName = "repo/grampus/trainjob/gpu/new"
  30. //NPU
  31. tplGrampusTrainJobNPUNew base.TplName = "repo/grampus/trainjob/npu/new"
  32. )
  33. func GrampusTrainJobGPUNew(ctx *context.Context) {
  34. err := grampusGpuNewDataPrepare(ctx)
  35. if err != nil {
  36. ctx.ServerError("get new train-job info failed", err)
  37. return
  38. }
  39. ctx.HTML(http.StatusOK, tplGrampusTrainJobGPUNew)
  40. }
  41. func grampusGpuNewDataPrepare(ctx *context.Context) error {
  42. ctx.Data["PageIsCloudBrain"] = true
  43. t := time.Now()
  44. var displayJobName = jobNamePrefixValid(cutString(ctx.User.Name, 5)) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  45. ctx.Data["display_job_name"] = displayJobName
  46. //get valid images
  47. result, err := cloudbrain.GetImages()
  48. if err != nil {
  49. ctx.Data["error"] = err.Error()
  50. log.Error("cloudbrain.GetImages failed:", err.Error(), ctx.Data["MsgID"])
  51. }
  52. for i, payload := range result.Payload.ImageInfo {
  53. if strings.HasPrefix(result.Payload.ImageInfo[i].Place, "192.168") {
  54. result.Payload.ImageInfo[i].PlaceView = payload.Place[strings.Index(payload.Place, "/"):len(payload.Place)]
  55. } else {
  56. result.Payload.ImageInfo[i].PlaceView = payload.Place
  57. }
  58. }
  59. ctx.Data["images"] = result.Payload.ImageInfo
  60. resultPublic, err := cloudbrain.GetPublicImages()
  61. if err != nil {
  62. ctx.Data["error"] = err.Error()
  63. log.Error("cloudbrain.GetPublicImages failed:", err.Error(), ctx.Data["MsgID"])
  64. }
  65. for i, payload := range resultPublic.Payload.ImageInfo {
  66. if strings.HasPrefix(resultPublic.Payload.ImageInfo[i].Place, "192.168") {
  67. resultPublic.Payload.ImageInfo[i].PlaceView = payload.Place[strings.Index(payload.Place, "/"):len(payload.Place)]
  68. } else {
  69. resultPublic.Payload.ImageInfo[i].PlaceView = payload.Place
  70. }
  71. }
  72. ctx.Data["public_images"] = resultPublic.Payload.ImageInfo
  73. //get valid dataset
  74. attachs, err := models.GetAllUserAttachments(ctx.User.ID)
  75. if err != nil {
  76. log.Error("GetAllUserAttachments failed: %v", err, ctx.Data["MsgID"])
  77. return err
  78. }
  79. ctx.Data["attachments"] = attachs
  80. ctx.Data["command"] = cloudbrain.Command
  81. ctx.Data["code_path"] = cloudbrain.CodeMountPath
  82. ctx.Data["dataset_path"] = cloudbrain.DataSetMountPath
  83. ctx.Data["model_path"] = cloudbrain.ModelMountPath
  84. ctx.Data["benchmark_path"] = cloudbrain.BenchMarkMountPath
  85. ctx.Data["is_benchmark_enabled"] = setting.IsBenchmarkEnabled
  86. //get valid resource specs
  87. if categories == nil {
  88. json.Unmarshal([]byte(setting.BenchmarkCategory), &categories)
  89. }
  90. ctx.Data["benchmark_categories"] = categories.Category
  91. ctx.Data["benchmark_types"] = GetBenchmarkTypes(ctx).BenchmarkType
  92. if gpuInfos == nil {
  93. json.Unmarshal([]byte(setting.GpuTypes), &gpuInfos)
  94. }
  95. ctx.Data["gpu_types"] = gpuInfos.GpuInfo
  96. if trainGpuInfos == nil {
  97. json.Unmarshal([]byte(setting.TrainGpuTypes), &trainGpuInfos)
  98. }
  99. ctx.Data["train_gpu_types"] = trainGpuInfos.GpuInfo
  100. if benchmarkGpuInfos == nil {
  101. json.Unmarshal([]byte(setting.BenchmarkGpuTypes), &benchmarkGpuInfos)
  102. }
  103. ctx.Data["benchmark_gpu_types"] = benchmarkGpuInfos.GpuInfo
  104. if benchmarkResourceSpecs == nil {
  105. json.Unmarshal([]byte(setting.BenchmarkResourceSpecs), &benchmarkResourceSpecs)
  106. }
  107. ctx.Data["benchmark_resource_specs"] = benchmarkResourceSpecs.ResourceSpec
  108. if cloudbrain.ResourceSpecs == nil {
  109. json.Unmarshal([]byte(setting.ResourceSpecs), &cloudbrain.ResourceSpecs)
  110. }
  111. ctx.Data["resource_specs"] = cloudbrain.ResourceSpecs.ResourceSpec
  112. if cloudbrain.TrainResourceSpecs == nil {
  113. json.Unmarshal([]byte(setting.TrainResourceSpecs), &cloudbrain.TrainResourceSpecs)
  114. }
  115. ctx.Data["train_resource_specs"] = cloudbrain.TrainResourceSpecs.ResourceSpec
  116. branches, _, err := ctx.Repo.GitRepo.GetBranches(0, 0)
  117. if err != nil {
  118. log.Error("GetBranches error:", err)
  119. }
  120. ctx.Data["branches"] = branches
  121. ctx.Data["branchName"] = ctx.Repo.BranchName
  122. return nil
  123. }
  124. func GrampusTrainJobNPUNew(ctx *context.Context) {
  125. err := grampusTrainJobNpuNewDataPrepare(ctx)
  126. if err != nil {
  127. ctx.ServerError("get new train-job info failed", err)
  128. return
  129. }
  130. ctx.HTML(200, tplGrampusTrainJobNPUNew)
  131. }
  132. func grampusTrainJobNpuNewDataPrepare(ctx *context.Context) error {
  133. ctx.Data["PageIsCloudBrain"] = true
  134. t := time.Now()
  135. var displayJobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  136. ctx.Data["display_job_name"] = displayJobName
  137. //get valid dataset
  138. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  139. if err != nil {
  140. log.Error("GetModelArtsTrainAttachments failed:", err.Error())
  141. } else {
  142. ctx.Data["attachments"] = attachs
  143. }
  144. //get valid engines
  145. images, err := grampus.GetImages(grampus.ProcessorTypeNPU)
  146. if err != nil {
  147. log.Error("GetResourceSpecs failed:", err.Error())
  148. } else {
  149. ctx.Data["engine_versions"] = images.Infos
  150. }
  151. //get valid resource specs
  152. specs, err := grampus.GetResourceSpecs(grampus.ProcessorTypeNPU)
  153. if err != nil {
  154. log.Error("GetResourceSpecs failed:", err.Error())
  155. } else {
  156. ctx.Data["flavor_infos"] = specs.Infos
  157. }
  158. //get branches
  159. branches, _, err := ctx.Repo.GitRepo.GetBranches(0, 0)
  160. if err != nil {
  161. log.Error("GetBranches error:", err.Error())
  162. } else {
  163. ctx.Data["branches"] = branches
  164. }
  165. ctx.Data["branchName"] = ctx.Repo.BranchName
  166. return nil
  167. }
  168. func grampusParamCheckCreateTrainJob(form auth.CreateGrampusTrainJobForm) error {
  169. if !strings.HasSuffix(form.BootFile, ".py") {
  170. log.Error("the boot file(%s) must be a python file", form.BootFile)
  171. return errors.New("启动文件必须是python文件")
  172. }
  173. if form.BranchName == "" {
  174. log.Error("the branch must not be null!", form.BranchName)
  175. return errors.New("代码分支不能为空!")
  176. }
  177. return nil
  178. }
  179. func GrampusTrainJobNpuCreate(ctx *context.Context, form auth.CreateGrampusTrainJobForm) {
  180. VersionOutputPath := modelarts.GetOutputPathByCount(modelarts.TotalVersionCount)
  181. displayJobName := form.DisplayJobName
  182. jobName := util.ConvertDisplayJobNameToJobName(displayJobName)
  183. uuid := form.Attachment
  184. description := form.Description
  185. bootFile := form.BootFile
  186. params := form.Params
  187. repo := ctx.Repo.Repository
  188. codeLocalPath := setting.JobPath + jobName + modelarts.CodePath
  189. codeObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.CodePath
  190. dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/"
  191. branchName := form.BranchName
  192. isLatestVersion := modelarts.IsLatestVersion
  193. flavorName := form.FlavorName
  194. versionCount := modelarts.VersionCount
  195. engineName := form.EngineName
  196. //check count limit
  197. count, err := models.GetGrampusCountByUserID(ctx.User.ID, string(models.JobTypeTrain), models.NPUResource)
  198. if err != nil {
  199. log.Error("GetGrampusCountByUserID failed:%v", err, ctx.Data["MsgID"])
  200. grampusTrainJobNpuNewDataPrepare(ctx)
  201. ctx.RenderWithErr("system error", tplGrampusTrainJobNPUNew, &form)
  202. return
  203. } else {
  204. if count >= 1 {
  205. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  206. grampusTrainJobNpuNewDataPrepare(ctx)
  207. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplGrampusTrainJobNPUNew, &form)
  208. return
  209. }
  210. }
  211. //check param
  212. if err := grampusParamCheckCreateTrainJob(form); err != nil {
  213. log.Error("paramCheckCreateTrainJob failed:(%v)", err)
  214. grampusTrainJobNpuNewDataPrepare(ctx)
  215. ctx.RenderWithErr(err.Error(), tplGrampusTrainJobNPUNew, &form)
  216. return
  217. }
  218. //check whether the task name in the project is duplicated
  219. tasks, err := models.GetCloudbrainsByDisplayJobName(repo.ID, string(models.JobTypeTrain), displayJobName)
  220. if err == nil {
  221. if len(tasks) != 0 {
  222. log.Error("the job name did already exist", ctx.Data["MsgID"])
  223. grampusTrainJobNpuNewDataPrepare(ctx)
  224. ctx.RenderWithErr("the job name did already exist", tplGrampusTrainJobNPUNew, &form)
  225. return
  226. }
  227. } else {
  228. if !models.IsErrJobNotExist(err) {
  229. log.Error("system error, %v", err, ctx.Data["MsgID"])
  230. grampusTrainJobNpuNewDataPrepare(ctx)
  231. ctx.RenderWithErr("system error", tplGrampusTrainJobNPUNew, &form)
  232. return
  233. }
  234. }
  235. //check dataset
  236. attachment, err := models.GetAttachmentByUUID(uuid)
  237. if err != nil {
  238. log.Error("GetAttachmentByUUID failed:", err.Error(), ctx.Data["MsgID"])
  239. grampusTrainJobNpuNewDataPrepare(ctx)
  240. ctx.RenderWithErr("dataset is not exist", tplGrampusTrainJobNPUNew, &form)
  241. return
  242. }
  243. //prepare code and out path
  244. _, err = ioutil.ReadDir(codeLocalPath)
  245. if err == nil {
  246. os.RemoveAll(codeLocalPath)
  247. }
  248. if err := downloadCode(repo, codeLocalPath, branchName); err != nil {
  249. log.Error("downloadCode failed, server timed out: %s (%v)", repo.FullName(), err)
  250. grampusTrainJobNpuNewDataPrepare(ctx)
  251. ctx.RenderWithErr("Create task failed, server timed out", tplGrampusTrainJobNPUNew, &form)
  252. return
  253. }
  254. //todo: upload code (send to file_server todo this work?)
  255. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.OutputPath + VersionOutputPath + "/"); err != nil {
  256. log.Error("Failed to obsMkdir_output: %s (%v)", repo.FullName(), err)
  257. grampusTrainJobNpuNewDataPrepare(ctx)
  258. ctx.RenderWithErr("Failed to obsMkdir_output", tplGrampusTrainJobNPUNew, &form)
  259. return
  260. }
  261. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.LogPath + VersionOutputPath + "/"); err != nil {
  262. log.Error("Failed to obsMkdir_log: %s (%v)", repo.FullName(), err)
  263. grampusTrainJobNpuNewDataPrepare(ctx)
  264. ctx.RenderWithErr("Failed to obsMkdir_log", tplGrampusTrainJobNPUNew, &form)
  265. return
  266. }
  267. if err := uploadCodeToObs(codeLocalPath, jobName, ""); err != nil {
  268. log.Error("Failed to uploadCodeToObs: %s (%v)", repo.FullName(), err)
  269. grampusTrainJobNpuNewDataPrepare(ctx)
  270. ctx.RenderWithErr("Failed to uploadCodeToObs", tplGrampusTrainJobNPUNew, &form)
  271. return
  272. }
  273. //prepare command
  274. //todo: download code, download dataset, unzip dataset, exec code, upload model
  275. command, err := generateCommand(grampus.ProcessorTypeNPU, "obs:/"+codeObsPath, "obs:/"+dataPath, params, "", attachment.Name)
  276. log.Info(command)
  277. var parameters models.Parameters
  278. param := make([]models.Parameter, 0)
  279. if len(params) != 0 {
  280. err := json.Unmarshal([]byte(params), &parameters)
  281. if err != nil {
  282. log.Error("Failed to Unmarshal params: %s (%v)", params, err)
  283. grampusTrainJobNpuNewDataPrepare(ctx)
  284. ctx.RenderWithErr("运行参数错误", tplGrampusTrainJobNPUNew, &form)
  285. return
  286. }
  287. for _, parameter := range parameters.Parameter {
  288. param = append(param, models.Parameter{
  289. Label: parameter.Label,
  290. Value: parameter.Value,
  291. })
  292. }
  293. }
  294. param = append(param, models.Parameter{
  295. Label: modelarts.DeviceTarget,
  296. Value: modelarts.Ascend,
  297. })
  298. gitRepo, _ := git.OpenRepository(repo.RepoPath())
  299. commitID, _ := gitRepo.GetBranchCommitID(branchName)
  300. req := &grampus.GenerateTrainJobReq{
  301. JobName: jobName,
  302. DisplayJobName: displayJobName,
  303. ComputeResource: models.NPUResource,
  304. Command: command,
  305. ResourceSpecId: form.FlavorID,
  306. ImageUrl: "",
  307. ImageId: form.ImageID,
  308. DataUrl: dataPath,
  309. Description: description,
  310. CodeObsPath: codeObsPath,
  311. BootFileUrl: codeObsPath + bootFile,
  312. BootFile: bootFile,
  313. WorkServerNumber: form.WorkServerNumber,
  314. Uuid: uuid,
  315. CommitID: commitID,
  316. IsLatestVersion: isLatestVersion,
  317. BranchName: branchName,
  318. Params: form.Params,
  319. FlavorName: flavorName,
  320. EngineName: engineName,
  321. VersionCount: versionCount,
  322. TotalVersionCount: modelarts.TotalVersionCount,
  323. DatasetName: attachment.Name,
  324. }
  325. err = grampus.GenerateTrainJob(ctx, req)
  326. if err != nil {
  327. log.Error("GenerateTrainJob failed:%v", err.Error())
  328. grampusTrainJobNpuNewDataPrepare(ctx)
  329. ctx.RenderWithErr(err.Error(), tplGrampusTrainJobNPUNew, &form)
  330. return
  331. }
  332. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  333. }
  334. func GrampusStopJob(ctx *context.Context) {
  335. var ID = ctx.Params(":jobid")
  336. var resultCode = "0"
  337. var errorMsg = ""
  338. var status = ""
  339. task := ctx.Cloudbrain
  340. for {
  341. if task.Status == string(models.GrampusStatusStopped) || task.Status == string(models.GrampusStatusFailed) || task.Status == string(models.GrampusStatusSucceeded) {
  342. log.Error("the job(%s) has been stopped", task.JobName, ctx.Data["msgID"])
  343. resultCode = "-1"
  344. errorMsg = "system error"
  345. break
  346. }
  347. res, err := grampus.StopJob(task.JobID)
  348. if err != nil {
  349. log.Error("StopJob(%s) failed:%v", task.JobName, err, ctx.Data["msgID"])
  350. resultCode = strconv.Itoa(res.ErrorCode)
  351. errorMsg = res.ErrorMsg
  352. break
  353. }
  354. task.Status = string(models.GrampusStatusStopped)
  355. if task.EndTime == 0 {
  356. task.EndTime = timeutil.TimeStampNow()
  357. }
  358. task.ComputeAndSetDuration()
  359. err = models.UpdateJob(task)
  360. if err != nil {
  361. log.Error("UpdateJob(%s) failed:%v", task.JobName, err, ctx.Data["msgID"])
  362. resultCode = "-1"
  363. errorMsg = "system error"
  364. break
  365. }
  366. status = task.Status
  367. break
  368. }
  369. ctx.JSON(200, map[string]interface{}{
  370. "result_code": resultCode,
  371. "error_msg": errorMsg,
  372. "status": status,
  373. "id": ID,
  374. "StatusOK": 0,
  375. })
  376. }
  377. func GrampusTrainJobDel(ctx *context.Context) {
  378. var listType = ctx.Query("listType")
  379. if err := deleteGrampusJob(ctx); err != nil {
  380. log.Error("deleteGrampusJob failed: %v", err, ctx.Data["msgID"])
  381. ctx.ServerError(err.Error(), err)
  382. return
  383. }
  384. var isAdminPage = ctx.Query("isadminpage")
  385. var isHomePage = ctx.Query("ishomepage")
  386. if ctx.IsUserSiteAdmin() && isAdminPage == "true" {
  387. ctx.Redirect(setting.AppSubURL + "/admin" + "/cloudbrains")
  388. } else if isHomePage == "true" {
  389. ctx.Redirect(setting.AppSubURL + "/cloudbrains")
  390. } else {
  391. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job?listType=" + listType)
  392. }
  393. }
  394. func deleteGrampusJob(ctx *context.Context) error {
  395. task := ctx.Cloudbrain
  396. if task.Status != string(models.GrampusStatusStopped) && task.Status != string(models.GrampusStatusSucceeded) && task.Status != string(models.GrampusStatusFailed) {
  397. log.Error("the job(%s) has not been stopped", task.JobName, ctx.Data["msgID"])
  398. return errors.New("the job has not been stopped")
  399. }
  400. err := models.DeleteJob(task)
  401. if err != nil {
  402. log.Error("DeleteJob failed: %v", err, ctx.Data["msgID"])
  403. return err
  404. }
  405. storageType := models.TypeCloudBrainOne
  406. if task.ComputeResource == models.NPUResource {
  407. storageType = models.TypeCloudBrainTwo
  408. }
  409. deleteJobStorage(task.JobName, storageType)
  410. return nil
  411. }
  412. func GrampusTrainJobShow(ctx *context.Context) {
  413. ctx.Data["PageIsCloudBrain"] = true
  414. var task *models.Cloudbrain
  415. task, err := models.GetCloudbrainByJobIDWithDeleted(ctx.Params(":jobid"))
  416. if err != nil {
  417. log.Error("GetCloudbrainByJobID failed:" + err.Error())
  418. ctx.ServerError("system error", err)
  419. return
  420. }
  421. if task.DeletedAt.IsZero() { //normal record
  422. result, err := grampus.GetJob(task.JobID)
  423. if err != nil {
  424. log.Error("GetJob failed:" + err.Error())
  425. ctx.ServerError("GetJob failed", err)
  426. return
  427. }
  428. if result != nil {
  429. task.Status = grampus.TransTrainJobStatus(result.JobInfo.Status)
  430. if task.Status != result.JobInfo.Status || result.JobInfo.Status == models.GrampusStatusRunning {
  431. task.Duration = result.JobInfo.RunSec
  432. task.TrainJobDuration = models.ConvertDurationToStr(task.Duration)
  433. if task.StartTime == 0 && result.JobInfo.StartedAt > 0 {
  434. task.StartTime = timeutil.TimeStamp(result.JobInfo.StartedAt)
  435. }
  436. if task.EndTime == 0 && models.IsTrainJobTerminal(task.Status) && task.StartTime > 0 {
  437. task.EndTime = task.StartTime.Add(task.Duration)
  438. }
  439. task.CorrectCreateUnix()
  440. err = models.UpdateJob(task)
  441. if err != nil {
  442. log.Error("UpdateJob failed:" + err.Error())
  443. }
  444. }
  445. }
  446. }
  447. if len(task.Parameters) > 0 {
  448. var parameters models.Parameters
  449. err := json.Unmarshal([]byte(task.Parameters), &parameters)
  450. if err != nil {
  451. log.Error("Failed to Unmarshal Parameters: %s (%v)", task.Parameters, err)
  452. ctx.ServerError("system error", err)
  453. return
  454. }
  455. if len(parameters.Parameter) > 0 {
  456. paramTemp := ""
  457. for _, Parameter := range parameters.Parameter {
  458. param := Parameter.Label + " = " + Parameter.Value + "; "
  459. paramTemp = paramTemp + param
  460. }
  461. task.Parameters = paramTemp[:len(paramTemp)-2]
  462. } else {
  463. task.Parameters = ""
  464. }
  465. }
  466. taskList := make([]*models.Cloudbrain, 0)
  467. taskList = append(taskList, task)
  468. ctx.Data["version_list_task"] = taskList
  469. ctx.HTML(http.StatusOK, tplGrampusTrainJobShow)
  470. }
  471. func GrampusGetLog(ctx *context.Context) {
  472. jobID := ctx.Params(":jobid")
  473. job, err := models.GetCloudbrainByJobID(jobID)
  474. if err != nil {
  475. log.Error("GetCloudbrainByJobID failed: %v", err, ctx.Data["MsgID"])
  476. ctx.ServerError(err.Error(), err)
  477. return
  478. }
  479. content, err := grampus.GetTrainJobLog(job.JobID)
  480. if err != nil {
  481. log.Error("GetJobLog failed: %v", err, ctx.Data["MsgID"])
  482. ctx.ServerError(err.Error(), err)
  483. return
  484. }
  485. ctx.JSON(http.StatusOK, map[string]interface{}{
  486. "JobName": job.JobName,
  487. "Content": content,
  488. })
  489. return
  490. }
  491. func generateCommand(processorType, codeObsPath, dataObsPath, params, outputPath, datasetName string) (string, error) {
  492. var command string
  493. command += grampus.CommandPrepareScript
  494. //download code & dataset
  495. if processorType == grampus.ProcessorTypeNPU {
  496. commandDownload := "python " + grampus.ScriptSyncObsCodeAndDataset + " --access_key=" + setting.AccessKeyID + " --secret_key=" + setting.SecretAccessKey + " --project_id=" + setting.ProjectID + " --region_name=" + setting.Location + " --code_obs_dir=" + codeObsPath + " --data_obs_dir=" + dataObsPath + " --dataset_name=" + datasetName + ";"
  497. command += commandDownload
  498. } else if processorType == grampus.ProcessorTypeGPU {
  499. }
  500. //unzip dataset
  501. //exec code
  502. //upload models
  503. return command, nil
  504. }
  505. func generateCommandObsDownloadFile(srcObsFile, dstLocalDir string) (string, error) {
  506. var command string
  507. command = "python;"
  508. command += "from modelarts.session import Session \n"
  509. command += fmt.Sprintf("session = Session(access_key='%s',secret_key='%s', project_id='%s', region_name='%s') \n", setting.AccessKeyID, setting.SecretAccessKey, setting.ProjectID, setting.Location)
  510. if util.IsDir(srcObsFile) {
  511. command += fmt.Sprintf("session.obs.download_dir(src_obs_dir=\"%s\", dst_local_dir=\"%s\") \n", srcObsFile, dstLocalDir)
  512. } else {
  513. command += fmt.Sprintf("session.obs.download_file(src_obs_file=\"%s\", dst_local_dir=\"%s\") \n", srcObsFile, dstLocalDir)
  514. }
  515. return command, nil
  516. }