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.

modelarts.go 19 kB

3 years ago
3 years ago
4 years ago
3 years ago
4 years ago
4 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
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 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
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
3 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
3 years ago
3 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 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
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
4 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628
  1. // Copyright 2016 The Gogs Authors. All rights reserved.
  2. // Copyright 2018 The Gitea Authors. All rights reserved.
  3. // Use of this source code is governed by a MIT-style
  4. // license that can be found in the LICENSE file.
  5. package repo
  6. import (
  7. "encoding/json"
  8. "net/http"
  9. "path"
  10. "strconv"
  11. "strings"
  12. "code.gitea.io/gitea/modules/urfs_client/urchin"
  13. "code.gitea.io/gitea/modules/notification"
  14. "code.gitea.io/gitea/modules/grampus"
  15. "code.gitea.io/gitea/modules/setting"
  16. "code.gitea.io/gitea/models"
  17. "code.gitea.io/gitea/modules/cloudbrain"
  18. "code.gitea.io/gitea/modules/context"
  19. "code.gitea.io/gitea/modules/log"
  20. "code.gitea.io/gitea/modules/modelarts"
  21. "code.gitea.io/gitea/modules/storage"
  22. "code.gitea.io/gitea/modules/timeutil"
  23. routerRepo "code.gitea.io/gitea/routers/repo"
  24. cloudbrainService "code.gitea.io/gitea/services/cloudbrain"
  25. )
  26. func GetModelArtsNotebook2(ctx *context.APIContext) {
  27. var (
  28. err error
  29. )
  30. ID := ctx.Params(":id")
  31. job, err := models.GetCloudbrainByID(ID)
  32. if err != nil {
  33. ctx.NotFound(err)
  34. return
  35. }
  36. err = modelarts.HandleNotebookInfo(job)
  37. if err != nil {
  38. ctx.NotFound(err)
  39. return
  40. }
  41. ctx.JSON(http.StatusOK, map[string]interface{}{
  42. "ID": ID,
  43. "JobName": job.JobName,
  44. "JobStatus": job.Status,
  45. "JobDuration": job.TrainJobDuration,
  46. "StartTime": job.StartTime,
  47. })
  48. }
  49. func GetModelArtsTrainJob(ctx *context.APIContext) {
  50. var (
  51. err error
  52. )
  53. jobID := ctx.Params(":jobid")
  54. repoID := ctx.Repo.Repository.ID
  55. job, err := models.GetRepoCloudBrainByJobID(repoID, jobID)
  56. if err != nil {
  57. ctx.NotFound(err)
  58. return
  59. }
  60. result, err := modelarts.GetTrainJob(jobID, strconv.FormatInt(job.VersionID, 10))
  61. if err != nil {
  62. ctx.NotFound(err)
  63. return
  64. }
  65. oldStatus := job.Status
  66. job.Status = modelarts.TransTrainJobStatus(result.IntStatus)
  67. job.Duration = result.Duration
  68. job.TrainJobDuration = result.TrainJobDuration
  69. if oldStatus != job.Status {
  70. notification.NotifyChangeCloudbrainStatus(job, oldStatus)
  71. }
  72. err = models.UpdateJob(job)
  73. if err != nil {
  74. log.Error("UpdateJob failed:", err)
  75. }
  76. ctx.JSON(http.StatusOK, map[string]interface{}{
  77. "JobID": jobID,
  78. "JobStatus": job.Status,
  79. "JobDuration": job.Duration,
  80. })
  81. }
  82. func GetModelArtsTrainJobVersion(ctx *context.APIContext) {
  83. var (
  84. err error
  85. aiCenterName string
  86. )
  87. jobID := ctx.Params(":jobid")
  88. versionName := ctx.Query("version_name")
  89. job, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  90. if err != nil {
  91. ctx.NotFound(err)
  92. return
  93. }
  94. if job.Type == models.TypeCloudBrainOne {
  95. jobResult, err := cloudbrain.GetJob(job.JobID)
  96. if err != nil {
  97. ctx.NotFound(err)
  98. log.Error("GetJob failed:", err)
  99. return
  100. }
  101. result, err := models.ConvertToJobResultPayload(jobResult.Payload)
  102. if err != nil {
  103. ctx.NotFound(err)
  104. log.Error("ConvertToJobResultPayload failed:", err)
  105. return
  106. }
  107. oldStatus := job.Status
  108. job.Status = result.JobStatus.State
  109. if result.JobStatus.State != string(models.JobWaiting) && result.JobStatus.State != string(models.JobFailed) {
  110. taskRoles := result.TaskRoles
  111. taskRes, _ := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
  112. job.ContainerIp = taskRes.TaskStatuses[0].ContainerIP
  113. job.ContainerID = taskRes.TaskStatuses[0].ContainerID
  114. job.Status = taskRes.TaskStatuses[0].State
  115. }
  116. if result.JobStatus.State != string(models.JobWaiting) {
  117. models.ParseAndSetDurationFromCloudBrainOne(result, job)
  118. if oldStatus != job.Status {
  119. notification.NotifyChangeCloudbrainStatus(job, oldStatus)
  120. }
  121. err = models.UpdateJob(job)
  122. if err != nil {
  123. log.Error("UpdateJob failed:", err)
  124. }
  125. }
  126. } else if job.Type == models.TypeCloudBrainTwo {
  127. err := modelarts.HandleTrainJobInfo(job)
  128. if err != nil {
  129. ctx.NotFound(err)
  130. return
  131. }
  132. } else if job.Type == models.TypeC2Net {
  133. result, err := grampus.GetJob(jobID)
  134. if err != nil {
  135. log.Error("GetJob(%s) failed:%v", job.JobName, err)
  136. ctx.NotFound(err)
  137. return
  138. }
  139. if job.StartTime == 0 && result.JobInfo.StartedAt > 0 {
  140. job.StartTime = timeutil.TimeStamp(result.JobInfo.StartedAt)
  141. }
  142. oldStatus := job.Status
  143. job.Status = grampus.TransTrainJobStatus(result.JobInfo.Status)
  144. job.Duration = result.JobInfo.RunSec
  145. job.TrainJobDuration = models.ConvertDurationToStr(job.Duration)
  146. if job.EndTime == 0 && models.IsTrainJobTerminal(job.Status) && job.StartTime > 0 {
  147. job.EndTime = job.StartTime.Add(job.Duration)
  148. }
  149. job.CorrectCreateUnix()
  150. if len(job.AiCenter) == 0 {
  151. if len(result.JobInfo.Tasks) > 0 {
  152. if len(result.JobInfo.Tasks[0].CenterID) > 0 && len(result.JobInfo.Tasks[0].CenterName) > 0 {
  153. job.AiCenter = result.JobInfo.Tasks[0].CenterID[0] + "+" + result.JobInfo.Tasks[0].CenterName[0]
  154. aiCenterName = result.JobInfo.Tasks[0].CenterName[0]
  155. }
  156. }
  157. } else {
  158. aiCenterName = cloudbrainService.GetAiCenterShow(job.AiCenter, ctx.Context)
  159. }
  160. if oldStatus != job.Status {
  161. notification.NotifyChangeCloudbrainStatus(job, oldStatus)
  162. if models.IsTrainJobTerminal(job.Status) {
  163. if len(result.JobInfo.Tasks[0].CenterID) == 1 {
  164. urchin.GetBackNpuModel(job.ID, grampus.GetRemoteEndPoint(result.JobInfo.Tasks[0].CenterID[0]), grampus.BucketRemote, grampus.GetNpuModelObjectKey(job.JobName), grampus.GetCenterProxy(setting.Grampus.LocalCenterID))
  165. }
  166. }
  167. }
  168. err = models.UpdateTrainJobVersion(job)
  169. if err != nil {
  170. log.Error("UpdateJob failed:", err)
  171. }
  172. }
  173. ctx.JSON(http.StatusOK, map[string]interface{}{
  174. "JobID": jobID,
  175. "JobStatus": job.Status,
  176. "JobDuration": job.TrainJobDuration,
  177. "AiCenter": aiCenterName,
  178. "StartTime": job.StartTime,
  179. })
  180. }
  181. func TrainJobForModelConvertGetLog(ctx *context.APIContext) {
  182. var (
  183. err error
  184. )
  185. var jobID = ctx.Params(":id")
  186. var baseLine = ctx.Query("base_line")
  187. var order = ctx.Query("order")
  188. var lines = ctx.Query("lines")
  189. lines_int, err := strconv.Atoi(lines)
  190. if err != nil {
  191. log.Error("change lines(%d) string to int failed", lines_int)
  192. }
  193. if order != modelarts.OrderDesc && order != modelarts.OrderAsc {
  194. log.Error("order(%s) check failed", order)
  195. ctx.JSON(http.StatusBadRequest, map[string]interface{}{
  196. "err_msg": "order check failed",
  197. })
  198. return
  199. }
  200. resultLogFile, result, err := trainJobForModelConvertGetLogContent(jobID, baseLine, order, lines_int)
  201. if err != nil {
  202. log.Error("trainJobGetLog(%s) failed:%v", jobID, err.Error())
  203. // ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  204. ctx.JSON(http.StatusOK, map[string]interface{}{
  205. "JobID": jobID,
  206. "LogFileName": "",
  207. "StartLine": "0",
  208. "EndLine": "0",
  209. "Content": "",
  210. "Lines": 0,
  211. })
  212. return
  213. }
  214. ctx.Data["log_file_name"] = resultLogFile.LogFileList[0]
  215. ctx.JSON(http.StatusOK, map[string]interface{}{
  216. "JobID": jobID,
  217. "LogFileName": resultLogFile.LogFileList[0],
  218. "StartLine": result.StartLine,
  219. "EndLine": result.EndLine,
  220. "Content": result.Content,
  221. "Lines": result.Lines,
  222. })
  223. }
  224. func trainJobForModelConvertGetLogContent(jobID string, baseLine string, order string, lines int) (*models.GetTrainJobLogFileNamesResult, *models.GetTrainJobLogResult, error) {
  225. task, err := models.QueryModelConvertById(jobID)
  226. if err != nil {
  227. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  228. return nil, nil, err
  229. }
  230. resultLogFile, err := modelarts.GetTrainJobLogFileNames(task.CloudBrainTaskId, task.ModelArtsVersionId)
  231. if err != nil {
  232. log.Error("GetTrainJobLogFileNames(%s) failed:%v", task.CloudBrainTaskId, err.Error())
  233. return nil, nil, err
  234. }
  235. result, err := modelarts.GetTrainJobLog(task.CloudBrainTaskId, task.ModelArtsVersionId, baseLine, resultLogFile.LogFileList[0], order, lines)
  236. if err != nil {
  237. log.Error("GetTrainJobLog(%s) failed:%v", task.CloudBrainTaskId, err.Error())
  238. return nil, nil, err
  239. }
  240. return resultLogFile, result, err
  241. }
  242. func TrainJobGetLog(ctx *context.APIContext) {
  243. var (
  244. err error
  245. )
  246. var jobID = ctx.Params(":jobid")
  247. var versionName = ctx.Query("version_name")
  248. var baseLine = ctx.Query("base_line")
  249. var order = ctx.Query("order")
  250. var lines = ctx.Query("lines")
  251. lines_int, err := strconv.Atoi(lines)
  252. if err != nil {
  253. log.Error("change lines(%d) string to int failed", lines_int)
  254. }
  255. if order != modelarts.OrderDesc && order != modelarts.OrderAsc {
  256. log.Error("order(%s) check failed", order)
  257. ctx.JSON(http.StatusBadRequest, map[string]interface{}{
  258. "err_msg": "order check failed",
  259. })
  260. return
  261. }
  262. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  263. if err != nil {
  264. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  265. return
  266. }
  267. resultLogFile, result, err := trainJobGetLogContent(jobID, task.VersionID, baseLine, order, lines_int)
  268. if err != nil {
  269. log.Error("trainJobGetLog(%s) failed:%v", jobID, err.Error())
  270. // ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  271. return
  272. }
  273. prefix := strings.TrimPrefix(path.Join(setting.TrainJobModelPath, task.JobName, modelarts.LogPath, versionName), "/") + "/job"
  274. _, err = storage.GetObsLogFileName(prefix)
  275. var canLogDownload bool
  276. if err != nil {
  277. canLogDownload = false
  278. } else {
  279. canLogDownload = true
  280. }
  281. ctx.Data["log_file_name"] = resultLogFile.LogFileList[0]
  282. ctx.JSON(http.StatusOK, map[string]interface{}{
  283. "JobID": jobID,
  284. "LogFileName": resultLogFile.LogFileList[0],
  285. "StartLine": result.StartLine,
  286. "EndLine": result.EndLine,
  287. "Content": result.Content,
  288. "Lines": result.Lines,
  289. "CanLogDownload": canLogDownload,
  290. "StartTime": task.StartTime,
  291. })
  292. }
  293. func trainJobGetLogContent(jobID string, versionID int64, baseLine string, order string, lines int) (*models.GetTrainJobLogFileNamesResult, *models.GetTrainJobLogResult, error) {
  294. resultLogFile, err := modelarts.GetTrainJobLogFileNames(jobID, strconv.FormatInt(versionID, 10))
  295. if err != nil {
  296. log.Error("GetTrainJobLogFileNames(%s) failed:%v", jobID, err.Error())
  297. return nil, nil, err
  298. }
  299. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(versionID, 10), baseLine, resultLogFile.LogFileList[0], order, lines)
  300. if err != nil {
  301. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  302. return nil, nil, err
  303. }
  304. return resultLogFile, result, err
  305. }
  306. func DelTrainJobVersion(ctx *context.APIContext) {
  307. var (
  308. err error
  309. )
  310. var jobID = ctx.Params(":jobid")
  311. var versionName = ctx.Query("version_name")
  312. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  313. if err != nil {
  314. log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
  315. ctx.NotFound(err)
  316. return
  317. }
  318. if task.Status != string(models.ModelArtsTrainJobImageFailed) && task.Status != string(models.ModelArtsTrainJobSubmitFailed) && task.Status != string(models.ModelArtsTrainJobDeleteFailed) &&
  319. task.Status != string(models.ModelArtsTrainJobCompleted) && task.Status != string(models.ModelArtsTrainJobFailed) &&
  320. task.Status != string(models.ModelArtsTrainJobKilled) && task.Status != string(models.ModelArtsTrainJobCanceled) && task.Status != string(models.ModelArtsTrainJobLost) {
  321. log.Error("the job(%s) version has not been stopped", task.JobName)
  322. ctx.NotFound(err)
  323. return
  324. }
  325. //删除modelarts上的记录
  326. _, err = modelarts.DelTrainJobVersion(jobID, strconv.FormatInt(task.VersionID, 10))
  327. if err != nil {
  328. log.Error("DelTrainJobVersion(%s) failed:%v", task.JobName, err.Error())
  329. ctx.NotFound(err)
  330. return
  331. }
  332. //删除数据库记录
  333. err = models.DeleteJob(task)
  334. if err != nil {
  335. ctx.ServerError("DeleteJob failed", err)
  336. ctx.NotFound(err)
  337. return
  338. }
  339. //获取删除后的版本数量
  340. var jobTypes []string
  341. jobTypes = append(jobTypes, string(models.JobTypeTrain))
  342. repo := ctx.Repo.Repository
  343. VersionTaskList, VersionListCount, err := models.CloudbrainsVersionList(&models.CloudbrainsOptions{
  344. RepoID: repo.ID,
  345. Type: models.TypeCloudBrainTwo,
  346. JobTypes: jobTypes,
  347. JobID: jobID,
  348. })
  349. if err != nil {
  350. ctx.ServerError("get VersionListCount failed", err)
  351. return
  352. }
  353. if VersionListCount > 0 {
  354. // 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
  355. if task.IsLatestVersion == modelarts.IsLatestVersion {
  356. err = models.SetVersionCountAndLatestVersion(jobID, VersionTaskList[0].Cloudbrain.VersionName, VersionListCount, modelarts.IsLatestVersion, task.TotalVersionCount)
  357. if err != nil {
  358. ctx.ServerError("UpdateJobVersionCount failed", err)
  359. return
  360. }
  361. } else {
  362. err = models.SetVersionCountAndLatestVersion(jobID, VersionTaskList[0].VersionName, VersionListCount, modelarts.IsLatestVersion, VersionTaskList[0].Cloudbrain.TotalVersionCount)
  363. if err != nil {
  364. ctx.ServerError("UpdateJobVersionCount failed", err)
  365. return
  366. }
  367. }
  368. } else { //已删除该任务下的所有版本
  369. routerRepo.DeleteJobStorage(task.JobName)
  370. }
  371. ctx.JSON(http.StatusOK, map[string]interface{}{
  372. "JobID": jobID,
  373. "VersionName": versionName,
  374. "StatusOK": 0,
  375. "VersionListCount": VersionListCount,
  376. })
  377. }
  378. func StopTrainJobVersion(ctx *context.APIContext) {
  379. var (
  380. err error
  381. )
  382. var jobID = ctx.Params(":jobid")
  383. var versionName = ctx.Query("version_name")
  384. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  385. if err != nil {
  386. log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
  387. return
  388. }
  389. _, err = modelarts.StopTrainJob(jobID, strconv.FormatInt(task.VersionID, 10))
  390. if err != nil {
  391. log.Error("StopTrainJob(%s) failed:%v", task.JobName, err.Error())
  392. return
  393. }
  394. ctx.JSON(http.StatusOK, map[string]interface{}{
  395. "JobID": jobID,
  396. "VersionName": versionName,
  397. "StatusOK": 0,
  398. })
  399. }
  400. func ModelList(ctx *context.APIContext) {
  401. var (
  402. err error
  403. )
  404. var jobID = ctx.Params(":jobid")
  405. var versionName = ctx.Query("version_name")
  406. parentDir := ctx.Query("parentDir")
  407. dirArray := strings.Split(parentDir, "/")
  408. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  409. if err != nil {
  410. log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
  411. return
  412. }
  413. status := models.StorageScheduleSucceed
  414. var fileInfos []storage.FileInfo
  415. if task.ComputeResource == models.NPUResource {
  416. fileInfos, err = storage.GetObsListObject(task.JobName, "output/", parentDir, versionName)
  417. if err != nil {
  418. log.Info("get TrainJobListModel failed:", err)
  419. ctx.ServerError("GetObsListObject:", err)
  420. return
  421. }
  422. if task.Type == models.TypeC2Net {
  423. if len(fileInfos) > 0 {
  424. status = models.StorageScheduleSucceed
  425. } else {
  426. if models.IsTrainJobTerminal(task.Status) {
  427. if task.Status == models.GrampusStatusStopped {
  428. status = models.StorageNoFile
  429. } else {
  430. record, _ := models.GetScheduleRecordByCloudbrainID(task.ID)
  431. if record != nil {
  432. status = record.Status
  433. if status == models.StorageScheduleSucceed {
  434. status = models.StorageNoFile
  435. }
  436. } else {
  437. status = models.StorageScheduleProcessing
  438. }
  439. }
  440. } else {
  441. status = models.StorageScheduleWaiting
  442. }
  443. }
  444. }
  445. } else if task.ComputeResource == models.GPUResource {
  446. files, err := routerRepo.GetModelDirs(task.JobName, parentDir)
  447. if err != nil {
  448. log.Info("GetModelDirs failed:", err)
  449. ctx.ServerError("GetModelDirs:", err)
  450. return
  451. }
  452. err = json.Unmarshal([]byte(files), &fileInfos)
  453. if err != nil {
  454. log.Error("json.Unmarshal failed:%v", err.Error(), ctx.Data["msgID"])
  455. ctx.ServerError("json.Unmarshal failed:", err)
  456. return
  457. }
  458. }
  459. ctx.JSON(http.StatusOK, map[string]interface{}{
  460. "JobID": jobID,
  461. "VersionName": versionName,
  462. "StatusOK": status,
  463. "Path": dirArray,
  464. "Dirs": fileInfos,
  465. "task": task,
  466. "PageIsCloudBrain": true,
  467. })
  468. }
  469. func GetModelArtsInferenceJob(ctx *context.APIContext) {
  470. var (
  471. err error
  472. )
  473. jobID := ctx.Params(":jobid")
  474. job, err := models.GetCloudbrainByJobID(jobID)
  475. if err != nil {
  476. ctx.NotFound(err)
  477. return
  478. }
  479. err = modelarts.HandleTrainJobInfo(job)
  480. if err != nil {
  481. ctx.NotFound(err)
  482. return
  483. }
  484. ctx.JSON(http.StatusOK, map[string]interface{}{
  485. "JobID": jobID,
  486. "JobStatus": job.Status,
  487. "JobDuration": job.TrainJobDuration,
  488. "StartTime": job.StartTime,
  489. })
  490. }
  491. func ResultList(ctx *context.APIContext) {
  492. var (
  493. err error
  494. )
  495. var jobID = ctx.Params(":jobid")
  496. var versionName = ctx.Query("version_name")
  497. parentDir := ctx.Query("parentDir")
  498. dirArray := strings.Split(parentDir, "/")
  499. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  500. if err != nil {
  501. log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
  502. return
  503. }
  504. models, err := storage.GetObsListObject(task.JobName, "result/", parentDir, versionName)
  505. if err != nil {
  506. log.Info("get TrainJobListModel failed:", err)
  507. ctx.ServerError("GetObsListObject:", err)
  508. return
  509. }
  510. ctx.JSON(http.StatusOK, map[string]interface{}{
  511. "JobID": jobID,
  512. "VersionName": versionName,
  513. "StatusOK": 0,
  514. "Path": dirArray,
  515. "Dirs": models,
  516. "task": task,
  517. "PageIsCloudBrain": true,
  518. })
  519. }
  520. func TrainJobGetMetricStatistic(ctx *context.APIContext) {
  521. var (
  522. err error
  523. )
  524. var jobID = ctx.Params(":jobid")
  525. var versionName = ctx.Query("version_name")
  526. result, err := trainJobGetMetricStatistic(jobID, versionName)
  527. if err != nil {
  528. log.Error("trainJobGetMetricStatistic(%s) failed:%v", jobID, err.Error())
  529. return
  530. }
  531. ctx.JSON(http.StatusOK, map[string]interface{}{
  532. "JobID": jobID,
  533. "Interval": result.Interval,
  534. "MetricsInfo": result.MetricsInfo,
  535. })
  536. }
  537. func trainJobGetMetricStatistic(jobID string, versionName string) (*models.GetTrainJobMetricStatisticResult, error) {
  538. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  539. if err != nil {
  540. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  541. return nil, err
  542. }
  543. resultLogFile, err := modelarts.GetTrainJobLogFileNames(jobID, strconv.FormatInt(task.VersionID, 10))
  544. if err != nil {
  545. log.Error("GetTrainJobLogFileNames(%s) failed:%v", jobID, err.Error())
  546. return nil, err
  547. }
  548. result, err := modelarts.GetTrainJobMetricStatistic(jobID, strconv.FormatInt(task.VersionID, 10), resultLogFile.LogFileList[0])
  549. if err != nil {
  550. log.Error("GetTrainJobMetricStatistic(%s) failed:%v", jobID, err.Error())
  551. return nil, err
  552. }
  553. return result, err
  554. }