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 48 kB

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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
4 years ago
4 years ago
4 years ago
12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526
  1. package repo
  2. import (
  3. "encoding/json"
  4. "errors"
  5. "io"
  6. "net/http"
  7. "os"
  8. "path"
  9. "strconv"
  10. "strings"
  11. "time"
  12. "code.gitea.io/gitea/modules/cloudbrain"
  13. "code.gitea.io/gitea/models"
  14. "code.gitea.io/gitea/modules/auth"
  15. "code.gitea.io/gitea/modules/base"
  16. "code.gitea.io/gitea/modules/context"
  17. "code.gitea.io/gitea/modules/git"
  18. "code.gitea.io/gitea/modules/log"
  19. "code.gitea.io/gitea/modules/modelarts"
  20. "code.gitea.io/gitea/modules/obs"
  21. "code.gitea.io/gitea/modules/setting"
  22. "code.gitea.io/gitea/modules/storage"
  23. "github.com/unknwon/com"
  24. )
  25. const (
  26. tplDebugJobIndex base.TplName = "repo/debugjob/index"
  27. tplModelArtsNotebookIndex base.TplName = "repo/modelarts/notebook/index"
  28. tplModelArtsNotebookNew base.TplName = "repo/modelarts/notebook/new"
  29. tplModelArtsNotebookShow base.TplName = "repo/modelarts/notebook/show"
  30. tplModelArtsTrainJobIndex base.TplName = "repo/modelarts/trainjob/index"
  31. tplModelArtsTrainJobNew base.TplName = "repo/modelarts/trainjob/new"
  32. tplModelArtsTrainJobShow base.TplName = "repo/modelarts/trainjob/show"
  33. tplModelArtsTrainJobVersionNew base.TplName = "repo/modelarts/trainjob/version_new"
  34. )
  35. func DebugJobIndex(ctx *context.Context) {
  36. debugListType := ctx.Query("debugListType")
  37. MustEnableCloudbrain(ctx)
  38. repo := ctx.Repo.Repository
  39. page := ctx.QueryInt("page")
  40. if page <= 0 {
  41. page = 1
  42. }
  43. debugType := modelarts.DebugType
  44. jobType := string(models.JobTypeDebug)
  45. if debugListType == modelarts.GPUResource {
  46. debugType = models.TypeCloudBrainOne
  47. jobType = ""
  48. }
  49. if debugListType == modelarts.NPUResource {
  50. debugType = models.TypeCloudBrainTwo
  51. }
  52. ciTasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  53. ListOptions: models.ListOptions{
  54. Page: page,
  55. PageSize: setting.UI.IssuePagingNum,
  56. },
  57. RepoID: repo.ID,
  58. Type: debugType,
  59. JobType: jobType,
  60. })
  61. if err != nil {
  62. ctx.ServerError("Get debugjob faild:", err)
  63. return
  64. }
  65. for i, task := range ciTasks {
  66. if task.Cloudbrain.Type == models.TypeCloudBrainOne {
  67. ciTasks[i].CanDebug = cloudbrain.CanCreateOrDebugJob(ctx)
  68. ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  69. ciTasks[i].Cloudbrain.ComputeResource = modelarts.GPUResource
  70. }
  71. if task.Cloudbrain.Type == models.TypeCloudBrainTwo {
  72. ciTasks[i].CanDebug = cloudbrain.CanCreateOrDebugJob(ctx)
  73. ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  74. ciTasks[i].Cloudbrain.ComputeResource = modelarts.NPUResource
  75. }
  76. }
  77. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  78. pager.SetDefaultParams(ctx)
  79. ctx.Data["Page"] = pager
  80. ctx.Data["PageIsCloudBrain"] = true
  81. ctx.Data["Tasks"] = ciTasks
  82. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  83. ctx.HTML(200, tplDebugJobIndex)
  84. }
  85. // MustEnableDataset check if repository enable internal cb
  86. func MustEnableModelArts(ctx *context.Context) {
  87. if !ctx.Repo.CanRead(models.UnitTypeCloudBrain) {
  88. ctx.NotFound("MustEnableCloudbrain", nil)
  89. return
  90. }
  91. }
  92. func NotebookIndex(ctx *context.Context) {
  93. MustEnableModelArts(ctx)
  94. repo := ctx.Repo.Repository
  95. page := ctx.QueryInt("page")
  96. if page <= 0 {
  97. page = 1
  98. }
  99. ciTasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  100. ListOptions: models.ListOptions{
  101. Page: page,
  102. PageSize: setting.UI.IssuePagingNum,
  103. },
  104. RepoID: repo.ID,
  105. Type: models.TypeCloudBrainTwo,
  106. JobType: string(models.JobTypeDebug),
  107. })
  108. if err != nil {
  109. ctx.ServerError("Cloudbrain", err)
  110. return
  111. }
  112. for i, task := range ciTasks {
  113. ciTasks[i].CanDebug = cloudbrain.CanCreateOrDebugJob(ctx)
  114. ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  115. ciTasks[i].Cloudbrain.ComputeResource = modelarts.NPUResource
  116. }
  117. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  118. pager.SetDefaultParams(ctx)
  119. ctx.Data["Page"] = pager
  120. ctx.Data["PageIsCloudBrain"] = true
  121. ctx.Data["Tasks"] = ciTasks
  122. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  123. ctx.HTML(200, tplModelArtsNotebookIndex)
  124. }
  125. func NotebookNew(ctx *context.Context) {
  126. ctx.Data["PageIsCloudBrain"] = true
  127. t := time.Now()
  128. var jobName = jobNamePrefixValid(cutString(ctx.User.Name, 5)) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  129. ctx.Data["job_name"] = jobName
  130. attachs, err := models.GetModelArtsUserAttachments(ctx.User.ID)
  131. if err != nil {
  132. ctx.ServerError("GetAllUserAttachments failed:", err)
  133. return
  134. }
  135. ctx.Data["attachments"] = attachs
  136. ctx.Data["dataset_path"] = modelarts.DataSetMountPath
  137. ctx.Data["env"] = modelarts.NotebookEnv
  138. ctx.Data["notebook_type"] = modelarts.NotebookType
  139. if modelarts.FlavorInfos == nil {
  140. json.Unmarshal([]byte(setting.FlavorInfos), &modelarts.FlavorInfos)
  141. }
  142. ctx.Data["flavors"] = modelarts.FlavorInfos.FlavorInfo
  143. ctx.HTML(200, tplModelArtsNotebookNew)
  144. }
  145. func NotebookCreate(ctx *context.Context, form auth.CreateModelArtsNotebookForm) {
  146. ctx.Data["PageIsNotebook"] = true
  147. jobName := form.JobName
  148. uuid := form.Attachment
  149. description := form.Description
  150. flavor := form.Flavor
  151. count, err := models.GetCloudbrainNotebookCountByUserID(ctx.User.ID)
  152. if err != nil {
  153. log.Error("GetCloudbrainNotebookCountByUserID failed:%v", err, ctx.Data["MsgID"])
  154. cloudBrainNewDataPrepare(ctx)
  155. ctx.RenderWithErr("system error", tplModelArtsNotebookNew, &form)
  156. return
  157. } else {
  158. if count >= 1 {
  159. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  160. cloudBrainNewDataPrepare(ctx)
  161. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsNotebookNew, &form)
  162. return
  163. }
  164. }
  165. err = modelarts.GenerateTask(ctx, jobName, uuid, description, flavor)
  166. if err != nil {
  167. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookNew, &form)
  168. return
  169. }
  170. // ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/notebook")
  171. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob")
  172. }
  173. func NotebookShow(ctx *context.Context) {
  174. ctx.Data["PageIsCloudBrain"] = true
  175. var jobID = ctx.Params(":jobid")
  176. task, err := models.GetCloudbrainByJobID(jobID)
  177. if err != nil {
  178. ctx.Data["error"] = err.Error()
  179. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil)
  180. return
  181. }
  182. result, err := modelarts.GetJob(jobID)
  183. if err != nil {
  184. ctx.Data["error"] = err.Error()
  185. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil)
  186. return
  187. }
  188. if result != nil {
  189. task.Status = result.Status
  190. err = models.UpdateJob(task)
  191. if err != nil {
  192. ctx.Data["error"] = err.Error()
  193. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil)
  194. return
  195. }
  196. createTime, _ := com.StrTo(result.CreationTimestamp).Int64()
  197. result.CreateTime = time.Unix(int64(createTime/1000), 0).Format("2006-01-02 15:04:05")
  198. endTime, _ := com.StrTo(result.LatestUpdateTimestamp).Int64()
  199. result.LatestUpdateTime = time.Unix(int64(endTime/1000), 0).Format("2006-01-02 15:04:05")
  200. result.QueuingInfo.BeginTime = time.Unix(int64(result.QueuingInfo.BeginTimestamp/1000), 0).Format("2006-01-02 15:04:05")
  201. result.QueuingInfo.EndTime = time.Unix(int64(result.QueuingInfo.EndTimestamp/1000), 0).Format("2006-01-02 15:04:05")
  202. }
  203. ctx.Data["task"] = task
  204. ctx.Data["jobID"] = jobID
  205. ctx.Data["result"] = result
  206. ctx.HTML(200, tplModelArtsNotebookShow)
  207. }
  208. func NotebookDebug(ctx *context.Context) {
  209. var jobID = ctx.Params(":jobid")
  210. _, err := models.GetCloudbrainByJobID(jobID)
  211. if err != nil {
  212. ctx.ServerError("GetCloudbrainByJobID failed", err)
  213. return
  214. }
  215. result, err := modelarts.GetJob(jobID)
  216. if err != nil {
  217. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil)
  218. return
  219. }
  220. res, err := modelarts.GetJobToken(jobID)
  221. if err != nil {
  222. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil)
  223. return
  224. }
  225. urls := strings.Split(result.Spec.Annotations.Url, "/")
  226. urlPrefix := result.Spec.Annotations.TargetDomain
  227. for i, url := range urls {
  228. if i > 2 {
  229. urlPrefix += "/" + url
  230. }
  231. }
  232. debugUrl := urlPrefix + "?token=" + res.Token
  233. ctx.Redirect(debugUrl)
  234. }
  235. func NotebookStop(ctx *context.Context) {
  236. var jobID = ctx.Params(":jobid")
  237. debugListType := ctx.Query("debugListType")
  238. log.Info(jobID)
  239. task, err := models.GetCloudbrainByJobID(jobID)
  240. if err != nil {
  241. ctx.ServerError("GetCloudbrainByJobID failed", err)
  242. return
  243. }
  244. if task.Status != string(models.JobRunning) {
  245. log.Error("the job(%s) is not running", task.JobName)
  246. ctx.ServerError("the job is not running", errors.New("the job is not running"))
  247. return
  248. }
  249. param := models.NotebookAction{
  250. Action: models.ActionStop,
  251. }
  252. res, err := modelarts.StopJob(jobID, param)
  253. if err != nil {
  254. log.Error("StopJob(%s) failed:%v", task.JobName, err.Error())
  255. ctx.ServerError("StopJob failed", err)
  256. return
  257. }
  258. task.Status = res.CurrentStatus
  259. err = models.UpdateJob(task)
  260. if err != nil {
  261. ctx.ServerError("UpdateJob failed", err)
  262. return
  263. }
  264. if debugListType == modelarts.AllResource {
  265. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob")
  266. } else {
  267. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/notebook")
  268. }
  269. }
  270. func NotebookDel(ctx *context.Context) {
  271. var jobID = ctx.Params(":jobid")
  272. debugListType := ctx.Query("debugListType")
  273. task, err := models.GetCloudbrainByJobID(jobID)
  274. if err != nil {
  275. ctx.ServerError("GetCloudbrainByJobID failed", err)
  276. return
  277. }
  278. if task.Status != string(models.JobStopped) {
  279. log.Error("the job(%s) has not been stopped", task.JobName)
  280. ctx.ServerError("the job has not been stopped", errors.New("the job has not been stopped"))
  281. return
  282. }
  283. _, err = modelarts.DelNotebook(jobID)
  284. if err != nil {
  285. log.Error("DelJob(%s) failed:%v", task.JobName, err.Error())
  286. ctx.ServerError("DelJob failed", err)
  287. return
  288. }
  289. err = models.DeleteJob(task)
  290. if err != nil {
  291. ctx.ServerError("DeleteJob failed", err)
  292. return
  293. }
  294. if debugListType == modelarts.AllResource {
  295. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob")
  296. } else {
  297. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/notebook")
  298. }
  299. }
  300. func TrainJobIndex(ctx *context.Context) {
  301. MustEnableModelArts(ctx)
  302. repo := ctx.Repo.Repository
  303. page := ctx.QueryInt("page")
  304. if page <= 0 {
  305. page = 1
  306. }
  307. tasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  308. ListOptions: models.ListOptions{
  309. Page: page,
  310. PageSize: setting.UI.IssuePagingNum,
  311. },
  312. RepoID: repo.ID,
  313. Type: models.TypeCloudBrainTwo,
  314. JobType: string(models.JobTypeTrain),
  315. IsLatestVersion: modelarts.IsLatestVersion,
  316. })
  317. if err != nil {
  318. ctx.ServerError("Cloudbrain", err)
  319. return
  320. }
  321. for i, task := range tasks {
  322. tasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  323. tasks[i].CanModify = cloudbrain.CanModifyJob(ctx, &task.Cloudbrain)
  324. }
  325. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  326. pager.SetDefaultParams(ctx)
  327. ctx.Data["Page"] = pager
  328. ctx.Data["PageIsCloudBrain"] = true
  329. ctx.Data["Tasks"] = tasks
  330. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  331. ctx.HTML(200, tplModelArtsTrainJobIndex)
  332. }
  333. func TrainJobNew(ctx *context.Context) {
  334. err := trainJobNewDataPrepare(ctx)
  335. if err != nil {
  336. ctx.ServerError("get new train-job info failed", err)
  337. return
  338. }
  339. ctx.HTML(200, tplModelArtsTrainJobNew)
  340. }
  341. func trainJobNewDataPrepare(ctx *context.Context) error {
  342. ctx.Data["PageIsCloudBrain"] = true
  343. //can, err := canUserCreateTrainJob(ctx.User.ID)
  344. //if err != nil {
  345. // ctx.ServerError("canUserCreateTrainJob", err)
  346. // return
  347. //}
  348. //
  349. //if !can {
  350. // log.Error("the user can not create train-job")
  351. // ctx.ServerError("the user can not create train-job", fmt.Errorf("the user can not create train-job"))
  352. // return
  353. //}
  354. t := time.Now()
  355. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  356. ctx.Data["job_name"] = jobName
  357. attachs, err := models.GetModelArtsUserAttachments(ctx.User.ID)
  358. if err != nil {
  359. ctx.ServerError("GetAllUserAttachments failed:", err)
  360. return err
  361. }
  362. ctx.Data["attachments"] = attachs
  363. var resourcePools modelarts.ResourcePool
  364. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  365. ctx.ServerError("json.Unmarshal failed:", err)
  366. return err
  367. }
  368. ctx.Data["resource_pools"] = resourcePools.Info
  369. var engines modelarts.Engine
  370. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  371. ctx.ServerError("json.Unmarshal failed:", err)
  372. return err
  373. }
  374. ctx.Data["engines"] = engines.Info
  375. var versionInfos modelarts.VersionInfo
  376. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  377. ctx.ServerError("json.Unmarshal failed:", err)
  378. return err
  379. }
  380. ctx.Data["engine_versions"] = versionInfos.Version
  381. var flavorInfos modelarts.Flavor
  382. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  383. ctx.ServerError("json.Unmarshal failed:", err)
  384. return err
  385. }
  386. ctx.Data["flavor_infos"] = flavorInfos.Info
  387. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  388. ctx.Data["train_url"] = outputObsPath
  389. ctx.Data["params"] = ""
  390. ctx.Data["branchName"] = ctx.Repo.BranchName
  391. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  392. if err != nil {
  393. ctx.ServerError("getConfigList failed:", err)
  394. return err
  395. }
  396. ctx.Data["config_list"] = configList.ParaConfigs
  397. return nil
  398. }
  399. func trainJobErrorNewDataPrepare(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) error {
  400. ctx.Data["PageIsCloudBrain"] = true
  401. //can, err := canUserCreateTrainJob(ctx.User.ID)
  402. //if err != nil {
  403. // ctx.ServerError("canUserCreateTrainJob", err)
  404. // return
  405. //}
  406. //
  407. //if !can {
  408. // log.Error("the user can not create train-job")
  409. // ctx.ServerError("the user can not create train-job", fmt.Errorf("the user can not create train-job"))
  410. // return
  411. //}
  412. t := time.Now()
  413. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  414. ctx.Data["job_name"] = jobName
  415. attachs, err := models.GetModelArtsUserAttachments(ctx.User.ID)
  416. if err != nil {
  417. ctx.ServerError("GetAllUserAttachments failed:", err)
  418. return err
  419. }
  420. ctx.Data["attachments"] = attachs
  421. var resourcePools modelarts.ResourcePool
  422. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  423. ctx.ServerError("json.Unmarshal failed:", err)
  424. return err
  425. }
  426. ctx.Data["resource_pools"] = resourcePools.Info
  427. var engines modelarts.Engine
  428. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  429. ctx.ServerError("json.Unmarshal failed:", err)
  430. return err
  431. }
  432. ctx.Data["engines"] = engines.Info
  433. var versionInfos modelarts.VersionInfo
  434. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  435. ctx.ServerError("json.Unmarshal failed:", err)
  436. return err
  437. }
  438. ctx.Data["engine_versions"] = versionInfos.Version
  439. var flavorInfos modelarts.Flavor
  440. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  441. ctx.ServerError("json.Unmarshal failed:", err)
  442. return err
  443. }
  444. ctx.Data["flavor_infos"] = flavorInfos.Info
  445. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  446. ctx.Data["train_url"] = outputObsPath
  447. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  448. if err != nil {
  449. ctx.ServerError("getConfigList failed:", err)
  450. return err
  451. }
  452. var Parameters modelarts.Parameters
  453. if err = json.Unmarshal([]byte(form.Params), &Parameters); err != nil {
  454. ctx.ServerError("json.Unmarshal failed:", err)
  455. return err
  456. }
  457. ctx.Data["params"] = Parameters.Parameter
  458. ctx.Data["config_list"] = configList.ParaConfigs
  459. ctx.Data["bootFile"] = form.BootFile
  460. ctx.Data["uuid"] = form.Attachment
  461. ctx.Data["branch_name"] = form.BranchName
  462. return nil
  463. }
  464. func TrainJobNewVersion(ctx *context.Context) {
  465. err := trainJobNewVersionDataPrepare(ctx)
  466. if err != nil {
  467. ctx.ServerError("get new train-job info failed", err)
  468. return
  469. }
  470. ctx.HTML(200, tplModelArtsTrainJobVersionNew)
  471. }
  472. func trainJobNewVersionDataPrepare(ctx *context.Context) error {
  473. ctx.Data["PageIsCloudBrain"] = true
  474. var jobID = ctx.Params(":jobid")
  475. var versionName = ctx.Query("version_name")
  476. // canNewJob, err := canUserCreateTrainJobVersion(ctx, jobID, versionName)
  477. // if err != nil {
  478. // ctx.ServerError("canNewJob can info failed", err)
  479. // return err
  480. // }
  481. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  482. if err != nil {
  483. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  484. return err
  485. }
  486. t := time.Now()
  487. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  488. ctx.Data["job_name"] = task.JobName
  489. attachs, err := models.GetModelArtsUserAttachments(ctx.User.ID)
  490. if err != nil {
  491. ctx.ServerError("GetAllUserAttachments failed:", err)
  492. return err
  493. }
  494. ctx.Data["attachments"] = attachs
  495. var resourcePools modelarts.ResourcePool
  496. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  497. ctx.ServerError("json.Unmarshal failed:", err)
  498. return err
  499. }
  500. ctx.Data["resource_pools"] = resourcePools.Info
  501. var engines modelarts.Engine
  502. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  503. ctx.ServerError("json.Unmarshal failed:", err)
  504. return err
  505. }
  506. ctx.Data["engines"] = engines.Info
  507. var versionInfos modelarts.VersionInfo
  508. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  509. ctx.ServerError("json.Unmarshal failed:", err)
  510. return err
  511. }
  512. ctx.Data["engine_versions"] = versionInfos.Version
  513. var flavorInfos modelarts.Flavor
  514. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  515. ctx.ServerError("json.Unmarshal failed:", err)
  516. return err
  517. }
  518. ctx.Data["flavor_infos"] = flavorInfos.Info
  519. var Parameters modelarts.Parameters
  520. if err = json.Unmarshal([]byte(task.Parameters), &Parameters); err != nil {
  521. ctx.ServerError("json.Unmarshal failed:", err)
  522. return err
  523. }
  524. ctx.Data["params"] = Parameters.Parameter
  525. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  526. ctx.Data["train_url"] = outputObsPath
  527. branches, _, err := ctx.Repo.GitRepo.GetBranches(0, 0)
  528. if err != nil {
  529. ctx.ServerError("GetBranches error:", err)
  530. return err
  531. }
  532. ctx.Data["branches"] = branches
  533. ctx.Data["branch_name"] = task.BranchName
  534. ctx.Data["description"] = task.Description
  535. ctx.Data["boot_file"] = task.BootFile
  536. ctx.Data["dataset_name"] = task.DatasetName
  537. ctx.Data["work_server_number"] = task.WorkServerNumber
  538. ctx.Data["flavor_name"] = task.FlavorName
  539. ctx.Data["engine_name"] = task.EngineName
  540. ctx.Data["uuid"] = task.Uuid
  541. ctx.Data["flavor_code"] = task.FlavorCode
  542. ctx.Data["engine_id"] = task.EngineID
  543. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  544. if err != nil {
  545. ctx.ServerError("getConfigList failed:", err)
  546. return err
  547. }
  548. ctx.Data["config_list"] = configList.ParaConfigs
  549. return nil
  550. }
  551. func versionErrorDataPrepare(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) error {
  552. ctx.Data["PageIsCloudBrain"] = true
  553. var jobID = ctx.Params(":jobid")
  554. // var versionName = ctx.Params(":version-name")
  555. var versionName = ctx.Query("version_name")
  556. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  557. if err != nil {
  558. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  559. return err
  560. }
  561. t := time.Now()
  562. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  563. ctx.Data["job_name"] = task.JobName
  564. attachs, err := models.GetModelArtsUserAttachments(ctx.User.ID)
  565. if err != nil {
  566. ctx.ServerError("GetAllUserAttachments failed:", err)
  567. return err
  568. }
  569. ctx.Data["attachments"] = attachs
  570. var resourcePools modelarts.ResourcePool
  571. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  572. ctx.ServerError("json.Unmarshal failed:", err)
  573. return err
  574. }
  575. ctx.Data["resource_pools"] = resourcePools.Info
  576. var engines modelarts.Engine
  577. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  578. ctx.ServerError("json.Unmarshal failed:", err)
  579. return err
  580. }
  581. ctx.Data["engines"] = engines.Info
  582. var versionInfos modelarts.VersionInfo
  583. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  584. ctx.ServerError("json.Unmarshal failed:", err)
  585. return err
  586. }
  587. ctx.Data["engine_versions"] = versionInfos.Version
  588. var flavorInfos modelarts.Flavor
  589. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  590. ctx.ServerError("json.Unmarshal failed:", err)
  591. return err
  592. }
  593. ctx.Data["flavor_infos"] = flavorInfos.Info
  594. var Parameters modelarts.Parameters
  595. if err = json.Unmarshal([]byte(form.Params), &Parameters); err != nil {
  596. ctx.ServerError("json.Unmarshal failed:", err)
  597. return err
  598. }
  599. ctx.Data["params"] = Parameters.Parameter
  600. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  601. ctx.Data["train_url"] = outputObsPath
  602. branches, _, err := ctx.Repo.GitRepo.GetBranches(0, 0)
  603. if err != nil {
  604. ctx.ServerError("GetBranches error:", err)
  605. return err
  606. }
  607. ctx.Data["branches"] = branches
  608. ctx.Data["description"] = form.Description
  609. ctx.Data["dataset_name"] = task.DatasetName
  610. ctx.Data["work_server_number"] = form.WorkServerNumber
  611. ctx.Data["flavor_name"] = form.FlavorName
  612. ctx.Data["engine_name"] = form.EngineName
  613. ctx.Data["flavor_code"] = task.FlavorCode
  614. ctx.Data["engine_id"] = task.EngineID
  615. ctx.Data["version_name"] = form.VersionName
  616. ctx.Data["bootFile"] = form.BootFile
  617. ctx.Data["uuid"] = form.Attachment
  618. ctx.Data["branch_name"] = form.BranchName
  619. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  620. if err != nil {
  621. ctx.ServerError("getConfigList failed:", err)
  622. return err
  623. }
  624. ctx.Data["config_list"] = configList.ParaConfigs
  625. return nil
  626. }
  627. func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) {
  628. ctx.Data["PageIsTrainJob"] = true
  629. VersionOutputPath := modelarts.GetVersionOutputPathByTotalVersionCount(modelarts.TotalVersionCount)
  630. jobName := form.JobName
  631. uuid := form.Attachment
  632. description := form.Description
  633. workServerNumber := form.WorkServerNumber
  634. engineID := form.EngineID
  635. bootFile := form.BootFile
  636. flavorCode := form.Flavor
  637. params := form.Params
  638. poolID := form.PoolID
  639. isSaveParam := form.IsSaveParam
  640. repo := ctx.Repo.Repository
  641. codeLocalPath := setting.JobPath + jobName + modelarts.CodePath
  642. codeObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.CodePath
  643. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath + VersionOutputPath + "/"
  644. logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath + VersionOutputPath + "/"
  645. dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/"
  646. branch_name := form.BranchName
  647. isLatestVersion := modelarts.IsLatestVersion
  648. FlavorName := form.FlavorName
  649. VersionCount := modelarts.VersionCount
  650. EngineName := form.EngineName
  651. count, err := models.GetCloudbrainTrainJobCountByUserID(ctx.User.ID)
  652. if err != nil {
  653. log.Error("GetCloudbrainTrainJobCountByUserID failed:%v", err, ctx.Data["MsgID"])
  654. trainJobErrorNewDataPrepare(ctx, form)
  655. ctx.RenderWithErr("system error", tplModelArtsTrainJobNew, &form)
  656. return
  657. } else {
  658. if count >= 1 {
  659. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  660. trainJobErrorNewDataPrepare(ctx, form)
  661. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsTrainJobNew, &form)
  662. return
  663. }
  664. }
  665. if err := paramCheckCreateTrainJob(form); err != nil {
  666. log.Error("paramCheckCreateTrainJob failed:(%v)", err)
  667. trainJobErrorNewDataPrepare(ctx, form)
  668. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobNew, &form)
  669. return
  670. }
  671. // attach, err := models.GetAttachmentByUUID(uuid)
  672. // if err != nil {
  673. // log.Error("GetAttachmentByUUID(%s) failed:%v", uuid, err.Error())
  674. // return
  675. // }
  676. //todo: del the codeLocalPath
  677. // _, err := ioutil.ReadDir(codeLocalPath)
  678. // if err == nil {
  679. // os.RemoveAll(codeLocalPath)
  680. // }
  681. os.RemoveAll(codeLocalPath)
  682. gitRepo, _ := git.OpenRepository(repo.RepoPath())
  683. commitID, _ := gitRepo.GetBranchCommitID(branch_name)
  684. if err := git.Clone(repo.RepoPath(), codeLocalPath, git.CloneRepoOptions{
  685. Branch: branch_name,
  686. }); err != nil {
  687. log.Error("创建任务失败,服务器超时!: %s (%v)", repo.FullName(), err)
  688. trainJobErrorNewDataPrepare(ctx, form)
  689. ctx.RenderWithErr("创建任务失败,服务器超时!", tplModelArtsTrainJobNew, &form)
  690. return
  691. }
  692. //todo: upload code (send to file_server todo this work?)
  693. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.OutputPath + VersionOutputPath + "/"); err != nil {
  694. log.Error("Failed to obsMkdir_output: %s (%v)", repo.FullName(), err)
  695. trainJobErrorNewDataPrepare(ctx, form)
  696. ctx.RenderWithErr("Failed to obsMkdir_output", tplModelArtsTrainJobNew, &form)
  697. return
  698. }
  699. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.LogPath + VersionOutputPath + "/"); err != nil {
  700. log.Error("Failed to obsMkdir_log: %s (%v)", repo.FullName(), err)
  701. trainJobErrorNewDataPrepare(ctx, form)
  702. ctx.RenderWithErr("Failed to obsMkdir_log", tplModelArtsTrainJobNew, &form)
  703. return
  704. }
  705. // parentDir := VersionOutputPath + "/"
  706. if err := uploadCodeToObs(codeLocalPath, jobName, ""); err != nil {
  707. // if err := uploadCodeToObs(codeLocalPath, jobName, parentDir); err != nil {
  708. log.Error("Failed to uploadCodeToObs: %s (%v)", repo.FullName(), err)
  709. trainJobErrorNewDataPrepare(ctx, form)
  710. ctx.RenderWithErr("Failed to uploadCodeToObs", tplModelArtsTrainJobNew, &form)
  711. return
  712. }
  713. //todo: del local code?
  714. var parameters models.Parameters
  715. param := make([]models.Parameter, 0)
  716. param = append(param, models.Parameter{
  717. Label: modelarts.TrainUrl,
  718. Value: outputObsPath,
  719. }, models.Parameter{
  720. Label: modelarts.DataUrl,
  721. Value: dataPath,
  722. })
  723. if len(params) != 0 {
  724. err := json.Unmarshal([]byte(params), &parameters)
  725. if err != nil {
  726. log.Error("Failed to Unmarshal params: %s (%v)", params, err)
  727. trainJobErrorNewDataPrepare(ctx, form)
  728. ctx.RenderWithErr("运行参数错误", tplModelArtsTrainJobNew, &form)
  729. return
  730. }
  731. for _, parameter := range parameters.Parameter {
  732. if parameter.Label != modelarts.TrainUrl && parameter.Label != modelarts.DataUrl {
  733. param = append(param, models.Parameter{
  734. Label: parameter.Label,
  735. Value: parameter.Value,
  736. })
  737. }
  738. }
  739. }
  740. //save param config
  741. if isSaveParam == "on" {
  742. if form.ParameterTemplateName == "" {
  743. log.Error("ParameterTemplateName is empty")
  744. trainJobNewDataPrepare(ctx)
  745. ctx.RenderWithErr("保存作业参数时,作业参数名称不能为空", tplModelArtsTrainJobNew, &form)
  746. return
  747. }
  748. _, err := modelarts.CreateTrainJobConfig(models.CreateConfigParams{
  749. ConfigName: form.ParameterTemplateName,
  750. Description: form.PrameterDescription,
  751. DataUrl: dataPath,
  752. AppUrl: codeObsPath,
  753. BootFileUrl: codeObsPath + bootFile,
  754. TrainUrl: outputObsPath,
  755. Flavor: models.Flavor{
  756. Code: flavorCode,
  757. },
  758. WorkServerNum: workServerNumber,
  759. EngineID: int64(engineID),
  760. LogUrl: logObsPath,
  761. PoolID: poolID,
  762. Parameter: param,
  763. })
  764. if err != nil {
  765. log.Error("Failed to CreateTrainJobConfig: %v", err)
  766. trainJobErrorNewDataPrepare(ctx, form)
  767. ctx.RenderWithErr("保存作业参数失败:"+err.Error(), tplModelArtsTrainJobNew, &form)
  768. return
  769. }
  770. }
  771. req := &modelarts.GenerateTrainJobReq{
  772. JobName: jobName,
  773. DataUrl: dataPath,
  774. Description: description,
  775. CodeObsPath: codeObsPath,
  776. BootFileUrl: codeObsPath + bootFile,
  777. BootFile: bootFile,
  778. TrainUrl: outputObsPath,
  779. FlavorCode: flavorCode,
  780. WorkServerNumber: workServerNumber,
  781. EngineID: int64(engineID),
  782. LogUrl: logObsPath,
  783. PoolID: poolID,
  784. Uuid: uuid,
  785. Parameters: parameters.Parameter,
  786. CommitID: commitID,
  787. IsLatestVersion: isLatestVersion,
  788. BranchName: branch_name,
  789. Params: form.Params,
  790. FlavorName: FlavorName,
  791. EngineName: EngineName,
  792. VersionCount: VersionCount,
  793. TotalVersionCount: modelarts.TotalVersionCount,
  794. }
  795. //将params转换Parameters.Parameter,出错时返回给前端
  796. var Parameters modelarts.Parameters
  797. if err := json.Unmarshal([]byte(params), &Parameters); err != nil {
  798. ctx.ServerError("json.Unmarshal failed:", err)
  799. return
  800. }
  801. err = modelarts.GenerateTrainJob(ctx, req)
  802. if err != nil {
  803. log.Error("GenerateTrainJob failed:%v", err.Error())
  804. trainJobErrorNewDataPrepare(ctx, form)
  805. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobNew, &form)
  806. return
  807. }
  808. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  809. }
  810. func TrainJobCreateVersion(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) {
  811. ctx.Data["PageIsTrainJob"] = true
  812. var jobID = ctx.Params(":jobid")
  813. count, err := models.GetCloudbrainTrainJobCountByUserID(ctx.User.ID)
  814. if err != nil {
  815. log.Error("GetCloudbrainTrainJobCountByUserID failed:%v", err, ctx.Data["MsgID"])
  816. versionErrorDataPrepare(ctx, form)
  817. ctx.RenderWithErr("system error", tplModelArtsTrainJobVersionNew, &form)
  818. return
  819. } else {
  820. if count >= 1 {
  821. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  822. versionErrorDataPrepare(ctx, form)
  823. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsTrainJobVersionNew, &form)
  824. return
  825. }
  826. }
  827. latestTask, err := models.GetCloudbrainByJobIDAndIsLatestVersion(jobID, modelarts.IsLatestVersion)
  828. if err != nil {
  829. ctx.ServerError("GetCloudbrainByJobIDAndIsLatestVersion faild:", err)
  830. return
  831. }
  832. VersionOutputPath := modelarts.GetVersionOutputPathByTotalVersionCount(latestTask.TotalVersionCount + 1)
  833. jobName := form.JobName
  834. uuid := form.Attachment
  835. description := form.Description
  836. workServerNumber := form.WorkServerNumber
  837. engineID := form.EngineID
  838. bootFile := form.BootFile
  839. flavorCode := form.Flavor
  840. params := form.Params
  841. poolID := form.PoolID
  842. isSaveParam := form.IsSaveParam
  843. repo := ctx.Repo.Repository
  844. codeLocalPath := setting.JobPath + jobName + modelarts.CodePath
  845. codeObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.CodePath + VersionOutputPath + "/"
  846. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath + VersionOutputPath + "/"
  847. logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath + VersionOutputPath + "/"
  848. dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/"
  849. branch_name := form.BranchName
  850. PreVersionName := form.VersionName
  851. FlavorName := form.FlavorName
  852. EngineName := form.EngineName
  853. isLatestVersion := modelarts.IsLatestVersion
  854. //判断权限
  855. canNewJob, _ := canUserCreateTrainJobVersion(ctx, latestTask.UserID)
  856. if !canNewJob {
  857. ctx.RenderWithErr("user cann't new trainjob", tplModelArtsTrainJobVersionNew, &form)
  858. return
  859. }
  860. if err := paramCheckCreateTrainJob(form); err != nil {
  861. log.Error("paramCheckCreateTrainJob failed:(%v)", err)
  862. versionErrorDataPrepare(ctx, form)
  863. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  864. return
  865. }
  866. // attach, err := models.GetAttachmentByUUID(uuid)
  867. // if err != nil {
  868. // log.Error("GetAttachmentByUUID(%s) failed:%v", uuid, err.Error())
  869. // return
  870. // }
  871. //todo: del the codeLocalPath
  872. // _, err = ioutil.ReadDir(codeLocalPath)
  873. // if err == nil {
  874. // os.RemoveAll(codeLocalPath)
  875. // }
  876. os.RemoveAll(codeLocalPath)
  877. gitRepo, _ := git.OpenRepository(repo.RepoPath())
  878. commitID, _ := gitRepo.GetBranchCommitID(branch_name)
  879. if err := git.Clone(repo.RepoPath(), codeLocalPath, git.CloneRepoOptions{
  880. Branch: branch_name,
  881. }); err != nil {
  882. log.Error("创建任务失败,任务名称已存在!: %s (%v)", repo.FullName(), err)
  883. versionErrorDataPrepare(ctx, form)
  884. ctx.RenderWithErr("创建任务失败,任务名称已存在!", tplModelArtsTrainJobVersionNew, &form)
  885. return
  886. }
  887. //todo: upload code (send to file_server todo this work?)
  888. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.OutputPath + VersionOutputPath + "/"); err != nil {
  889. log.Error("Failed to obsMkdir_output: %s (%v)", repo.FullName(), err)
  890. versionErrorDataPrepare(ctx, form)
  891. ctx.RenderWithErr("Failed to obsMkdir_output", tplModelArtsTrainJobVersionNew, &form)
  892. return
  893. }
  894. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.LogPath + VersionOutputPath + "/"); err != nil {
  895. log.Error("Failed to obsMkdir_log: %s (%v)", repo.FullName(), err)
  896. versionErrorDataPrepare(ctx, form)
  897. ctx.RenderWithErr("Failed to obsMkdir_log", tplModelArtsTrainJobVersionNew, &form)
  898. return
  899. }
  900. parentDir := VersionOutputPath + "/"
  901. // parentDir := ""
  902. // if err := uploadCodeToObs(codeLocalPath, jobName, ""); err != nil {
  903. if err := uploadCodeToObs(codeLocalPath, jobName, parentDir); err != nil {
  904. log.Error("Failed to uploadCodeToObs: %s (%v)", repo.FullName(), err)
  905. versionErrorDataPrepare(ctx, form)
  906. ctx.RenderWithErr("Failed to uploadCodeToObs", tplModelArtsTrainJobVersionNew, &form)
  907. return
  908. }
  909. //todo: del local code?
  910. var parameters models.Parameters
  911. param := make([]models.Parameter, 0)
  912. param = append(param, models.Parameter{
  913. Label: modelarts.TrainUrl,
  914. Value: outputObsPath,
  915. }, models.Parameter{
  916. Label: modelarts.DataUrl,
  917. Value: dataPath,
  918. })
  919. if len(params) != 0 {
  920. err := json.Unmarshal([]byte(params), &parameters)
  921. if err != nil {
  922. log.Error("Failed to Unmarshal params: %s (%v)", params, err)
  923. versionErrorDataPrepare(ctx, form)
  924. ctx.RenderWithErr("运行参数错误", tplModelArtsTrainJobVersionNew, &form)
  925. return
  926. }
  927. for _, parameter := range parameters.Parameter {
  928. if parameter.Label != modelarts.TrainUrl && parameter.Label != modelarts.DataUrl {
  929. param = append(param, models.Parameter{
  930. Label: parameter.Label,
  931. Value: parameter.Value,
  932. })
  933. }
  934. }
  935. }
  936. //save param config
  937. if isSaveParam == "on" {
  938. if form.ParameterTemplateName == "" {
  939. log.Error("ParameterTemplateName is empty")
  940. versionErrorDataPrepare(ctx, form)
  941. ctx.RenderWithErr("保存作业参数时,作业参数名称不能为空", tplModelArtsTrainJobVersionNew, &form)
  942. return
  943. }
  944. _, err := modelarts.CreateTrainJobConfig(models.CreateConfigParams{
  945. ConfigName: form.ParameterTemplateName,
  946. Description: form.PrameterDescription,
  947. DataUrl: dataPath,
  948. AppUrl: codeObsPath,
  949. BootFileUrl: codeObsPath + bootFile,
  950. TrainUrl: outputObsPath,
  951. Flavor: models.Flavor{
  952. Code: flavorCode,
  953. },
  954. WorkServerNum: workServerNumber,
  955. EngineID: int64(engineID),
  956. LogUrl: logObsPath,
  957. PoolID: poolID,
  958. Parameter: parameters.Parameter,
  959. })
  960. if err != nil {
  961. log.Error("Failed to CreateTrainJobConfig: %v", err)
  962. versionErrorDataPrepare(ctx, form)
  963. ctx.RenderWithErr("保存作业参数失败:"+err.Error(), tplModelArtsTrainJobVersionNew, &form)
  964. return
  965. }
  966. }
  967. if err != nil {
  968. log.Error("getFlavorNameByEngineID(%s) failed:%v", engineID, err.Error())
  969. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  970. return
  971. }
  972. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, PreVersionName)
  973. if err != nil {
  974. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  975. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  976. return
  977. }
  978. req := &modelarts.GenerateTrainJobReq{
  979. JobName: task.JobName,
  980. DataUrl: dataPath,
  981. Description: description,
  982. CodeObsPath: codeObsPath,
  983. BootFileUrl: codeObsPath + bootFile,
  984. BootFile: bootFile,
  985. TrainUrl: outputObsPath,
  986. FlavorCode: flavorCode,
  987. WorkServerNumber: workServerNumber,
  988. IsLatestVersion: isLatestVersion,
  989. EngineID: int64(engineID),
  990. LogUrl: logObsPath,
  991. PoolID: poolID,
  992. Uuid: uuid,
  993. Params: form.Params,
  994. Parameters: parameters.Parameter,
  995. PreVersionId: task.VersionID,
  996. CommitID: commitID,
  997. BranchName: branch_name,
  998. FlavorName: FlavorName,
  999. EngineName: EngineName,
  1000. PreVersionName: PreVersionName,
  1001. TotalVersionCount: latestTask.TotalVersionCount + 1,
  1002. }
  1003. err = modelarts.GenerateTrainJobVersion(ctx, req, jobID)
  1004. if err != nil {
  1005. log.Error("GenerateTrainJob failed:%v", err.Error())
  1006. versionErrorDataPrepare(ctx, form)
  1007. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  1008. return
  1009. }
  1010. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job/" + jobID)
  1011. // ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1012. }
  1013. // readDir reads the directory named by dirname and returns
  1014. // a list of directory entries sorted by filename.
  1015. func readDir(dirname string) ([]os.FileInfo, error) {
  1016. f, err := os.Open(dirname)
  1017. if err != nil {
  1018. return nil, err
  1019. }
  1020. list, err := f.Readdir(100)
  1021. f.Close()
  1022. if err != nil {
  1023. //todo: can not upload empty folder
  1024. if err == io.EOF {
  1025. return nil, nil
  1026. }
  1027. return nil, err
  1028. }
  1029. //sort.Slice(list, func(i, j int) bool { return list[i].Name() < list[j].Name() })
  1030. return list, nil
  1031. }
  1032. func uploadCodeToObs(codePath, jobName, parentDir string) error {
  1033. files, err := readDir(codePath)
  1034. if err != nil {
  1035. log.Error("readDir(%s) failed: %s", codePath, err.Error())
  1036. return err
  1037. }
  1038. for _, file := range files {
  1039. if file.IsDir() {
  1040. input := &obs.PutObjectInput{}
  1041. input.Bucket = setting.Bucket
  1042. input.Key = parentDir + file.Name() + "/"
  1043. _, err = storage.ObsCli.PutObject(input)
  1044. if err != nil {
  1045. log.Error("PutObject(%s) failed: %s", input.Key, err.Error())
  1046. return err
  1047. }
  1048. if err = uploadCodeToObs(codePath+file.Name()+"/", jobName, parentDir+file.Name()+"/"); err != nil {
  1049. log.Error("uploadCodeToObs(%s) failed: %s", file.Name(), err.Error())
  1050. return err
  1051. }
  1052. } else {
  1053. input := &obs.PutFileInput{}
  1054. input.Bucket = setting.Bucket
  1055. input.Key = setting.CodePathPrefix + jobName + "/code/" + parentDir + file.Name()
  1056. input.SourceFile = codePath + file.Name()
  1057. _, err = storage.ObsCli.PutFile(input)
  1058. if err != nil {
  1059. log.Error("PutFile(%s) failed: %s", input.SourceFile, err.Error())
  1060. return err
  1061. }
  1062. }
  1063. }
  1064. return nil
  1065. }
  1066. func obsMkdir(dir string) error {
  1067. input := &obs.PutObjectInput{}
  1068. input.Bucket = setting.Bucket
  1069. input.Key = dir
  1070. _, err := storage.ObsCli.PutObject(input)
  1071. if err != nil {
  1072. log.Error("PutObject(%s) failed: %s", input.Key, err.Error())
  1073. return err
  1074. }
  1075. return nil
  1076. }
  1077. func paramCheckCreateTrainJob(form auth.CreateModelArtsTrainJobForm) error {
  1078. if !strings.HasSuffix(form.BootFile, ".py") {
  1079. log.Error("the boot file(%s) must be a python file", form.BootFile)
  1080. return errors.New("启动文件必须是python文件")
  1081. }
  1082. if form.WorkServerNumber > 25 || form.WorkServerNumber < 1 {
  1083. log.Error("the WorkServerNumber(%d) must be in (1,25)", form.WorkServerNumber)
  1084. return errors.New("计算节点数必须在1-25之间")
  1085. }
  1086. return nil
  1087. }
  1088. func TrainJobShow(ctx *context.Context) {
  1089. ctx.Data["PageIsCloudBrain"] = true
  1090. var jobID = ctx.Params(":jobid")
  1091. repo := ctx.Repo.Repository
  1092. page := ctx.QueryInt("page")
  1093. if page <= 0 {
  1094. page = 1
  1095. }
  1096. VersionListTasks, VersionListCount, err := models.CloudbrainsVersionList(&models.CloudbrainsOptions{
  1097. ListOptions: models.ListOptions{
  1098. Page: page,
  1099. PageSize: setting.UI.IssuePagingNum,
  1100. },
  1101. RepoID: repo.ID,
  1102. Type: models.TypeCloudBrainTwo,
  1103. JobType: string(models.JobTypeTrain),
  1104. JobID: jobID,
  1105. })
  1106. if err != nil {
  1107. log.Error("GetVersionListTasks(%s) failed:%v", jobID, err.Error())
  1108. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1109. return
  1110. }
  1111. //设置权限
  1112. canNewJob, err := canUserCreateTrainJobVersion(ctx, VersionListTasks[0].UserID)
  1113. if err != nil {
  1114. ctx.ServerError("canNewJob failed", err)
  1115. return
  1116. }
  1117. ctx.Data["canNewJob"] = canNewJob
  1118. //将运行参数转化为epoch_size = 3, device_target = Ascend的格式
  1119. for i, _ := range VersionListTasks {
  1120. var parameters models.Parameters
  1121. err := json.Unmarshal([]byte(VersionListTasks[i].Parameters), &parameters)
  1122. if err != nil {
  1123. log.Error("Failed to Unmarshal Parameters: %s (%v)", VersionListTasks[i].Parameters, err)
  1124. trainJobNewDataPrepare(ctx)
  1125. return
  1126. }
  1127. if len(parameters.Parameter) > 0 {
  1128. paramTemp := ""
  1129. for _, Parameter := range parameters.Parameter {
  1130. param := Parameter.Label + " = " + Parameter.Value + "; "
  1131. paramTemp = paramTemp + param
  1132. }
  1133. VersionListTasks[i].Parameters = paramTemp[:len(paramTemp)-2]
  1134. } else {
  1135. VersionListTasks[i].Parameters = ""
  1136. }
  1137. }
  1138. pager := context.NewPagination(VersionListCount, setting.UI.IssuePagingNum, page, 5)
  1139. pager.SetDefaultParams(ctx)
  1140. ctx.Data["Page"] = pager
  1141. ctx.Data["jobID"] = jobID
  1142. ctx.Data["jobName"] = VersionListTasks[0].JobName
  1143. ctx.Data["version_list_task"] = VersionListTasks
  1144. ctx.Data["version_list_count"] = VersionListCount
  1145. ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1146. }
  1147. func TrainJobGetLog(ctx *context.Context) {
  1148. ctx.Data["PageIsTrainJob"] = true
  1149. var jobID = ctx.Params(":jobid")
  1150. var logFileName = ctx.Query("file_name")
  1151. var baseLine = ctx.Query("base_line")
  1152. var order = ctx.Query("order")
  1153. if order != modelarts.OrderDesc && order != modelarts.OrderAsc {
  1154. log.Error("order(%s) check failed", order)
  1155. ctx.HTML(http.StatusBadRequest, tplModelArtsTrainJobShow)
  1156. return
  1157. }
  1158. task, err := models.GetCloudbrainByJobID(jobID)
  1159. if err != nil {
  1160. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  1161. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1162. return
  1163. }
  1164. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), baseLine, logFileName, order, modelarts.Lines)
  1165. if err != nil {
  1166. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  1167. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1168. return
  1169. }
  1170. ctx.Data["log"] = result
  1171. //ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1172. }
  1173. func trainJobGetLog(jobID string) (*models.GetTrainJobLogFileNamesResult, *models.GetTrainJobLogResult, error) {
  1174. task, err := models.GetCloudbrainByJobID(jobID)
  1175. if err != nil {
  1176. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  1177. return nil, nil, err
  1178. }
  1179. resultLogFile, err := modelarts.GetTrainJobLogFileNames(jobID, strconv.FormatInt(task.VersionID, 10))
  1180. if err != nil {
  1181. log.Error("GetTrainJobLogFileNames(%s) failed:%v", jobID, err.Error())
  1182. return nil, nil, err
  1183. }
  1184. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), "", resultLogFile.LogFileList[0], modelarts.OrderDesc, modelarts.Lines)
  1185. if err != nil {
  1186. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  1187. return nil, nil, err
  1188. }
  1189. return resultLogFile, result, err
  1190. }
  1191. func TrainJobDel(ctx *context.Context) {
  1192. var jobID = ctx.Params(":jobid")
  1193. repo := ctx.Repo.Repository
  1194. VersionListTasks, _, err := models.CloudbrainsVersionList(&models.CloudbrainsOptions{
  1195. RepoID: repo.ID,
  1196. Type: models.TypeCloudBrainTwo,
  1197. JobType: string(models.JobTypeTrain),
  1198. JobID: jobID,
  1199. })
  1200. if err != nil {
  1201. ctx.ServerError("get VersionListTasks failed", err)
  1202. return
  1203. }
  1204. //删除modelarts上的任务记录
  1205. _, err = modelarts.DelTrainJob(jobID)
  1206. if err != nil {
  1207. log.Error("DelTrainJob(%s) failed:%v", jobID, err.Error())
  1208. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil)
  1209. return
  1210. }
  1211. //删除数据库Cloudbrain表的记录
  1212. for _, task := range VersionListTasks {
  1213. err = models.DeleteJob(&task.Cloudbrain)
  1214. if err != nil {
  1215. ctx.ServerError("DeleteJob failed", err)
  1216. return
  1217. }
  1218. }
  1219. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  1220. }
  1221. func TrainJobStop(ctx *context.Context) {
  1222. var jobID = ctx.Params(":jobid")
  1223. task, err := models.GetCloudbrainByJobID(jobID)
  1224. if err != nil {
  1225. log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
  1226. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil)
  1227. return
  1228. }
  1229. _, err = modelarts.StopTrainJob(jobID, strconv.FormatInt(task.VersionID, 10))
  1230. if err != nil {
  1231. log.Error("StopTrainJob(%s) failed:%v", task.JobName, err.Error())
  1232. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil)
  1233. return
  1234. }
  1235. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  1236. }
  1237. func canUserCreateTrainJob(uid int64) (bool, error) {
  1238. org, err := models.GetOrgByName(setting.AllowedOrg)
  1239. if err != nil {
  1240. log.Error("get allowed org failed: ", setting.AllowedOrg)
  1241. return false, err
  1242. }
  1243. return org.IsOrgMember(uid)
  1244. }
  1245. func canUserCreateTrainJobVersion(ctx *context.Context, userID int64) (bool, error) {
  1246. if ctx == nil || ctx.User == nil {
  1247. log.Error("user unlogin!")
  1248. return false, nil
  1249. }
  1250. if userID == ctx.User.ID || ctx.User.IsAdmin {
  1251. return true, nil
  1252. } else {
  1253. log.Error("Only user itself and admin can new trainjob!")
  1254. return false, nil
  1255. }
  1256. }
  1257. func TrainJobGetConfigList(ctx *context.Context) {
  1258. ctx.Data["PageIsTrainJob"] = true
  1259. var jobID = ctx.Params(":jobid")
  1260. var logFileName = ctx.Query("file_name")
  1261. var baseLine = ctx.Query("base_line")
  1262. var order = ctx.Query("order")
  1263. if order != modelarts.OrderDesc && order != modelarts.OrderAsc {
  1264. log.Error("order(%s) check failed", order)
  1265. ctx.HTML(http.StatusBadRequest, tplModelArtsTrainJobShow)
  1266. return
  1267. }
  1268. task, err := models.GetCloudbrainByJobID(jobID)
  1269. if err != nil {
  1270. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  1271. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1272. return
  1273. }
  1274. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), baseLine, logFileName, order, modelarts.Lines)
  1275. if err != nil {
  1276. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  1277. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1278. return
  1279. }
  1280. ctx.Data["log"] = result
  1281. //ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1282. }
  1283. func getConfigList(perPage, page int, sortBy, order, searchContent, configType string) (*models.GetConfigListResult, error) {
  1284. var result models.GetConfigListResult
  1285. list, err := modelarts.GetConfigList(perPage, page, sortBy, order, searchContent, configType)
  1286. if err != nil {
  1287. log.Error("GetConfigList failed:", err)
  1288. return &result, err
  1289. }
  1290. for _, config := range list.ParaConfigs {
  1291. paraConfig, err := modelarts.GetParaConfig(config.ConfigName, configType)
  1292. if err != nil {
  1293. log.Error("GetParaConfig failed:", err)
  1294. return &result, err
  1295. }
  1296. config.Result = paraConfig
  1297. }
  1298. return list, nil
  1299. }
  1300. func ModelDownload(ctx *context.Context) {
  1301. var (
  1302. err error
  1303. )
  1304. var jobID = ctx.Params(":jobid")
  1305. versionName := ctx.Query("version_name")
  1306. parentDir := ctx.Query("parent_dir")
  1307. fileName := ctx.Query("file_name")
  1308. log.Info("DownloadSingleModelFile start.")
  1309. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  1310. if err != nil {
  1311. log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
  1312. return
  1313. }
  1314. path := strings.TrimPrefix(path.Join(setting.TrainJobModelPath, task.JobName, setting.OutPutPath, versionName, parentDir, fileName), "/")
  1315. log.Info("Download path is:%s", path)
  1316. url, err := storage.GetObsCreateSignedUrlByBucketAndKey(setting.Bucket, path)
  1317. if err != nil {
  1318. log.Error("GetObsCreateSignedUrl failed: %v", err.Error(), ctx.Data["msgID"])
  1319. ctx.ServerError("GetObsCreateSignedUrl", err)
  1320. return
  1321. }
  1322. http.Redirect(ctx.Resp, ctx.Req.Request, url, http.StatusMovedPermanently)
  1323. }