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 50 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
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 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
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
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
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 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
4 years ago
4 years ago
4 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
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
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
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
3 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
4 years ago
4 years ago
3 years ago
4 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
4 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
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
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
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
4 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
3 years ago
3 years ago
3 years ago
3 years ago
3 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
3 years ago
4 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
4 years ago
4 years ago
3 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
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
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
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
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
4 years ago
4 years ago
3 years ago
3 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
4 years ago
4 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
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
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
3 years ago
4 years ago
3 years ago
3 years ago
4 years ago
3 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
4 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
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565
  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/models"
  13. "code.gitea.io/gitea/modules/auth"
  14. "code.gitea.io/gitea/modules/base"
  15. "code.gitea.io/gitea/modules/cloudbrain"
  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. ctx.Data["ListType"] = debugListType
  38. MustEnableCloudbrain(ctx)
  39. repo := ctx.Repo.Repository
  40. page := ctx.QueryInt("page")
  41. if page <= 0 {
  42. page = 1
  43. }
  44. debugType := modelarts.DebugType
  45. if debugListType == models.GPUResource {
  46. debugType = models.TypeCloudBrainOne
  47. } else if debugListType == models.NPUResource {
  48. debugType = models.TypeCloudBrainTwo
  49. }
  50. ciTasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  51. ListOptions: models.ListOptions{
  52. Page: page,
  53. PageSize: setting.UI.IssuePagingNum,
  54. },
  55. RepoID: repo.ID,
  56. Type: debugType,
  57. JobTypeNot: true,
  58. JobType: string(models.JobTypeTrain),
  59. })
  60. if err != nil {
  61. ctx.ServerError("Get debugjob faild:", err)
  62. return
  63. }
  64. for i, task := range ciTasks {
  65. ciTasks[i].CanDebug = cloudbrain.CanModifyJob(ctx, &task.Cloudbrain)
  66. ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  67. ciTasks[i].Cloudbrain.ComputeResource = task.ComputeResource
  68. }
  69. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  70. pager.AddParam(ctx, "debugListType", "ListType")
  71. ctx.Data["Page"] = pager
  72. ctx.Data["PageIsCloudBrain"] = true
  73. ctx.Data["Tasks"] = ciTasks
  74. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  75. ctx.Data["RepoIsEmpty"] = repo.IsEmpty
  76. ctx.HTML(200, tplDebugJobIndex)
  77. }
  78. // MustEnableDataset check if repository enable internal cb
  79. func MustEnableModelArts(ctx *context.Context) {
  80. if !ctx.Repo.CanRead(models.UnitTypeCloudBrain) {
  81. ctx.NotFound("MustEnableCloudbrain", nil)
  82. return
  83. }
  84. }
  85. func NotebookNew(ctx *context.Context) {
  86. ctx.Data["PageIsCloudBrain"] = true
  87. t := time.Now()
  88. var jobName = jobNamePrefixValid(cutString(ctx.User.Name, 5)) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  89. ctx.Data["job_name"] = jobName
  90. attachs, err := models.GetModelArtsUserAttachments(ctx.User.ID)
  91. if err != nil {
  92. ctx.ServerError("GetAllUserAttachments failed:", err)
  93. return
  94. }
  95. ctx.Data["attachments"] = attachs
  96. ctx.Data["dataset_path"] = modelarts.DataSetMountPath
  97. ctx.Data["env"] = modelarts.NotebookEnv
  98. ctx.Data["notebook_type"] = modelarts.NotebookType
  99. if modelarts.FlavorInfos == nil {
  100. json.Unmarshal([]byte(setting.FlavorInfos), &modelarts.FlavorInfos)
  101. }
  102. ctx.Data["flavors"] = modelarts.FlavorInfos.FlavorInfo
  103. ctx.HTML(200, tplModelArtsNotebookNew)
  104. }
  105. func NotebookCreate(ctx *context.Context, form auth.CreateModelArtsNotebookForm) {
  106. ctx.Data["PageIsNotebook"] = true
  107. jobName := form.JobName
  108. uuid := form.Attachment
  109. description := form.Description
  110. flavor := form.Flavor
  111. count, err := models.GetCloudbrainNotebookCountByUserID(ctx.User.ID)
  112. if err != nil {
  113. log.Error("GetCloudbrainNotebookCountByUserID failed:%v", err, ctx.Data["MsgID"])
  114. cloudBrainNewDataPrepare(ctx)
  115. ctx.RenderWithErr("system error", tplModelArtsNotebookNew, &form)
  116. return
  117. } else {
  118. if count >= 1 {
  119. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  120. cloudBrainNewDataPrepare(ctx)
  121. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsNotebookNew, &form)
  122. return
  123. }
  124. }
  125. _, err = models.GetCloudbrainByName(jobName)
  126. if err == nil {
  127. log.Error("the job name did already exist", ctx.Data["MsgID"])
  128. cloudBrainNewDataPrepare(ctx)
  129. ctx.RenderWithErr("the job name did already exist", tplModelArtsNotebookNew, &form)
  130. return
  131. } else {
  132. if !models.IsErrJobNotExist(err) {
  133. log.Error("system error, %v", err, ctx.Data["MsgID"])
  134. cloudBrainNewDataPrepare(ctx)
  135. ctx.RenderWithErr("system error", tplModelArtsNotebookNew, &form)
  136. return
  137. }
  138. }
  139. err = modelarts.GenerateTask(ctx, jobName, uuid, description, flavor)
  140. if err != nil {
  141. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookNew, &form)
  142. return
  143. }
  144. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=all")
  145. }
  146. func NotebookShow(ctx *context.Context) {
  147. ctx.Data["PageIsCloudBrain"] = true
  148. var jobID = ctx.Params(":jobid")
  149. task, err := models.GetCloudbrainByJobID(jobID)
  150. if err != nil {
  151. ctx.Data["error"] = err.Error()
  152. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil)
  153. return
  154. }
  155. result, err := modelarts.GetJob(jobID)
  156. if err != nil {
  157. ctx.Data["error"] = err.Error()
  158. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil)
  159. return
  160. }
  161. if result != nil {
  162. task.Status = result.Status
  163. err = models.UpdateJob(task)
  164. if err != nil {
  165. ctx.Data["error"] = err.Error()
  166. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil)
  167. return
  168. }
  169. createTime, _ := com.StrTo(result.CreationTimestamp).Int64()
  170. result.CreateTime = time.Unix(int64(createTime/1000), 0).Format("2006-01-02 15:04:05")
  171. endTime, _ := com.StrTo(result.LatestUpdateTimestamp).Int64()
  172. result.LatestUpdateTime = time.Unix(int64(endTime/1000), 0).Format("2006-01-02 15:04:05")
  173. result.QueuingInfo.BeginTime = time.Unix(int64(result.QueuingInfo.BeginTimestamp/1000), 0).Format("2006-01-02 15:04:05")
  174. result.QueuingInfo.EndTime = time.Unix(int64(result.QueuingInfo.EndTimestamp/1000), 0).Format("2006-01-02 15:04:05")
  175. }
  176. ctx.Data["task"] = task
  177. ctx.Data["jobID"] = jobID
  178. ctx.Data["result"] = result
  179. ctx.HTML(200, tplModelArtsNotebookShow)
  180. }
  181. func NotebookDebug(ctx *context.Context) {
  182. var jobID = ctx.Params(":jobid")
  183. result, err := modelarts.GetJob(jobID)
  184. if err != nil {
  185. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil)
  186. return
  187. }
  188. res, err := modelarts.GetJobToken(jobID)
  189. if err != nil {
  190. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil)
  191. return
  192. }
  193. urls := strings.Split(result.Spec.Annotations.Url, "/")
  194. urlPrefix := result.Spec.Annotations.TargetDomain
  195. for i, url := range urls {
  196. if i > 2 {
  197. urlPrefix += "/" + url
  198. }
  199. }
  200. debugUrl := urlPrefix + "?token=" + res.Token
  201. ctx.Redirect(debugUrl)
  202. }
  203. func NotebookManage(ctx *context.Context) {
  204. var jobID = ctx.Params(":jobid")
  205. var action = ctx.Params(":action")
  206. var resultCode = "0"
  207. var errorMsg = ""
  208. var status = ""
  209. for {
  210. task, err := models.GetCloudbrainByJobID(jobID)
  211. if err != nil {
  212. log.Error("GetCloudbrainByJobID failed:%v", err, ctx.Data["MsgID"])
  213. resultCode = "-1"
  214. errorMsg = "system error"
  215. break
  216. }
  217. if action == models.ActionStop {
  218. if task.Status != string(models.ModelArtsRunning) {
  219. log.Error("the job(%s) is not running", task.JobName, ctx.Data["MsgID"])
  220. resultCode = "-1"
  221. errorMsg = "the job is not running"
  222. break
  223. }
  224. if !ctx.IsSigned || (ctx.User.ID != task.UserID && !ctx.IsUserSiteAdmin() && !ctx.IsUserRepoOwner()) {
  225. log.Error("the user has no right ro stop the job", task.JobName, ctx.Data["MsgID"])
  226. resultCode = "-1"
  227. errorMsg = "you have no right to stop the job"
  228. break
  229. }
  230. } else if action == models.ActionRestart {
  231. if task.Status != string(models.ModelArtsStopped) && task.Status != string(models.ModelArtsStartFailed) && task.Status != string(models.ModelArtsCreateFailed) {
  232. log.Error("the job(%s) is not stopped", task.JobName, ctx.Data["MsgID"])
  233. resultCode = "-1"
  234. errorMsg = "the job is not stopped"
  235. break
  236. }
  237. if !ctx.IsSigned || (ctx.User.ID != task.UserID && !ctx.IsUserSiteAdmin()) {
  238. log.Error("the user has no right ro restart the job", task.JobName, ctx.Data["MsgID"])
  239. resultCode = "-1"
  240. errorMsg = "you have no right to restart the job"
  241. break
  242. }
  243. count, err := models.GetCloudbrainNotebookCountByUserID(ctx.User.ID)
  244. if err != nil {
  245. log.Error("GetCloudbrainNotebookCountByUserID failed:%v", err, ctx.Data["MsgID"])
  246. resultCode = "-1"
  247. errorMsg = "system error"
  248. break
  249. } else {
  250. if count >= 1 {
  251. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  252. resultCode = "-1"
  253. errorMsg = "you have already a running or waiting task, can not create more"
  254. break
  255. }
  256. }
  257. action = models.ActionStart
  258. } else {
  259. log.Error("the action(%s) is illegal", action, ctx.Data["MsgID"])
  260. resultCode = "-1"
  261. errorMsg = "非法操作"
  262. break
  263. }
  264. param := models.NotebookAction{
  265. Action: action,
  266. }
  267. res, err := modelarts.ManageNotebook(jobID, param)
  268. if err != nil {
  269. log.Error("ManageNotebook(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"])
  270. resultCode = "-1"
  271. errorMsg = "启动失败"
  272. break
  273. }
  274. task.Status = res.CurrentStatus
  275. err = models.UpdateJob(task)
  276. if err != nil {
  277. log.Error("UpdateJob(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"])
  278. resultCode = "-1"
  279. errorMsg = "system error"
  280. break
  281. }
  282. status = task.Status
  283. break
  284. }
  285. ctx.JSON(200, map[string]string{
  286. "result_code": resultCode,
  287. "error_msg": errorMsg,
  288. "status": status,
  289. "job_id": jobID,
  290. })
  291. }
  292. func NotebookDel(ctx *context.Context) {
  293. var jobID = ctx.Params(":jobid")
  294. task := ctx.Cloudbrain
  295. if task.Status != string(models.ModelArtsCreateFailed) && task.Status != string(models.ModelArtsStartFailed) && task.Status != string(models.ModelArtsStopped) {
  296. log.Error("the job(%s) has not been stopped", task.JobName)
  297. ctx.ServerError("the job has not been stopped", errors.New("the job has not been stopped"))
  298. return
  299. }
  300. _, err := modelarts.DelNotebook(jobID)
  301. if err != nil {
  302. log.Error("DelJob(%s) failed:%v", task.JobName, err.Error())
  303. ctx.ServerError("DelJob failed", err)
  304. return
  305. }
  306. err = models.DeleteJob(task)
  307. if err != nil {
  308. ctx.ServerError("DeleteJob failed", err)
  309. return
  310. }
  311. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=all")
  312. }
  313. func TrainJobIndex(ctx *context.Context) {
  314. MustEnableModelArts(ctx)
  315. repo := ctx.Repo.Repository
  316. page := ctx.QueryInt("page")
  317. if page <= 0 {
  318. page = 1
  319. }
  320. tasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  321. ListOptions: models.ListOptions{
  322. Page: page,
  323. PageSize: setting.UI.IssuePagingNum,
  324. },
  325. RepoID: repo.ID,
  326. Type: models.TypeCloudBrainTwo,
  327. JobTypeNot: false,
  328. JobType: string(models.JobTypeTrain),
  329. IsLatestVersion: modelarts.IsLatestVersion,
  330. })
  331. if err != nil {
  332. ctx.ServerError("Cloudbrain", err)
  333. return
  334. }
  335. for i, task := range tasks {
  336. tasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  337. tasks[i].CanModify = cloudbrain.CanModifyJob(ctx, &task.Cloudbrain)
  338. tasks[i].ComputeResource = models.NPUResource
  339. }
  340. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  341. pager.SetDefaultParams(ctx)
  342. ctx.Data["Page"] = pager
  343. ctx.Data["PageIsCloudBrain"] = true
  344. ctx.Data["Tasks"] = tasks
  345. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  346. ctx.Data["RepoIsEmpty"] = repo.IsEmpty
  347. ctx.HTML(200, tplModelArtsTrainJobIndex)
  348. }
  349. func TrainJobNew(ctx *context.Context) {
  350. err := trainJobNewDataPrepare(ctx)
  351. if err != nil {
  352. ctx.ServerError("get new train-job info failed", err)
  353. return
  354. }
  355. ctx.HTML(200, tplModelArtsTrainJobNew)
  356. }
  357. func trainJobNewDataPrepare(ctx *context.Context) error {
  358. ctx.Data["PageIsCloudBrain"] = true
  359. //can, err := canUserCreateTrainJob(ctx.User.ID)
  360. //if err != nil {
  361. // ctx.ServerError("canUserCreateTrainJob", err)
  362. // return
  363. //}
  364. //
  365. //if !can {
  366. // log.Error("the user can not create train-job")
  367. // ctx.ServerError("the user can not create train-job", fmt.Errorf("the user can not create train-job"))
  368. // return
  369. //}
  370. t := time.Now()
  371. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  372. ctx.Data["job_name"] = jobName
  373. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  374. if err != nil {
  375. ctx.ServerError("GetAllUserAttachments failed:", err)
  376. return err
  377. }
  378. ctx.Data["attachments"] = attachs
  379. var resourcePools modelarts.ResourcePool
  380. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  381. ctx.ServerError("json.Unmarshal failed:", err)
  382. return err
  383. }
  384. ctx.Data["resource_pools"] = resourcePools.Info
  385. var engines modelarts.Engine
  386. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  387. ctx.ServerError("json.Unmarshal failed:", err)
  388. return err
  389. }
  390. ctx.Data["engines"] = engines.Info
  391. var versionInfos modelarts.VersionInfo
  392. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  393. ctx.ServerError("json.Unmarshal failed:", err)
  394. return err
  395. }
  396. ctx.Data["engine_versions"] = versionInfos.Version
  397. var flavorInfos modelarts.Flavor
  398. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  399. ctx.ServerError("json.Unmarshal failed:", err)
  400. return err
  401. }
  402. ctx.Data["flavor_infos"] = flavorInfos.Info
  403. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  404. ctx.Data["train_url"] = outputObsPath
  405. ctx.Data["params"] = ""
  406. ctx.Data["branchName"] = ctx.Repo.BranchName
  407. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  408. if err != nil {
  409. ctx.ServerError("getConfigList failed:", err)
  410. return err
  411. }
  412. ctx.Data["config_list"] = configList.ParaConfigs
  413. return nil
  414. }
  415. func trainJobErrorNewDataPrepare(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) error {
  416. ctx.Data["PageIsCloudBrain"] = true
  417. //can, err := canUserCreateTrainJob(ctx.User.ID)
  418. //if err != nil {
  419. // ctx.ServerError("canUserCreateTrainJob", err)
  420. // return
  421. //}
  422. //
  423. //if !can {
  424. // log.Error("the user can not create train-job")
  425. // ctx.ServerError("the user can not create train-job", fmt.Errorf("the user can not create train-job"))
  426. // return
  427. //}
  428. t := time.Now()
  429. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  430. ctx.Data["job_name"] = jobName
  431. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  432. if err != nil {
  433. ctx.ServerError("GetAllUserAttachments failed:", err)
  434. return err
  435. }
  436. ctx.Data["attachments"] = attachs
  437. var resourcePools modelarts.ResourcePool
  438. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  439. ctx.ServerError("json.Unmarshal failed:", err)
  440. return err
  441. }
  442. ctx.Data["resource_pools"] = resourcePools.Info
  443. var engines modelarts.Engine
  444. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  445. ctx.ServerError("json.Unmarshal failed:", err)
  446. return err
  447. }
  448. ctx.Data["engines"] = engines.Info
  449. var versionInfos modelarts.VersionInfo
  450. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  451. ctx.ServerError("json.Unmarshal failed:", err)
  452. return err
  453. }
  454. ctx.Data["engine_versions"] = versionInfos.Version
  455. var flavorInfos modelarts.Flavor
  456. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  457. ctx.ServerError("json.Unmarshal failed:", err)
  458. return err
  459. }
  460. ctx.Data["flavor_infos"] = flavorInfos.Info
  461. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  462. ctx.Data["train_url"] = outputObsPath
  463. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  464. if err != nil {
  465. ctx.ServerError("getConfigList failed:", err)
  466. return err
  467. }
  468. var Parameters modelarts.Parameters
  469. if err = json.Unmarshal([]byte(form.Params), &Parameters); err != nil {
  470. ctx.ServerError("json.Unmarshal failed:", err)
  471. return err
  472. }
  473. ctx.Data["params"] = Parameters.Parameter
  474. ctx.Data["config_list"] = configList.ParaConfigs
  475. ctx.Data["bootFile"] = form.BootFile
  476. ctx.Data["uuid"] = form.Attachment
  477. ctx.Data["branch_name"] = form.BranchName
  478. return nil
  479. }
  480. func TrainJobNewVersion(ctx *context.Context) {
  481. err := trainJobNewVersionDataPrepare(ctx)
  482. if err != nil {
  483. ctx.ServerError("get new train-job info failed", err)
  484. return
  485. }
  486. ctx.HTML(200, tplModelArtsTrainJobVersionNew)
  487. }
  488. func trainJobNewVersionDataPrepare(ctx *context.Context) error {
  489. ctx.Data["PageIsCloudBrain"] = true
  490. var jobID = ctx.Params(":jobid")
  491. var versionName = ctx.Query("version_name")
  492. // canNewJob, err := canUserCreateTrainJobVersion(ctx, jobID, versionName)
  493. // if err != nil {
  494. // ctx.ServerError("canNewJob can info failed", err)
  495. // return err
  496. // }
  497. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  498. if err != nil {
  499. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  500. return err
  501. }
  502. t := time.Now()
  503. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  504. ctx.Data["job_name"] = task.JobName
  505. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  506. if err != nil {
  507. ctx.ServerError("GetAllUserAttachments failed:", err)
  508. return err
  509. }
  510. ctx.Data["attachments"] = attachs
  511. var resourcePools modelarts.ResourcePool
  512. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  513. ctx.ServerError("json.Unmarshal failed:", err)
  514. return err
  515. }
  516. ctx.Data["resource_pools"] = resourcePools.Info
  517. var engines modelarts.Engine
  518. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  519. ctx.ServerError("json.Unmarshal failed:", err)
  520. return err
  521. }
  522. ctx.Data["engines"] = engines.Info
  523. var versionInfos modelarts.VersionInfo
  524. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  525. ctx.ServerError("json.Unmarshal failed:", err)
  526. return err
  527. }
  528. ctx.Data["engine_versions"] = versionInfos.Version
  529. var flavorInfos modelarts.Flavor
  530. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  531. ctx.ServerError("json.Unmarshal failed:", err)
  532. return err
  533. }
  534. ctx.Data["flavor_infos"] = flavorInfos.Info
  535. var Parameters modelarts.Parameters
  536. if err = json.Unmarshal([]byte(task.Parameters), &Parameters); err != nil {
  537. ctx.ServerError("json.Unmarshal failed:", err)
  538. return err
  539. }
  540. ctx.Data["params"] = Parameters.Parameter
  541. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  542. ctx.Data["train_url"] = outputObsPath
  543. branches, _, err := ctx.Repo.GitRepo.GetBranches(0, 0)
  544. if err != nil {
  545. ctx.ServerError("GetBranches error:", err)
  546. return err
  547. }
  548. ctx.Data["branches"] = branches
  549. ctx.Data["branch_name"] = task.BranchName
  550. ctx.Data["description"] = task.Description
  551. ctx.Data["boot_file"] = task.BootFile
  552. ctx.Data["dataset_name"] = task.DatasetName
  553. ctx.Data["work_server_number"] = task.WorkServerNumber
  554. ctx.Data["flavor_name"] = task.FlavorName
  555. ctx.Data["engine_name"] = task.EngineName
  556. ctx.Data["uuid"] = task.Uuid
  557. ctx.Data["flavor_code"] = task.FlavorCode
  558. ctx.Data["engine_id"] = task.EngineID
  559. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  560. if err != nil {
  561. ctx.ServerError("getConfigList failed:", err)
  562. return err
  563. }
  564. ctx.Data["config_list"] = configList.ParaConfigs
  565. return nil
  566. }
  567. func versionErrorDataPrepare(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) error {
  568. ctx.Data["PageIsCloudBrain"] = true
  569. var jobID = ctx.Params(":jobid")
  570. // var versionName = ctx.Params(":version-name")
  571. var versionName = ctx.Query("version_name")
  572. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  573. if err != nil {
  574. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  575. return err
  576. }
  577. t := time.Now()
  578. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  579. ctx.Data["job_name"] = task.JobName
  580. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  581. if err != nil {
  582. ctx.ServerError("GetAllUserAttachments failed:", err)
  583. return err
  584. }
  585. ctx.Data["attachments"] = attachs
  586. var resourcePools modelarts.ResourcePool
  587. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  588. ctx.ServerError("json.Unmarshal failed:", err)
  589. return err
  590. }
  591. ctx.Data["resource_pools"] = resourcePools.Info
  592. var engines modelarts.Engine
  593. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  594. ctx.ServerError("json.Unmarshal failed:", err)
  595. return err
  596. }
  597. ctx.Data["engines"] = engines.Info
  598. var versionInfos modelarts.VersionInfo
  599. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  600. ctx.ServerError("json.Unmarshal failed:", err)
  601. return err
  602. }
  603. ctx.Data["engine_versions"] = versionInfos.Version
  604. var flavorInfos modelarts.Flavor
  605. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  606. ctx.ServerError("json.Unmarshal failed:", err)
  607. return err
  608. }
  609. ctx.Data["flavor_infos"] = flavorInfos.Info
  610. var Parameters modelarts.Parameters
  611. if err = json.Unmarshal([]byte(form.Params), &Parameters); err != nil {
  612. ctx.ServerError("json.Unmarshal failed:", err)
  613. return err
  614. }
  615. ctx.Data["params"] = Parameters.Parameter
  616. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  617. ctx.Data["train_url"] = outputObsPath
  618. branches, _, err := ctx.Repo.GitRepo.GetBranches(0, 0)
  619. if err != nil {
  620. ctx.ServerError("GetBranches error:", err)
  621. return err
  622. }
  623. ctx.Data["branches"] = branches
  624. ctx.Data["description"] = form.Description
  625. ctx.Data["dataset_name"] = task.DatasetName
  626. ctx.Data["work_server_number"] = form.WorkServerNumber
  627. ctx.Data["flavor_name"] = form.FlavorName
  628. ctx.Data["engine_name"] = form.EngineName
  629. ctx.Data["flavor_code"] = task.FlavorCode
  630. ctx.Data["engine_id"] = task.EngineID
  631. ctx.Data["version_name"] = form.VersionName
  632. ctx.Data["bootFile"] = form.BootFile
  633. ctx.Data["uuid"] = form.Attachment
  634. ctx.Data["branch_name"] = form.BranchName
  635. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  636. if err != nil {
  637. ctx.ServerError("getConfigList failed:", err)
  638. return err
  639. }
  640. ctx.Data["config_list"] = configList.ParaConfigs
  641. return nil
  642. }
  643. func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) {
  644. ctx.Data["PageIsTrainJob"] = true
  645. VersionOutputPath := modelarts.GetVersionOutputPathByTotalVersionCount(modelarts.TotalVersionCount)
  646. jobName := form.JobName
  647. uuid := form.Attachment
  648. description := form.Description
  649. workServerNumber := form.WorkServerNumber
  650. engineID := form.EngineID
  651. bootFile := form.BootFile
  652. flavorCode := form.Flavor
  653. params := form.Params
  654. poolID := form.PoolID
  655. isSaveParam := form.IsSaveParam
  656. repo := ctx.Repo.Repository
  657. codeLocalPath := setting.JobPath + jobName + modelarts.CodePath
  658. codeObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.CodePath
  659. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath + VersionOutputPath + "/"
  660. logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath + VersionOutputPath + "/"
  661. dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/"
  662. branch_name := form.BranchName
  663. isLatestVersion := modelarts.IsLatestVersion
  664. FlavorName := form.FlavorName
  665. VersionCount := modelarts.VersionCount
  666. EngineName := form.EngineName
  667. count, err := models.GetCloudbrainTrainJobCountByUserID(ctx.User.ID)
  668. if err != nil {
  669. log.Error("GetCloudbrainTrainJobCountByUserID failed:%v", err, ctx.Data["MsgID"])
  670. trainJobErrorNewDataPrepare(ctx, form)
  671. ctx.RenderWithErr("system error", tplModelArtsTrainJobNew, &form)
  672. return
  673. } else {
  674. if count >= 1 {
  675. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  676. trainJobErrorNewDataPrepare(ctx, form)
  677. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsTrainJobNew, &form)
  678. return
  679. }
  680. }
  681. if err := paramCheckCreateTrainJob(form); err != nil {
  682. log.Error("paramCheckCreateTrainJob failed:(%v)", err)
  683. trainJobErrorNewDataPrepare(ctx, form)
  684. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobNew, &form)
  685. return
  686. }
  687. // attach, err := models.GetAttachmentByUUID(uuid)
  688. // if err != nil {
  689. // log.Error("GetAttachmentByUUID(%s) failed:%v", uuid, err.Error())
  690. // return
  691. // }
  692. //todo: del the codeLocalPath
  693. // _, err := ioutil.ReadDir(codeLocalPath)
  694. // if err == nil {
  695. // os.RemoveAll(codeLocalPath)
  696. // }
  697. os.RemoveAll(codeLocalPath)
  698. gitRepo, _ := git.OpenRepository(repo.RepoPath())
  699. commitID, _ := gitRepo.GetBranchCommitID(branch_name)
  700. if err := git.Clone(repo.RepoPath(), codeLocalPath, git.CloneRepoOptions{
  701. Branch: branch_name,
  702. }); err != nil {
  703. log.Error("创建任务失败,服务器超时!: %s (%v)", repo.FullName(), err)
  704. trainJobErrorNewDataPrepare(ctx, form)
  705. ctx.RenderWithErr("创建任务失败,服务器超时!", tplModelArtsTrainJobNew, &form)
  706. return
  707. }
  708. //todo: upload code (send to file_server todo this work?)
  709. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.OutputPath + VersionOutputPath + "/"); err != nil {
  710. log.Error("Failed to obsMkdir_output: %s (%v)", repo.FullName(), err)
  711. trainJobErrorNewDataPrepare(ctx, form)
  712. ctx.RenderWithErr("Failed to obsMkdir_output", tplModelArtsTrainJobNew, &form)
  713. return
  714. }
  715. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.LogPath + VersionOutputPath + "/"); err != nil {
  716. log.Error("Failed to obsMkdir_log: %s (%v)", repo.FullName(), err)
  717. trainJobErrorNewDataPrepare(ctx, form)
  718. ctx.RenderWithErr("Failed to obsMkdir_log", tplModelArtsTrainJobNew, &form)
  719. return
  720. }
  721. // parentDir := VersionOutputPath + "/"
  722. if err := uploadCodeToObs(codeLocalPath, jobName, ""); err != nil {
  723. // if err := uploadCodeToObs(codeLocalPath, jobName, parentDir); err != nil {
  724. log.Error("Failed to uploadCodeToObs: %s (%v)", repo.FullName(), err)
  725. trainJobErrorNewDataPrepare(ctx, form)
  726. ctx.RenderWithErr("Failed to uploadCodeToObs", tplModelArtsTrainJobNew, &form)
  727. return
  728. }
  729. //todo: del local code?
  730. var parameters models.Parameters
  731. param := make([]models.Parameter, 0)
  732. param = append(param, models.Parameter{
  733. Label: modelarts.TrainUrl,
  734. Value: outputObsPath,
  735. }, models.Parameter{
  736. Label: modelarts.DataUrl,
  737. Value: dataPath,
  738. })
  739. if len(params) != 0 {
  740. err := json.Unmarshal([]byte(params), &parameters)
  741. if err != nil {
  742. log.Error("Failed to Unmarshal params: %s (%v)", params, err)
  743. trainJobErrorNewDataPrepare(ctx, form)
  744. ctx.RenderWithErr("运行参数错误", tplModelArtsTrainJobNew, &form)
  745. return
  746. }
  747. for _, parameter := range parameters.Parameter {
  748. if parameter.Label != modelarts.TrainUrl && parameter.Label != modelarts.DataUrl {
  749. param = append(param, models.Parameter{
  750. Label: parameter.Label,
  751. Value: parameter.Value,
  752. })
  753. }
  754. }
  755. }
  756. //save param config
  757. if isSaveParam == "on" {
  758. if form.ParameterTemplateName == "" {
  759. log.Error("ParameterTemplateName is empty")
  760. trainJobNewDataPrepare(ctx)
  761. ctx.RenderWithErr("保存作业参数时,作业参数名称不能为空", tplModelArtsTrainJobNew, &form)
  762. return
  763. }
  764. _, err := modelarts.CreateTrainJobConfig(models.CreateConfigParams{
  765. ConfigName: form.ParameterTemplateName,
  766. Description: form.PrameterDescription,
  767. DataUrl: dataPath,
  768. AppUrl: codeObsPath,
  769. BootFileUrl: codeObsPath + bootFile,
  770. TrainUrl: outputObsPath,
  771. Flavor: models.Flavor{
  772. Code: flavorCode,
  773. },
  774. WorkServerNum: workServerNumber,
  775. EngineID: int64(engineID),
  776. LogUrl: logObsPath,
  777. PoolID: poolID,
  778. Parameter: param,
  779. })
  780. if err != nil {
  781. log.Error("Failed to CreateTrainJobConfig: %v", err)
  782. trainJobErrorNewDataPrepare(ctx, form)
  783. ctx.RenderWithErr("保存作业参数失败:"+err.Error(), tplModelArtsTrainJobNew, &form)
  784. return
  785. }
  786. }
  787. req := &modelarts.GenerateTrainJobReq{
  788. JobName: jobName,
  789. DataUrl: dataPath,
  790. Description: description,
  791. CodeObsPath: codeObsPath,
  792. BootFileUrl: codeObsPath + bootFile,
  793. BootFile: bootFile,
  794. TrainUrl: outputObsPath,
  795. FlavorCode: flavorCode,
  796. WorkServerNumber: workServerNumber,
  797. EngineID: int64(engineID),
  798. LogUrl: logObsPath,
  799. PoolID: poolID,
  800. Uuid: uuid,
  801. Parameters: parameters.Parameter,
  802. CommitID: commitID,
  803. IsLatestVersion: isLatestVersion,
  804. BranchName: branch_name,
  805. Params: form.Params,
  806. FlavorName: FlavorName,
  807. EngineName: EngineName,
  808. VersionCount: VersionCount,
  809. TotalVersionCount: modelarts.TotalVersionCount,
  810. }
  811. //将params转换Parameters.Parameter,出错时返回给前端
  812. var Parameters modelarts.Parameters
  813. if err := json.Unmarshal([]byte(params), &Parameters); err != nil {
  814. ctx.ServerError("json.Unmarshal failed:", err)
  815. return
  816. }
  817. err = modelarts.GenerateTrainJob(ctx, req)
  818. if err != nil {
  819. log.Error("GenerateTrainJob failed:%v", err.Error())
  820. trainJobErrorNewDataPrepare(ctx, form)
  821. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobNew, &form)
  822. return
  823. }
  824. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  825. }
  826. func TrainJobCreateVersion(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) {
  827. ctx.Data["PageIsTrainJob"] = true
  828. var jobID = ctx.Params(":jobid")
  829. count, err := models.GetCloudbrainTrainJobCountByUserID(ctx.User.ID)
  830. if err != nil {
  831. log.Error("GetCloudbrainTrainJobCountByUserID failed:%v", err, ctx.Data["MsgID"])
  832. versionErrorDataPrepare(ctx, form)
  833. ctx.RenderWithErr("system error", tplModelArtsTrainJobVersionNew, &form)
  834. return
  835. } else {
  836. if count >= 1 {
  837. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  838. versionErrorDataPrepare(ctx, form)
  839. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsTrainJobVersionNew, &form)
  840. return
  841. }
  842. }
  843. latestTask, err := models.GetCloudbrainByJobIDAndIsLatestVersion(jobID, modelarts.IsLatestVersion)
  844. if err != nil {
  845. ctx.ServerError("GetCloudbrainByJobIDAndIsLatestVersion faild:", err)
  846. return
  847. }
  848. VersionOutputPath := modelarts.GetVersionOutputPathByTotalVersionCount(latestTask.TotalVersionCount + 1)
  849. jobName := form.JobName
  850. uuid := form.Attachment
  851. description := form.Description
  852. workServerNumber := form.WorkServerNumber
  853. engineID := form.EngineID
  854. bootFile := form.BootFile
  855. flavorCode := form.Flavor
  856. params := form.Params
  857. poolID := form.PoolID
  858. isSaveParam := form.IsSaveParam
  859. repo := ctx.Repo.Repository
  860. codeLocalPath := setting.JobPath + jobName + modelarts.CodePath
  861. codeObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.CodePath + VersionOutputPath + "/"
  862. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath + VersionOutputPath + "/"
  863. logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath + VersionOutputPath + "/"
  864. dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/"
  865. branch_name := form.BranchName
  866. PreVersionName := form.VersionName
  867. FlavorName := form.FlavorName
  868. EngineName := form.EngineName
  869. isLatestVersion := modelarts.IsLatestVersion
  870. //判断权限
  871. canNewJob, _ := canUserCreateTrainJobVersion(ctx, latestTask.UserID)
  872. if !canNewJob {
  873. ctx.RenderWithErr("user cann't new trainjob", tplModelArtsTrainJobVersionNew, &form)
  874. return
  875. }
  876. if err := paramCheckCreateTrainJob(form); err != nil {
  877. log.Error("paramCheckCreateTrainJob failed:(%v)", err)
  878. versionErrorDataPrepare(ctx, form)
  879. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  880. return
  881. }
  882. // attach, err := models.GetAttachmentByUUID(uuid)
  883. // if err != nil {
  884. // log.Error("GetAttachmentByUUID(%s) failed:%v", uuid, err.Error())
  885. // return
  886. // }
  887. //todo: del the codeLocalPath
  888. // _, err = ioutil.ReadDir(codeLocalPath)
  889. // if err == nil {
  890. // os.RemoveAll(codeLocalPath)
  891. // }
  892. os.RemoveAll(codeLocalPath)
  893. gitRepo, _ := git.OpenRepository(repo.RepoPath())
  894. commitID, _ := gitRepo.GetBranchCommitID(branch_name)
  895. if err := git.Clone(repo.RepoPath(), codeLocalPath, git.CloneRepoOptions{
  896. Branch: branch_name,
  897. }); err != nil {
  898. log.Error("创建任务失败,任务名称已存在!: %s (%v)", repo.FullName(), err)
  899. versionErrorDataPrepare(ctx, form)
  900. ctx.RenderWithErr("创建任务失败,任务名称已存在!", tplModelArtsTrainJobVersionNew, &form)
  901. return
  902. }
  903. //todo: upload code (send to file_server todo this work?)
  904. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.OutputPath + VersionOutputPath + "/"); err != nil {
  905. log.Error("Failed to obsMkdir_output: %s (%v)", repo.FullName(), err)
  906. versionErrorDataPrepare(ctx, form)
  907. ctx.RenderWithErr("Failed to obsMkdir_output", tplModelArtsTrainJobVersionNew, &form)
  908. return
  909. }
  910. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.LogPath + VersionOutputPath + "/"); err != nil {
  911. log.Error("Failed to obsMkdir_log: %s (%v)", repo.FullName(), err)
  912. versionErrorDataPrepare(ctx, form)
  913. ctx.RenderWithErr("Failed to obsMkdir_log", tplModelArtsTrainJobVersionNew, &form)
  914. return
  915. }
  916. parentDir := VersionOutputPath + "/"
  917. // parentDir := ""
  918. // if err := uploadCodeToObs(codeLocalPath, jobName, ""); err != nil {
  919. if err := uploadCodeToObs(codeLocalPath, jobName, parentDir); err != nil {
  920. log.Error("Failed to uploadCodeToObs: %s (%v)", repo.FullName(), err)
  921. versionErrorDataPrepare(ctx, form)
  922. ctx.RenderWithErr("Failed to uploadCodeToObs", tplModelArtsTrainJobVersionNew, &form)
  923. return
  924. }
  925. //todo: del local code?
  926. var parameters models.Parameters
  927. param := make([]models.Parameter, 0)
  928. param = append(param, models.Parameter{
  929. Label: modelarts.TrainUrl,
  930. Value: outputObsPath,
  931. }, models.Parameter{
  932. Label: modelarts.DataUrl,
  933. Value: dataPath,
  934. })
  935. if len(params) != 0 {
  936. err := json.Unmarshal([]byte(params), &parameters)
  937. if err != nil {
  938. log.Error("Failed to Unmarshal params: %s (%v)", params, err)
  939. versionErrorDataPrepare(ctx, form)
  940. ctx.RenderWithErr("运行参数错误", tplModelArtsTrainJobVersionNew, &form)
  941. return
  942. }
  943. for _, parameter := range parameters.Parameter {
  944. if parameter.Label != modelarts.TrainUrl && parameter.Label != modelarts.DataUrl {
  945. param = append(param, models.Parameter{
  946. Label: parameter.Label,
  947. Value: parameter.Value,
  948. })
  949. }
  950. }
  951. }
  952. //save param config
  953. if isSaveParam == "on" {
  954. if form.ParameterTemplateName == "" {
  955. log.Error("ParameterTemplateName is empty")
  956. versionErrorDataPrepare(ctx, form)
  957. ctx.RenderWithErr("保存作业参数时,作业参数名称不能为空", tplModelArtsTrainJobVersionNew, &form)
  958. return
  959. }
  960. _, err := modelarts.CreateTrainJobConfig(models.CreateConfigParams{
  961. ConfigName: form.ParameterTemplateName,
  962. Description: form.PrameterDescription,
  963. DataUrl: dataPath,
  964. AppUrl: codeObsPath,
  965. BootFileUrl: codeObsPath + bootFile,
  966. TrainUrl: outputObsPath,
  967. Flavor: models.Flavor{
  968. Code: flavorCode,
  969. },
  970. WorkServerNum: workServerNumber,
  971. EngineID: int64(engineID),
  972. LogUrl: logObsPath,
  973. PoolID: poolID,
  974. Parameter: parameters.Parameter,
  975. })
  976. if err != nil {
  977. log.Error("Failed to CreateTrainJobConfig: %v", err)
  978. versionErrorDataPrepare(ctx, form)
  979. ctx.RenderWithErr("保存作业参数失败:"+err.Error(), tplModelArtsTrainJobVersionNew, &form)
  980. return
  981. }
  982. }
  983. if err != nil {
  984. log.Error("getFlavorNameByEngineID(%s) failed:%v", engineID, err.Error())
  985. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  986. return
  987. }
  988. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, PreVersionName)
  989. if err != nil {
  990. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  991. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  992. return
  993. }
  994. req := &modelarts.GenerateTrainJobReq{
  995. JobName: task.JobName,
  996. DataUrl: dataPath,
  997. Description: description,
  998. CodeObsPath: codeObsPath,
  999. BootFileUrl: codeObsPath + bootFile,
  1000. BootFile: bootFile,
  1001. TrainUrl: outputObsPath,
  1002. FlavorCode: flavorCode,
  1003. WorkServerNumber: workServerNumber,
  1004. IsLatestVersion: isLatestVersion,
  1005. EngineID: int64(engineID),
  1006. LogUrl: logObsPath,
  1007. PoolID: poolID,
  1008. Uuid: uuid,
  1009. Params: form.Params,
  1010. Parameters: parameters.Parameter,
  1011. PreVersionId: task.VersionID,
  1012. CommitID: commitID,
  1013. BranchName: branch_name,
  1014. FlavorName: FlavorName,
  1015. EngineName: EngineName,
  1016. PreVersionName: PreVersionName,
  1017. TotalVersionCount: latestTask.TotalVersionCount + 1,
  1018. }
  1019. err = modelarts.GenerateTrainJobVersion(ctx, req, jobID)
  1020. if err != nil {
  1021. log.Error("GenerateTrainJob failed:%v", err.Error())
  1022. versionErrorDataPrepare(ctx, form)
  1023. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  1024. return
  1025. }
  1026. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job/" + jobID)
  1027. // ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1028. }
  1029. // readDir reads the directory named by dirname and returns
  1030. // a list of directory entries sorted by filename.
  1031. func readDir(dirname string) ([]os.FileInfo, error) {
  1032. f, err := os.Open(dirname)
  1033. if err != nil {
  1034. return nil, err
  1035. }
  1036. list, err := f.Readdir(100)
  1037. f.Close()
  1038. if err != nil {
  1039. //todo: can not upload empty folder
  1040. if err == io.EOF {
  1041. return nil, nil
  1042. }
  1043. return nil, err
  1044. }
  1045. //sort.Slice(list, func(i, j int) bool { return list[i].Name() < list[j].Name() })
  1046. return list, nil
  1047. }
  1048. func uploadCodeToObs(codePath, jobName, parentDir string) error {
  1049. files, err := readDir(codePath)
  1050. if err != nil {
  1051. log.Error("readDir(%s) failed: %s", codePath, err.Error())
  1052. return err
  1053. }
  1054. for _, file := range files {
  1055. if file.IsDir() {
  1056. input := &obs.PutObjectInput{}
  1057. input.Bucket = setting.Bucket
  1058. input.Key = parentDir + file.Name() + "/"
  1059. _, err = storage.ObsCli.PutObject(input)
  1060. if err != nil {
  1061. log.Error("PutObject(%s) failed: %s", input.Key, err.Error())
  1062. return err
  1063. }
  1064. if err = uploadCodeToObs(codePath+file.Name()+"/", jobName, parentDir+file.Name()+"/"); err != nil {
  1065. log.Error("uploadCodeToObs(%s) failed: %s", file.Name(), err.Error())
  1066. return err
  1067. }
  1068. } else {
  1069. input := &obs.PutFileInput{}
  1070. input.Bucket = setting.Bucket
  1071. input.Key = setting.CodePathPrefix + jobName + "/code/" + parentDir + file.Name()
  1072. input.SourceFile = codePath + file.Name()
  1073. _, err = storage.ObsCli.PutFile(input)
  1074. if err != nil {
  1075. log.Error("PutFile(%s) failed: %s", input.SourceFile, err.Error())
  1076. return err
  1077. }
  1078. }
  1079. }
  1080. return nil
  1081. }
  1082. func obsMkdir(dir string) error {
  1083. input := &obs.PutObjectInput{}
  1084. input.Bucket = setting.Bucket
  1085. input.Key = dir
  1086. _, err := storage.ObsCli.PutObject(input)
  1087. if err != nil {
  1088. log.Error("PutObject(%s) failed: %s", input.Key, err.Error())
  1089. return err
  1090. }
  1091. return nil
  1092. }
  1093. func paramCheckCreateTrainJob(form auth.CreateModelArtsTrainJobForm) error {
  1094. if !strings.HasSuffix(form.BootFile, ".py") {
  1095. log.Error("the boot file(%s) must be a python file", form.BootFile)
  1096. return errors.New("启动文件必须是python文件")
  1097. }
  1098. if form.WorkServerNumber > 25 || form.WorkServerNumber < 1 {
  1099. log.Error("the WorkServerNumber(%d) must be in (1,25)", form.WorkServerNumber)
  1100. return errors.New("计算节点数必须在1-25之间")
  1101. }
  1102. if form.BranchName == "" {
  1103. log.Error("the branch must not be null!", form.BranchName)
  1104. return errors.New("代码分支不能为空!")
  1105. }
  1106. return nil
  1107. }
  1108. func TrainJobShow(ctx *context.Context) {
  1109. ctx.Data["PageIsCloudBrain"] = true
  1110. var jobID = ctx.Params(":jobid")
  1111. repo := ctx.Repo.Repository
  1112. page := ctx.QueryInt("page")
  1113. if page <= 0 {
  1114. page = 1
  1115. }
  1116. VersionListTasks, VersionListCount, err := models.CloudbrainsVersionList(&models.CloudbrainsOptions{
  1117. ListOptions: models.ListOptions{
  1118. Page: page,
  1119. PageSize: setting.UI.IssuePagingNum,
  1120. },
  1121. RepoID: repo.ID,
  1122. Type: models.TypeCloudBrainTwo,
  1123. JobType: string(models.JobTypeTrain),
  1124. JobID: jobID,
  1125. })
  1126. if err != nil {
  1127. log.Error("GetVersionListTasks(%s) failed:%v", jobID, err.Error())
  1128. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1129. return
  1130. }
  1131. //设置权限
  1132. canNewJob, err := canUserCreateTrainJobVersion(ctx, VersionListTasks[0].UserID)
  1133. if err != nil {
  1134. ctx.ServerError("canNewJob failed", err)
  1135. return
  1136. }
  1137. ctx.Data["canNewJob"] = canNewJob
  1138. //将运行参数转化为epoch_size = 3, device_target = Ascend的格式
  1139. for i, task := range VersionListTasks {
  1140. var parameters models.Parameters
  1141. err := json.Unmarshal([]byte(VersionListTasks[i].Parameters), &parameters)
  1142. if err != nil {
  1143. log.Error("Failed to Unmarshal Parameters: %s (%v)", VersionListTasks[i].Parameters, err)
  1144. trainJobNewDataPrepare(ctx)
  1145. return
  1146. }
  1147. if len(parameters.Parameter) > 0 {
  1148. paramTemp := ""
  1149. for _, Parameter := range parameters.Parameter {
  1150. param := Parameter.Label + " = " + Parameter.Value + "; "
  1151. paramTemp = paramTemp + param
  1152. }
  1153. VersionListTasks[i].Parameters = paramTemp[:len(paramTemp)-2]
  1154. } else {
  1155. VersionListTasks[i].Parameters = ""
  1156. }
  1157. VersionListTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  1158. VersionListTasks[i].CanModify = cloudbrain.CanModifyJob(ctx, &task.Cloudbrain)
  1159. }
  1160. pager := context.NewPagination(VersionListCount, setting.UI.IssuePagingNum, page, 5)
  1161. pager.SetDefaultParams(ctx)
  1162. ctx.Data["Page"] = pager
  1163. ctx.Data["jobID"] = jobID
  1164. ctx.Data["jobName"] = VersionListTasks[0].JobName
  1165. ctx.Data["version_list_task"] = VersionListTasks
  1166. ctx.Data["version_list_count"] = VersionListCount
  1167. ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1168. }
  1169. func TrainJobGetLog(ctx *context.Context) {
  1170. ctx.Data["PageIsTrainJob"] = true
  1171. var jobID = ctx.Params(":jobid")
  1172. var logFileName = ctx.Query("file_name")
  1173. var baseLine = ctx.Query("base_line")
  1174. var order = ctx.Query("order")
  1175. if order != modelarts.OrderDesc && order != modelarts.OrderAsc {
  1176. log.Error("order(%s) check failed", order)
  1177. ctx.HTML(http.StatusBadRequest, tplModelArtsTrainJobShow)
  1178. return
  1179. }
  1180. task, err := models.GetCloudbrainByJobID(jobID)
  1181. if err != nil {
  1182. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  1183. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1184. return
  1185. }
  1186. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), baseLine, logFileName, order, modelarts.Lines)
  1187. if err != nil {
  1188. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  1189. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1190. return
  1191. }
  1192. ctx.Data["log"] = result
  1193. //ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1194. }
  1195. func trainJobGetLog(jobID string) (*models.GetTrainJobLogFileNamesResult, *models.GetTrainJobLogResult, error) {
  1196. task, err := models.GetCloudbrainByJobID(jobID)
  1197. if err != nil {
  1198. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  1199. return nil, nil, err
  1200. }
  1201. resultLogFile, err := modelarts.GetTrainJobLogFileNames(jobID, strconv.FormatInt(task.VersionID, 10))
  1202. if err != nil {
  1203. log.Error("GetTrainJobLogFileNames(%s) failed:%v", jobID, err.Error())
  1204. return nil, nil, err
  1205. }
  1206. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), "", resultLogFile.LogFileList[0], modelarts.OrderDesc, modelarts.Lines)
  1207. if err != nil {
  1208. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  1209. return nil, nil, err
  1210. }
  1211. return resultLogFile, result, err
  1212. }
  1213. func TrainJobDel(ctx *context.Context) {
  1214. var jobID = ctx.Params(":jobid")
  1215. repo := ctx.Repo.Repository
  1216. VersionListTasks, _, err := models.CloudbrainsVersionList(&models.CloudbrainsOptions{
  1217. RepoID: repo.ID,
  1218. Type: models.TypeCloudBrainTwo,
  1219. JobType: string(models.JobTypeTrain),
  1220. JobID: jobID,
  1221. })
  1222. if err != nil {
  1223. ctx.ServerError("get VersionListTasks failed", err)
  1224. return
  1225. }
  1226. //删除modelarts上的任务记录
  1227. _, err = modelarts.DelTrainJob(jobID)
  1228. if err != nil {
  1229. log.Error("DelTrainJob(%s) failed:%v", jobID, err.Error())
  1230. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil)
  1231. return
  1232. }
  1233. //删除数据库Cloudbrain表的记录
  1234. for _, task := range VersionListTasks {
  1235. err = models.DeleteJob(&task.Cloudbrain)
  1236. if err != nil {
  1237. ctx.ServerError("DeleteJob failed", err)
  1238. return
  1239. }
  1240. }
  1241. //删除存储
  1242. if len(VersionListTasks) > 0 {
  1243. DeleteJobStorage(VersionListTasks[0].JobName)
  1244. }
  1245. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  1246. }
  1247. func TrainJobStop(ctx *context.Context) {
  1248. var jobID = ctx.Params(":jobid")
  1249. task := ctx.Cloudbrain
  1250. _, err := modelarts.StopTrainJob(jobID, strconv.FormatInt(task.VersionID, 10))
  1251. if err != nil {
  1252. log.Error("StopTrainJob(%s) failed:%v", task.JobName, err.Error())
  1253. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil)
  1254. return
  1255. }
  1256. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  1257. }
  1258. func canUserCreateTrainJob(uid int64) (bool, error) {
  1259. org, err := models.GetOrgByName(setting.AllowedOrg)
  1260. if err != nil {
  1261. log.Error("get allowed org failed: ", setting.AllowedOrg)
  1262. return false, err
  1263. }
  1264. return org.IsOrgMember(uid)
  1265. }
  1266. func canUserCreateTrainJobVersion(ctx *context.Context, userID int64) (bool, error) {
  1267. if ctx == nil || ctx.User == nil {
  1268. log.Error("user unlogin!")
  1269. return false, nil
  1270. }
  1271. if userID == ctx.User.ID || ctx.User.IsAdmin {
  1272. return true, nil
  1273. } else {
  1274. log.Error("Only user itself and admin can new trainjob!")
  1275. return false, nil
  1276. }
  1277. }
  1278. func TrainJobGetConfigList(ctx *context.Context) {
  1279. ctx.Data["PageIsTrainJob"] = true
  1280. var jobID = ctx.Params(":jobid")
  1281. var logFileName = ctx.Query("file_name")
  1282. var baseLine = ctx.Query("base_line")
  1283. var order = ctx.Query("order")
  1284. if order != modelarts.OrderDesc && order != modelarts.OrderAsc {
  1285. log.Error("order(%s) check failed", order)
  1286. ctx.HTML(http.StatusBadRequest, tplModelArtsTrainJobShow)
  1287. return
  1288. }
  1289. task, err := models.GetCloudbrainByJobID(jobID)
  1290. if err != nil {
  1291. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  1292. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1293. return
  1294. }
  1295. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), baseLine, logFileName, order, modelarts.Lines)
  1296. if err != nil {
  1297. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  1298. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1299. return
  1300. }
  1301. ctx.Data["log"] = result
  1302. //ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1303. }
  1304. func getConfigList(perPage, page int, sortBy, order, searchContent, configType string) (*models.GetConfigListResult, error) {
  1305. var result models.GetConfigListResult
  1306. list, err := modelarts.GetConfigList(perPage, page, sortBy, order, searchContent, configType)
  1307. if err != nil {
  1308. log.Error("GetConfigList failed:", err)
  1309. return &result, err
  1310. }
  1311. for _, config := range list.ParaConfigs {
  1312. paraConfig, err := modelarts.GetParaConfig(config.ConfigName, configType)
  1313. if err != nil {
  1314. log.Error("GetParaConfig failed:", err)
  1315. return &result, err
  1316. }
  1317. config.Result = paraConfig
  1318. }
  1319. return list, nil
  1320. }
  1321. func ModelDownload(ctx *context.Context) {
  1322. var (
  1323. err error
  1324. )
  1325. var jobID = ctx.Params(":jobid")
  1326. versionName := ctx.Query("version_name")
  1327. parentDir := ctx.Query("parent_dir")
  1328. fileName := ctx.Query("file_name")
  1329. log.Info("DownloadSingleModelFile start.")
  1330. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  1331. if err != nil {
  1332. log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
  1333. return
  1334. }
  1335. path := strings.TrimPrefix(path.Join(setting.TrainJobModelPath, task.JobName, setting.OutPutPath, versionName, parentDir, fileName), "/")
  1336. log.Info("Download path is:%s", path)
  1337. url, err := storage.GetObsCreateSignedUrlByBucketAndKey(setting.Bucket, path)
  1338. if err != nil {
  1339. log.Error("GetObsCreateSignedUrl failed: %v", err.Error(), ctx.Data["msgID"])
  1340. ctx.ServerError("GetObsCreateSignedUrl", err)
  1341. return
  1342. }
  1343. http.Redirect(ctx.Resp, ctx.Req.Request, url, http.StatusMovedPermanently)
  1344. }
  1345. func DeleteJobStorage(jobName string) error {
  1346. //delete local
  1347. localJobPath := setting.JobPath + jobName
  1348. err := os.RemoveAll(localJobPath)
  1349. if err != nil {
  1350. log.Error("RemoveAll(%s) failed:%v", localJobPath, err)
  1351. }
  1352. //delete oss
  1353. dirPath := setting.CodePathPrefix + jobName + "/"
  1354. err = storage.ObsRemoveObject(setting.Bucket, dirPath)
  1355. if err != nil {
  1356. log.Error("ObsRemoveObject(%s) failed:%v", localJobPath, err)
  1357. }
  1358. return nil
  1359. }