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 49 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
4 years ago
4 years ago
4 years ago
4 years ago
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
3 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
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
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
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
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
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
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
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
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
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
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
12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562
  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. jobType := string(models.JobTypeDebug)
  46. if debugListType == models.GPUResource {
  47. debugType = models.TypeCloudBrainOne
  48. jobType = ""
  49. }
  50. if debugListType == models.NPUResource {
  51. debugType = models.TypeCloudBrainTwo
  52. }
  53. ciTasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  54. ListOptions: models.ListOptions{
  55. Page: page,
  56. PageSize: setting.UI.IssuePagingNum,
  57. },
  58. RepoID: repo.ID,
  59. Type: debugType,
  60. JobType: jobType,
  61. })
  62. if err != nil {
  63. ctx.ServerError("Get debugjob faild:", err)
  64. return
  65. }
  66. for i, task := range ciTasks {
  67. ciTasks[i].CanDebug = cloudbrain.CanModifyJob(ctx, &task.Cloudbrain)
  68. ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  69. ciTasks[i].Cloudbrain.ComputeResource = task.ComputeResource
  70. }
  71. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  72. pager.AddParam(ctx, "debugListType", "ListType")
  73. ctx.Data["Page"] = pager
  74. ctx.Data["PageIsCloudBrain"] = true
  75. ctx.Data["Tasks"] = ciTasks
  76. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  77. ctx.Data["RepoIsEmpty"] = repo.IsEmpty
  78. ctx.HTML(200, tplDebugJobIndex)
  79. }
  80. // MustEnableDataset check if repository enable internal cb
  81. func MustEnableModelArts(ctx *context.Context) {
  82. if !ctx.Repo.CanRead(models.UnitTypeCloudBrain) {
  83. ctx.NotFound("MustEnableCloudbrain", nil)
  84. return
  85. }
  86. }
  87. func NotebookNew(ctx *context.Context) {
  88. ctx.Data["PageIsCloudBrain"] = true
  89. t := time.Now()
  90. var jobName = jobNamePrefixValid(cutString(ctx.User.Name, 5)) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  91. ctx.Data["job_name"] = jobName
  92. attachs, err := models.GetModelArtsUserAttachments(ctx.User.ID)
  93. if err != nil {
  94. ctx.ServerError("GetAllUserAttachments failed:", err)
  95. return
  96. }
  97. ctx.Data["attachments"] = attachs
  98. ctx.Data["dataset_path"] = modelarts.DataSetMountPath
  99. ctx.Data["env"] = modelarts.NotebookEnv
  100. ctx.Data["notebook_type"] = modelarts.NotebookType
  101. if modelarts.FlavorInfos == nil {
  102. json.Unmarshal([]byte(setting.FlavorInfos), &modelarts.FlavorInfos)
  103. }
  104. ctx.Data["flavors"] = modelarts.FlavorInfos.FlavorInfo
  105. ctx.HTML(200, tplModelArtsNotebookNew)
  106. }
  107. func NotebookCreate(ctx *context.Context, form auth.CreateModelArtsNotebookForm) {
  108. ctx.Data["PageIsNotebook"] = true
  109. jobName := form.JobName
  110. uuid := form.Attachment
  111. description := form.Description
  112. flavor := form.Flavor
  113. count, err := models.GetCloudbrainNotebookCountByUserID(ctx.User.ID)
  114. if err != nil {
  115. log.Error("GetCloudbrainNotebookCountByUserID failed:%v", err, ctx.Data["MsgID"])
  116. cloudBrainNewDataPrepare(ctx)
  117. ctx.RenderWithErr("system error", tplModelArtsNotebookNew, &form)
  118. return
  119. } else {
  120. if count >= 1 {
  121. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  122. cloudBrainNewDataPrepare(ctx)
  123. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsNotebookNew, &form)
  124. return
  125. }
  126. }
  127. err = modelarts.GenerateTask(ctx, jobName, uuid, description, flavor)
  128. if err != nil {
  129. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookNew, &form)
  130. return
  131. }
  132. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=all")
  133. }
  134. func NotebookShow(ctx *context.Context) {
  135. ctx.Data["PageIsCloudBrain"] = true
  136. var jobID = ctx.Params(":jobid")
  137. task, err := models.GetCloudbrainByJobID(jobID)
  138. if err != nil {
  139. ctx.Data["error"] = err.Error()
  140. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil)
  141. return
  142. }
  143. result, err := modelarts.GetJob(jobID)
  144. if err != nil {
  145. ctx.Data["error"] = err.Error()
  146. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil)
  147. return
  148. }
  149. if result != nil {
  150. task.Status = result.Status
  151. err = models.UpdateJob(task)
  152. if err != nil {
  153. ctx.Data["error"] = err.Error()
  154. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil)
  155. return
  156. }
  157. createTime, _ := com.StrTo(result.CreationTimestamp).Int64()
  158. result.CreateTime = time.Unix(int64(createTime/1000), 0).Format("2006-01-02 15:04:05")
  159. endTime, _ := com.StrTo(result.LatestUpdateTimestamp).Int64()
  160. result.LatestUpdateTime = time.Unix(int64(endTime/1000), 0).Format("2006-01-02 15:04:05")
  161. result.QueuingInfo.BeginTime = time.Unix(int64(result.QueuingInfo.BeginTimestamp/1000), 0).Format("2006-01-02 15:04:05")
  162. result.QueuingInfo.EndTime = time.Unix(int64(result.QueuingInfo.EndTimestamp/1000), 0).Format("2006-01-02 15:04:05")
  163. }
  164. ctx.Data["task"] = task
  165. ctx.Data["jobID"] = jobID
  166. ctx.Data["result"] = result
  167. ctx.HTML(200, tplModelArtsNotebookShow)
  168. }
  169. func NotebookDebug(ctx *context.Context) {
  170. var jobID = ctx.Params(":jobid")
  171. _, err := models.GetCloudbrainByJobID(jobID)
  172. if err != nil {
  173. ctx.ServerError("GetCloudbrainByJobID failed", err)
  174. return
  175. }
  176. result, err := modelarts.GetJob(jobID)
  177. if err != nil {
  178. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil)
  179. return
  180. }
  181. res, err := modelarts.GetJobToken(jobID)
  182. if err != nil {
  183. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil)
  184. return
  185. }
  186. urls := strings.Split(result.Spec.Annotations.Url, "/")
  187. urlPrefix := result.Spec.Annotations.TargetDomain
  188. for i, url := range urls {
  189. if i > 2 {
  190. urlPrefix += "/" + url
  191. }
  192. }
  193. debugUrl := urlPrefix + "?token=" + res.Token
  194. ctx.Redirect(debugUrl)
  195. }
  196. func NotebookManage(ctx *context.Context) {
  197. var jobID = ctx.Params(":jobid")
  198. var action = ctx.Params(":action")
  199. var resultCode = "0"
  200. var errorMsg = ""
  201. var status = ""
  202. for {
  203. task, err := models.GetCloudbrainByJobID(jobID)
  204. if err != nil {
  205. log.Error("GetCloudbrainByJobID failed:%v", err, ctx.Data["MsgID"])
  206. resultCode = "-1"
  207. errorMsg = "system error"
  208. break
  209. }
  210. if action == models.ActionStop {
  211. if task.Status != string(models.ModelArtsRunning) {
  212. log.Error("the job(%s) is not running", task.JobName, ctx.Data["MsgID"])
  213. resultCode = "-1"
  214. errorMsg = "the job is not running"
  215. break
  216. }
  217. if !ctx.IsSigned || (ctx.User.ID != task.UserID && !ctx.IsUserSiteAdmin() && !ctx.IsUserRepoOwner()) {
  218. log.Error("the user has no right ro stop the job", task.JobName, ctx.Data["MsgID"])
  219. resultCode = "-1"
  220. errorMsg = "you have no right to stop the job"
  221. break
  222. }
  223. } else if action == models.ActionRestart {
  224. if task.Status != string(models.ModelArtsStopped) && task.Status != string(models.ModelArtsStartFailed) && task.Status != string(models.ModelArtsCreateFailed) {
  225. log.Error("the job(%s) is not stopped", task.JobName, ctx.Data["MsgID"])
  226. resultCode = "-1"
  227. errorMsg = "the job is not stopped"
  228. break
  229. }
  230. if !ctx.IsSigned || (ctx.User.ID != task.UserID && !ctx.IsUserSiteAdmin()) {
  231. log.Error("the user has no right ro restart the job", task.JobName, ctx.Data["MsgID"])
  232. resultCode = "-1"
  233. errorMsg = "you have no right to restart the job"
  234. break
  235. }
  236. count, err := models.GetCloudbrainNotebookCountByUserID(ctx.User.ID)
  237. if err != nil {
  238. log.Error("GetCloudbrainNotebookCountByUserID failed:%v", err, ctx.Data["MsgID"])
  239. resultCode = "-1"
  240. errorMsg = "system error"
  241. break
  242. } else {
  243. if count >= 1 {
  244. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  245. resultCode = "-1"
  246. errorMsg = "you have already a running or waiting task, can not create more"
  247. break
  248. }
  249. }
  250. action = models.ActionStart
  251. } else {
  252. log.Error("the action(%s) is illegal", action, ctx.Data["MsgID"])
  253. resultCode = "-1"
  254. errorMsg = "非法操作"
  255. break
  256. }
  257. param := models.NotebookAction{
  258. Action: action,
  259. }
  260. res, err := modelarts.ManageNotebook(jobID, param)
  261. if err != nil {
  262. log.Error("ManageNotebook(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"])
  263. resultCode = "-1"
  264. errorMsg = "启动失败"
  265. break
  266. }
  267. task.Status = res.CurrentStatus
  268. err = models.UpdateJob(task)
  269. if err != nil {
  270. log.Error("UpdateJob(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"])
  271. resultCode = "-1"
  272. errorMsg = "system error"
  273. break
  274. }
  275. status = task.Status
  276. break
  277. }
  278. ctx.JSON(200, map[string]string{
  279. "result_code": resultCode,
  280. "error_msg": errorMsg,
  281. "status": status,
  282. "job_id": jobID,
  283. })
  284. }
  285. func NotebookDel(ctx *context.Context) {
  286. var jobID = ctx.Params(":jobid")
  287. task, err := models.GetCloudbrainByJobID(jobID)
  288. if err != nil {
  289. ctx.ServerError("GetCloudbrainByJobID failed", err)
  290. return
  291. }
  292. if task.Status != string(models.ModelArtsCreateFailed) && task.Status != string(models.ModelArtsStartFailed) && task.Status != string(models.ModelArtsStopped) {
  293. log.Error("the job(%s) has not been stopped", task.JobName)
  294. ctx.ServerError("the job has not been stopped", errors.New("the job has not been stopped"))
  295. return
  296. }
  297. _, err = modelarts.DelNotebook(jobID)
  298. if err != nil {
  299. log.Error("DelJob(%s) failed:%v", task.JobName, err.Error())
  300. ctx.ServerError("DelJob failed", err)
  301. return
  302. }
  303. err = models.DeleteJob(task)
  304. if err != nil {
  305. ctx.ServerError("DeleteJob failed", err)
  306. return
  307. }
  308. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=all")
  309. }
  310. func TrainJobIndex(ctx *context.Context) {
  311. MustEnableModelArts(ctx)
  312. repo := ctx.Repo.Repository
  313. page := ctx.QueryInt("page")
  314. if page <= 0 {
  315. page = 1
  316. }
  317. tasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  318. ListOptions: models.ListOptions{
  319. Page: page,
  320. PageSize: setting.UI.IssuePagingNum,
  321. },
  322. RepoID: repo.ID,
  323. Type: models.TypeCloudBrainTwo,
  324. JobType: string(models.JobTypeTrain),
  325. IsLatestVersion: modelarts.IsLatestVersion,
  326. })
  327. if err != nil {
  328. ctx.ServerError("Cloudbrain", err)
  329. return
  330. }
  331. for i, task := range tasks {
  332. tasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  333. tasks[i].CanModify = cloudbrain.CanModifyJob(ctx, &task.Cloudbrain)
  334. tasks[i].ComputeResource = models.NPUResource
  335. }
  336. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  337. pager.SetDefaultParams(ctx)
  338. ctx.Data["Page"] = pager
  339. ctx.Data["PageIsCloudBrain"] = true
  340. ctx.Data["Tasks"] = tasks
  341. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  342. ctx.Data["RepoIsEmpty"] = repo.IsEmpty
  343. ctx.HTML(200, tplModelArtsTrainJobIndex)
  344. }
  345. func TrainJobNew(ctx *context.Context) {
  346. err := trainJobNewDataPrepare(ctx)
  347. if err != nil {
  348. ctx.ServerError("get new train-job info failed", err)
  349. return
  350. }
  351. ctx.HTML(200, tplModelArtsTrainJobNew)
  352. }
  353. func trainJobNewDataPrepare(ctx *context.Context) error {
  354. ctx.Data["PageIsCloudBrain"] = true
  355. //can, err := canUserCreateTrainJob(ctx.User.ID)
  356. //if err != nil {
  357. // ctx.ServerError("canUserCreateTrainJob", err)
  358. // return
  359. //}
  360. //
  361. //if !can {
  362. // log.Error("the user can not create train-job")
  363. // ctx.ServerError("the user can not create train-job", fmt.Errorf("the user can not create train-job"))
  364. // return
  365. //}
  366. t := time.Now()
  367. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  368. ctx.Data["job_name"] = jobName
  369. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  370. if err != nil {
  371. ctx.ServerError("GetAllUserAttachments failed:", err)
  372. return err
  373. }
  374. ctx.Data["attachments"] = attachs
  375. var resourcePools modelarts.ResourcePool
  376. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  377. ctx.ServerError("json.Unmarshal failed:", err)
  378. return err
  379. }
  380. ctx.Data["resource_pools"] = resourcePools.Info
  381. var engines modelarts.Engine
  382. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  383. ctx.ServerError("json.Unmarshal failed:", err)
  384. return err
  385. }
  386. ctx.Data["engines"] = engines.Info
  387. var versionInfos modelarts.VersionInfo
  388. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  389. ctx.ServerError("json.Unmarshal failed:", err)
  390. return err
  391. }
  392. ctx.Data["engine_versions"] = versionInfos.Version
  393. var flavorInfos modelarts.Flavor
  394. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  395. ctx.ServerError("json.Unmarshal failed:", err)
  396. return err
  397. }
  398. ctx.Data["flavor_infos"] = flavorInfos.Info
  399. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  400. ctx.Data["train_url"] = outputObsPath
  401. ctx.Data["params"] = ""
  402. ctx.Data["branchName"] = ctx.Repo.BranchName
  403. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  404. if err != nil {
  405. ctx.ServerError("getConfigList failed:", err)
  406. return err
  407. }
  408. ctx.Data["config_list"] = configList.ParaConfigs
  409. return nil
  410. }
  411. func trainJobErrorNewDataPrepare(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) error {
  412. ctx.Data["PageIsCloudBrain"] = true
  413. //can, err := canUserCreateTrainJob(ctx.User.ID)
  414. //if err != nil {
  415. // ctx.ServerError("canUserCreateTrainJob", err)
  416. // return
  417. //}
  418. //
  419. //if !can {
  420. // log.Error("the user can not create train-job")
  421. // ctx.ServerError("the user can not create train-job", fmt.Errorf("the user can not create train-job"))
  422. // return
  423. //}
  424. t := time.Now()
  425. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  426. ctx.Data["job_name"] = jobName
  427. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  428. if err != nil {
  429. ctx.ServerError("GetAllUserAttachments failed:", err)
  430. return err
  431. }
  432. ctx.Data["attachments"] = attachs
  433. var resourcePools modelarts.ResourcePool
  434. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  435. ctx.ServerError("json.Unmarshal failed:", err)
  436. return err
  437. }
  438. ctx.Data["resource_pools"] = resourcePools.Info
  439. var engines modelarts.Engine
  440. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  441. ctx.ServerError("json.Unmarshal failed:", err)
  442. return err
  443. }
  444. ctx.Data["engines"] = engines.Info
  445. var versionInfos modelarts.VersionInfo
  446. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  447. ctx.ServerError("json.Unmarshal failed:", err)
  448. return err
  449. }
  450. ctx.Data["engine_versions"] = versionInfos.Version
  451. var flavorInfos modelarts.Flavor
  452. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  453. ctx.ServerError("json.Unmarshal failed:", err)
  454. return err
  455. }
  456. ctx.Data["flavor_infos"] = flavorInfos.Info
  457. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  458. ctx.Data["train_url"] = outputObsPath
  459. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  460. if err != nil {
  461. ctx.ServerError("getConfigList failed:", err)
  462. return err
  463. }
  464. var Parameters modelarts.Parameters
  465. if err = json.Unmarshal([]byte(form.Params), &Parameters); err != nil {
  466. ctx.ServerError("json.Unmarshal failed:", err)
  467. return err
  468. }
  469. ctx.Data["params"] = Parameters.Parameter
  470. ctx.Data["config_list"] = configList.ParaConfigs
  471. ctx.Data["bootFile"] = form.BootFile
  472. ctx.Data["uuid"] = form.Attachment
  473. ctx.Data["branch_name"] = form.BranchName
  474. return nil
  475. }
  476. func TrainJobNewVersion(ctx *context.Context) {
  477. err := trainJobNewVersionDataPrepare(ctx)
  478. if err != nil {
  479. ctx.ServerError("get new train-job info failed", err)
  480. return
  481. }
  482. ctx.HTML(200, tplModelArtsTrainJobVersionNew)
  483. }
  484. func trainJobNewVersionDataPrepare(ctx *context.Context) error {
  485. ctx.Data["PageIsCloudBrain"] = true
  486. var jobID = ctx.Params(":jobid")
  487. var versionName = ctx.Query("version_name")
  488. // canNewJob, err := canUserCreateTrainJobVersion(ctx, jobID, versionName)
  489. // if err != nil {
  490. // ctx.ServerError("canNewJob can info failed", err)
  491. // return err
  492. // }
  493. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  494. if err != nil {
  495. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  496. return err
  497. }
  498. t := time.Now()
  499. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  500. ctx.Data["job_name"] = task.JobName
  501. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  502. if err != nil {
  503. ctx.ServerError("GetAllUserAttachments failed:", err)
  504. return err
  505. }
  506. ctx.Data["attachments"] = attachs
  507. var resourcePools modelarts.ResourcePool
  508. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  509. ctx.ServerError("json.Unmarshal failed:", err)
  510. return err
  511. }
  512. ctx.Data["resource_pools"] = resourcePools.Info
  513. var engines modelarts.Engine
  514. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  515. ctx.ServerError("json.Unmarshal failed:", err)
  516. return err
  517. }
  518. ctx.Data["engines"] = engines.Info
  519. var versionInfos modelarts.VersionInfo
  520. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  521. ctx.ServerError("json.Unmarshal failed:", err)
  522. return err
  523. }
  524. ctx.Data["engine_versions"] = versionInfos.Version
  525. var flavorInfos modelarts.Flavor
  526. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  527. ctx.ServerError("json.Unmarshal failed:", err)
  528. return err
  529. }
  530. ctx.Data["flavor_infos"] = flavorInfos.Info
  531. var Parameters modelarts.Parameters
  532. if err = json.Unmarshal([]byte(task.Parameters), &Parameters); err != nil {
  533. ctx.ServerError("json.Unmarshal failed:", err)
  534. return err
  535. }
  536. ctx.Data["params"] = Parameters.Parameter
  537. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  538. ctx.Data["train_url"] = outputObsPath
  539. branches, _, err := ctx.Repo.GitRepo.GetBranches(0, 0)
  540. if err != nil {
  541. ctx.ServerError("GetBranches error:", err)
  542. return err
  543. }
  544. ctx.Data["branches"] = branches
  545. ctx.Data["branch_name"] = task.BranchName
  546. ctx.Data["description"] = task.Description
  547. ctx.Data["boot_file"] = task.BootFile
  548. ctx.Data["dataset_name"] = task.DatasetName
  549. ctx.Data["work_server_number"] = task.WorkServerNumber
  550. ctx.Data["flavor_name"] = task.FlavorName
  551. ctx.Data["engine_name"] = task.EngineName
  552. ctx.Data["uuid"] = task.Uuid
  553. ctx.Data["flavor_code"] = task.FlavorCode
  554. ctx.Data["engine_id"] = task.EngineID
  555. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  556. if err != nil {
  557. ctx.ServerError("getConfigList failed:", err)
  558. return err
  559. }
  560. ctx.Data["config_list"] = configList.ParaConfigs
  561. return nil
  562. }
  563. func versionErrorDataPrepare(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) error {
  564. ctx.Data["PageIsCloudBrain"] = true
  565. var jobID = ctx.Params(":jobid")
  566. // var versionName = ctx.Params(":version-name")
  567. var versionName = ctx.Query("version_name")
  568. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  569. if err != nil {
  570. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  571. return err
  572. }
  573. t := time.Now()
  574. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  575. ctx.Data["job_name"] = task.JobName
  576. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  577. if err != nil {
  578. ctx.ServerError("GetAllUserAttachments failed:", err)
  579. return err
  580. }
  581. ctx.Data["attachments"] = attachs
  582. var resourcePools modelarts.ResourcePool
  583. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  584. ctx.ServerError("json.Unmarshal failed:", err)
  585. return err
  586. }
  587. ctx.Data["resource_pools"] = resourcePools.Info
  588. var engines modelarts.Engine
  589. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  590. ctx.ServerError("json.Unmarshal failed:", err)
  591. return err
  592. }
  593. ctx.Data["engines"] = engines.Info
  594. var versionInfos modelarts.VersionInfo
  595. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  596. ctx.ServerError("json.Unmarshal failed:", err)
  597. return err
  598. }
  599. ctx.Data["engine_versions"] = versionInfos.Version
  600. var flavorInfos modelarts.Flavor
  601. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  602. ctx.ServerError("json.Unmarshal failed:", err)
  603. return err
  604. }
  605. ctx.Data["flavor_infos"] = flavorInfos.Info
  606. var Parameters modelarts.Parameters
  607. if err = json.Unmarshal([]byte(form.Params), &Parameters); err != nil {
  608. ctx.ServerError("json.Unmarshal failed:", err)
  609. return err
  610. }
  611. ctx.Data["params"] = Parameters.Parameter
  612. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  613. ctx.Data["train_url"] = outputObsPath
  614. branches, _, err := ctx.Repo.GitRepo.GetBranches(0, 0)
  615. if err != nil {
  616. ctx.ServerError("GetBranches error:", err)
  617. return err
  618. }
  619. ctx.Data["branches"] = branches
  620. ctx.Data["description"] = form.Description
  621. ctx.Data["dataset_name"] = task.DatasetName
  622. ctx.Data["work_server_number"] = form.WorkServerNumber
  623. ctx.Data["flavor_name"] = form.FlavorName
  624. ctx.Data["engine_name"] = form.EngineName
  625. ctx.Data["flavor_code"] = task.FlavorCode
  626. ctx.Data["engine_id"] = task.EngineID
  627. ctx.Data["version_name"] = form.VersionName
  628. ctx.Data["bootFile"] = form.BootFile
  629. ctx.Data["uuid"] = form.Attachment
  630. ctx.Data["branch_name"] = form.BranchName
  631. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  632. if err != nil {
  633. ctx.ServerError("getConfigList failed:", err)
  634. return err
  635. }
  636. ctx.Data["config_list"] = configList.ParaConfigs
  637. return nil
  638. }
  639. func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) {
  640. ctx.Data["PageIsTrainJob"] = true
  641. VersionOutputPath := modelarts.GetVersionOutputPathByTotalVersionCount(modelarts.TotalVersionCount)
  642. jobName := form.JobName
  643. uuid := form.Attachment
  644. description := form.Description
  645. workServerNumber := form.WorkServerNumber
  646. engineID := form.EngineID
  647. bootFile := form.BootFile
  648. flavorCode := form.Flavor
  649. params := form.Params
  650. poolID := form.PoolID
  651. isSaveParam := form.IsSaveParam
  652. repo := ctx.Repo.Repository
  653. codeLocalPath := setting.JobPath + jobName + modelarts.CodePath
  654. codeObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.CodePath
  655. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath + VersionOutputPath + "/"
  656. logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath + VersionOutputPath + "/"
  657. dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/"
  658. branch_name := form.BranchName
  659. isLatestVersion := modelarts.IsLatestVersion
  660. FlavorName := form.FlavorName
  661. VersionCount := modelarts.VersionCount
  662. EngineName := form.EngineName
  663. count, err := models.GetCloudbrainTrainJobCountByUserID(ctx.User.ID)
  664. if err != nil {
  665. log.Error("GetCloudbrainTrainJobCountByUserID failed:%v", err, ctx.Data["MsgID"])
  666. trainJobErrorNewDataPrepare(ctx, form)
  667. ctx.RenderWithErr("system error", tplModelArtsTrainJobNew, &form)
  668. return
  669. } else {
  670. if count >= 1 {
  671. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  672. trainJobErrorNewDataPrepare(ctx, form)
  673. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsTrainJobNew, &form)
  674. return
  675. }
  676. }
  677. if err := paramCheckCreateTrainJob(form); err != nil {
  678. log.Error("paramCheckCreateTrainJob failed:(%v)", err)
  679. trainJobErrorNewDataPrepare(ctx, form)
  680. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobNew, &form)
  681. return
  682. }
  683. // attach, err := models.GetAttachmentByUUID(uuid)
  684. // if err != nil {
  685. // log.Error("GetAttachmentByUUID(%s) failed:%v", uuid, err.Error())
  686. // return
  687. // }
  688. //todo: del the codeLocalPath
  689. // _, err := ioutil.ReadDir(codeLocalPath)
  690. // if err == nil {
  691. // os.RemoveAll(codeLocalPath)
  692. // }
  693. os.RemoveAll(codeLocalPath)
  694. gitRepo, _ := git.OpenRepository(repo.RepoPath())
  695. commitID, _ := gitRepo.GetBranchCommitID(branch_name)
  696. if err := git.Clone(repo.RepoPath(), codeLocalPath, git.CloneRepoOptions{
  697. Branch: branch_name,
  698. }); err != nil {
  699. log.Error("创建任务失败,服务器超时!: %s (%v)", repo.FullName(), err)
  700. trainJobErrorNewDataPrepare(ctx, form)
  701. ctx.RenderWithErr("创建任务失败,服务器超时!", tplModelArtsTrainJobNew, &form)
  702. return
  703. }
  704. //todo: upload code (send to file_server todo this work?)
  705. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.OutputPath + VersionOutputPath + "/"); err != nil {
  706. log.Error("Failed to obsMkdir_output: %s (%v)", repo.FullName(), err)
  707. trainJobErrorNewDataPrepare(ctx, form)
  708. ctx.RenderWithErr("Failed to obsMkdir_output", tplModelArtsTrainJobNew, &form)
  709. return
  710. }
  711. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.LogPath + VersionOutputPath + "/"); err != nil {
  712. log.Error("Failed to obsMkdir_log: %s (%v)", repo.FullName(), err)
  713. trainJobErrorNewDataPrepare(ctx, form)
  714. ctx.RenderWithErr("Failed to obsMkdir_log", tplModelArtsTrainJobNew, &form)
  715. return
  716. }
  717. // parentDir := VersionOutputPath + "/"
  718. if err := uploadCodeToObs(codeLocalPath, jobName, ""); err != nil {
  719. // if err := uploadCodeToObs(codeLocalPath, jobName, parentDir); err != nil {
  720. log.Error("Failed to uploadCodeToObs: %s (%v)", repo.FullName(), err)
  721. trainJobErrorNewDataPrepare(ctx, form)
  722. ctx.RenderWithErr("Failed to uploadCodeToObs", tplModelArtsTrainJobNew, &form)
  723. return
  724. }
  725. //todo: del local code?
  726. var parameters models.Parameters
  727. param := make([]models.Parameter, 0)
  728. param = append(param, models.Parameter{
  729. Label: modelarts.TrainUrl,
  730. Value: outputObsPath,
  731. }, models.Parameter{
  732. Label: modelarts.DataUrl,
  733. Value: dataPath,
  734. })
  735. if len(params) != 0 {
  736. err := json.Unmarshal([]byte(params), &parameters)
  737. if err != nil {
  738. log.Error("Failed to Unmarshal params: %s (%v)", params, err)
  739. trainJobErrorNewDataPrepare(ctx, form)
  740. ctx.RenderWithErr("运行参数错误", tplModelArtsTrainJobNew, &form)
  741. return
  742. }
  743. for _, parameter := range parameters.Parameter {
  744. if parameter.Label != modelarts.TrainUrl && parameter.Label != modelarts.DataUrl {
  745. param = append(param, models.Parameter{
  746. Label: parameter.Label,
  747. Value: parameter.Value,
  748. })
  749. }
  750. }
  751. }
  752. //save param config
  753. if isSaveParam == "on" {
  754. if form.ParameterTemplateName == "" {
  755. log.Error("ParameterTemplateName is empty")
  756. trainJobNewDataPrepare(ctx)
  757. ctx.RenderWithErr("保存作业参数时,作业参数名称不能为空", tplModelArtsTrainJobNew, &form)
  758. return
  759. }
  760. _, err := modelarts.CreateTrainJobConfig(models.CreateConfigParams{
  761. ConfigName: form.ParameterTemplateName,
  762. Description: form.PrameterDescription,
  763. DataUrl: dataPath,
  764. AppUrl: codeObsPath,
  765. BootFileUrl: codeObsPath + bootFile,
  766. TrainUrl: outputObsPath,
  767. Flavor: models.Flavor{
  768. Code: flavorCode,
  769. },
  770. WorkServerNum: workServerNumber,
  771. EngineID: int64(engineID),
  772. LogUrl: logObsPath,
  773. PoolID: poolID,
  774. Parameter: param,
  775. })
  776. if err != nil {
  777. log.Error("Failed to CreateTrainJobConfig: %v", err)
  778. trainJobErrorNewDataPrepare(ctx, form)
  779. ctx.RenderWithErr("保存作业参数失败:"+err.Error(), tplModelArtsTrainJobNew, &form)
  780. return
  781. }
  782. }
  783. req := &modelarts.GenerateTrainJobReq{
  784. JobName: jobName,
  785. DataUrl: dataPath,
  786. Description: description,
  787. CodeObsPath: codeObsPath,
  788. BootFileUrl: codeObsPath + bootFile,
  789. BootFile: bootFile,
  790. TrainUrl: outputObsPath,
  791. FlavorCode: flavorCode,
  792. WorkServerNumber: workServerNumber,
  793. EngineID: int64(engineID),
  794. LogUrl: logObsPath,
  795. PoolID: poolID,
  796. Uuid: uuid,
  797. Parameters: parameters.Parameter,
  798. CommitID: commitID,
  799. IsLatestVersion: isLatestVersion,
  800. BranchName: branch_name,
  801. Params: form.Params,
  802. FlavorName: FlavorName,
  803. EngineName: EngineName,
  804. VersionCount: VersionCount,
  805. TotalVersionCount: modelarts.TotalVersionCount,
  806. }
  807. //将params转换Parameters.Parameter,出错时返回给前端
  808. var Parameters modelarts.Parameters
  809. if err := json.Unmarshal([]byte(params), &Parameters); err != nil {
  810. ctx.ServerError("json.Unmarshal failed:", err)
  811. return
  812. }
  813. err = modelarts.GenerateTrainJob(ctx, req)
  814. if err != nil {
  815. log.Error("GenerateTrainJob failed:%v", err.Error())
  816. trainJobErrorNewDataPrepare(ctx, form)
  817. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobNew, &form)
  818. return
  819. }
  820. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  821. }
  822. func TrainJobCreateVersion(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) {
  823. ctx.Data["PageIsTrainJob"] = true
  824. var jobID = ctx.Params(":jobid")
  825. count, err := models.GetCloudbrainTrainJobCountByUserID(ctx.User.ID)
  826. if err != nil {
  827. log.Error("GetCloudbrainTrainJobCountByUserID failed:%v", err, ctx.Data["MsgID"])
  828. versionErrorDataPrepare(ctx, form)
  829. ctx.RenderWithErr("system error", tplModelArtsTrainJobVersionNew, &form)
  830. return
  831. } else {
  832. if count >= 1 {
  833. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  834. versionErrorDataPrepare(ctx, form)
  835. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsTrainJobVersionNew, &form)
  836. return
  837. }
  838. }
  839. latestTask, err := models.GetCloudbrainByJobIDAndIsLatestVersion(jobID, modelarts.IsLatestVersion)
  840. if err != nil {
  841. ctx.ServerError("GetCloudbrainByJobIDAndIsLatestVersion faild:", err)
  842. return
  843. }
  844. VersionOutputPath := modelarts.GetVersionOutputPathByTotalVersionCount(latestTask.TotalVersionCount + 1)
  845. jobName := form.JobName
  846. uuid := form.Attachment
  847. description := form.Description
  848. workServerNumber := form.WorkServerNumber
  849. engineID := form.EngineID
  850. bootFile := form.BootFile
  851. flavorCode := form.Flavor
  852. params := form.Params
  853. poolID := form.PoolID
  854. isSaveParam := form.IsSaveParam
  855. repo := ctx.Repo.Repository
  856. codeLocalPath := setting.JobPath + jobName + modelarts.CodePath
  857. codeObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.CodePath + VersionOutputPath + "/"
  858. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath + VersionOutputPath + "/"
  859. logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath + VersionOutputPath + "/"
  860. dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/"
  861. branch_name := form.BranchName
  862. PreVersionName := form.VersionName
  863. FlavorName := form.FlavorName
  864. EngineName := form.EngineName
  865. isLatestVersion := modelarts.IsLatestVersion
  866. //判断权限
  867. canNewJob, _ := canUserCreateTrainJobVersion(ctx, latestTask.UserID)
  868. if !canNewJob {
  869. ctx.RenderWithErr("user cann't new trainjob", tplModelArtsTrainJobVersionNew, &form)
  870. return
  871. }
  872. if err := paramCheckCreateTrainJob(form); err != nil {
  873. log.Error("paramCheckCreateTrainJob failed:(%v)", err)
  874. versionErrorDataPrepare(ctx, form)
  875. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  876. return
  877. }
  878. // attach, err := models.GetAttachmentByUUID(uuid)
  879. // if err != nil {
  880. // log.Error("GetAttachmentByUUID(%s) failed:%v", uuid, err.Error())
  881. // return
  882. // }
  883. //todo: del the codeLocalPath
  884. // _, err = ioutil.ReadDir(codeLocalPath)
  885. // if err == nil {
  886. // os.RemoveAll(codeLocalPath)
  887. // }
  888. os.RemoveAll(codeLocalPath)
  889. gitRepo, _ := git.OpenRepository(repo.RepoPath())
  890. commitID, _ := gitRepo.GetBranchCommitID(branch_name)
  891. if err := git.Clone(repo.RepoPath(), codeLocalPath, git.CloneRepoOptions{
  892. Branch: branch_name,
  893. }); err != nil {
  894. log.Error("创建任务失败,任务名称已存在!: %s (%v)", repo.FullName(), err)
  895. versionErrorDataPrepare(ctx, form)
  896. ctx.RenderWithErr("创建任务失败,任务名称已存在!", tplModelArtsTrainJobVersionNew, &form)
  897. return
  898. }
  899. //todo: upload code (send to file_server todo this work?)
  900. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.OutputPath + VersionOutputPath + "/"); err != nil {
  901. log.Error("Failed to obsMkdir_output: %s (%v)", repo.FullName(), err)
  902. versionErrorDataPrepare(ctx, form)
  903. ctx.RenderWithErr("Failed to obsMkdir_output", tplModelArtsTrainJobVersionNew, &form)
  904. return
  905. }
  906. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.LogPath + VersionOutputPath + "/"); err != nil {
  907. log.Error("Failed to obsMkdir_log: %s (%v)", repo.FullName(), err)
  908. versionErrorDataPrepare(ctx, form)
  909. ctx.RenderWithErr("Failed to obsMkdir_log", tplModelArtsTrainJobVersionNew, &form)
  910. return
  911. }
  912. parentDir := VersionOutputPath + "/"
  913. // parentDir := ""
  914. // if err := uploadCodeToObs(codeLocalPath, jobName, ""); err != nil {
  915. if err := uploadCodeToObs(codeLocalPath, jobName, parentDir); err != nil {
  916. log.Error("Failed to uploadCodeToObs: %s (%v)", repo.FullName(), err)
  917. versionErrorDataPrepare(ctx, form)
  918. ctx.RenderWithErr("Failed to uploadCodeToObs", tplModelArtsTrainJobVersionNew, &form)
  919. return
  920. }
  921. //todo: del local code?
  922. var parameters models.Parameters
  923. param := make([]models.Parameter, 0)
  924. param = append(param, models.Parameter{
  925. Label: modelarts.TrainUrl,
  926. Value: outputObsPath,
  927. }, models.Parameter{
  928. Label: modelarts.DataUrl,
  929. Value: dataPath,
  930. })
  931. if len(params) != 0 {
  932. err := json.Unmarshal([]byte(params), &parameters)
  933. if err != nil {
  934. log.Error("Failed to Unmarshal params: %s (%v)", params, err)
  935. versionErrorDataPrepare(ctx, form)
  936. ctx.RenderWithErr("运行参数错误", tplModelArtsTrainJobVersionNew, &form)
  937. return
  938. }
  939. for _, parameter := range parameters.Parameter {
  940. if parameter.Label != modelarts.TrainUrl && parameter.Label != modelarts.DataUrl {
  941. param = append(param, models.Parameter{
  942. Label: parameter.Label,
  943. Value: parameter.Value,
  944. })
  945. }
  946. }
  947. }
  948. //save param config
  949. if isSaveParam == "on" {
  950. if form.ParameterTemplateName == "" {
  951. log.Error("ParameterTemplateName is empty")
  952. versionErrorDataPrepare(ctx, form)
  953. ctx.RenderWithErr("保存作业参数时,作业参数名称不能为空", tplModelArtsTrainJobVersionNew, &form)
  954. return
  955. }
  956. _, err := modelarts.CreateTrainJobConfig(models.CreateConfigParams{
  957. ConfigName: form.ParameterTemplateName,
  958. Description: form.PrameterDescription,
  959. DataUrl: dataPath,
  960. AppUrl: codeObsPath,
  961. BootFileUrl: codeObsPath + bootFile,
  962. TrainUrl: outputObsPath,
  963. Flavor: models.Flavor{
  964. Code: flavorCode,
  965. },
  966. WorkServerNum: workServerNumber,
  967. EngineID: int64(engineID),
  968. LogUrl: logObsPath,
  969. PoolID: poolID,
  970. Parameter: parameters.Parameter,
  971. })
  972. if err != nil {
  973. log.Error("Failed to CreateTrainJobConfig: %v", err)
  974. versionErrorDataPrepare(ctx, form)
  975. ctx.RenderWithErr("保存作业参数失败:"+err.Error(), tplModelArtsTrainJobVersionNew, &form)
  976. return
  977. }
  978. }
  979. if err != nil {
  980. log.Error("getFlavorNameByEngineID(%s) failed:%v", engineID, err.Error())
  981. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  982. return
  983. }
  984. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, PreVersionName)
  985. if err != nil {
  986. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  987. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  988. return
  989. }
  990. req := &modelarts.GenerateTrainJobReq{
  991. JobName: task.JobName,
  992. DataUrl: dataPath,
  993. Description: description,
  994. CodeObsPath: codeObsPath,
  995. BootFileUrl: codeObsPath + bootFile,
  996. BootFile: bootFile,
  997. TrainUrl: outputObsPath,
  998. FlavorCode: flavorCode,
  999. WorkServerNumber: workServerNumber,
  1000. IsLatestVersion: isLatestVersion,
  1001. EngineID: int64(engineID),
  1002. LogUrl: logObsPath,
  1003. PoolID: poolID,
  1004. Uuid: uuid,
  1005. Params: form.Params,
  1006. Parameters: parameters.Parameter,
  1007. PreVersionId: task.VersionID,
  1008. CommitID: commitID,
  1009. BranchName: branch_name,
  1010. FlavorName: FlavorName,
  1011. EngineName: EngineName,
  1012. PreVersionName: PreVersionName,
  1013. TotalVersionCount: latestTask.TotalVersionCount + 1,
  1014. }
  1015. err = modelarts.GenerateTrainJobVersion(ctx, req, jobID)
  1016. if err != nil {
  1017. log.Error("GenerateTrainJob failed:%v", err.Error())
  1018. versionErrorDataPrepare(ctx, form)
  1019. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  1020. return
  1021. }
  1022. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job/" + jobID)
  1023. // ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1024. }
  1025. // readDir reads the directory named by dirname and returns
  1026. // a list of directory entries sorted by filename.
  1027. func readDir(dirname string) ([]os.FileInfo, error) {
  1028. f, err := os.Open(dirname)
  1029. if err != nil {
  1030. return nil, err
  1031. }
  1032. list, err := f.Readdir(100)
  1033. f.Close()
  1034. if err != nil {
  1035. //todo: can not upload empty folder
  1036. if err == io.EOF {
  1037. return nil, nil
  1038. }
  1039. return nil, err
  1040. }
  1041. //sort.Slice(list, func(i, j int) bool { return list[i].Name() < list[j].Name() })
  1042. return list, nil
  1043. }
  1044. func uploadCodeToObs(codePath, jobName, parentDir string) error {
  1045. files, err := readDir(codePath)
  1046. if err != nil {
  1047. log.Error("readDir(%s) failed: %s", codePath, err.Error())
  1048. return err
  1049. }
  1050. for _, file := range files {
  1051. if file.IsDir() {
  1052. input := &obs.PutObjectInput{}
  1053. input.Bucket = setting.Bucket
  1054. input.Key = parentDir + file.Name() + "/"
  1055. _, err = storage.ObsCli.PutObject(input)
  1056. if err != nil {
  1057. log.Error("PutObject(%s) failed: %s", input.Key, err.Error())
  1058. return err
  1059. }
  1060. if err = uploadCodeToObs(codePath+file.Name()+"/", jobName, parentDir+file.Name()+"/"); err != nil {
  1061. log.Error("uploadCodeToObs(%s) failed: %s", file.Name(), err.Error())
  1062. return err
  1063. }
  1064. } else {
  1065. input := &obs.PutFileInput{}
  1066. input.Bucket = setting.Bucket
  1067. input.Key = setting.CodePathPrefix + jobName + "/code/" + parentDir + file.Name()
  1068. input.SourceFile = codePath + file.Name()
  1069. _, err = storage.ObsCli.PutFile(input)
  1070. if err != nil {
  1071. log.Error("PutFile(%s) failed: %s", input.SourceFile, err.Error())
  1072. return err
  1073. }
  1074. }
  1075. }
  1076. return nil
  1077. }
  1078. func obsMkdir(dir string) error {
  1079. input := &obs.PutObjectInput{}
  1080. input.Bucket = setting.Bucket
  1081. input.Key = dir
  1082. _, err := storage.ObsCli.PutObject(input)
  1083. if err != nil {
  1084. log.Error("PutObject(%s) failed: %s", input.Key, err.Error())
  1085. return err
  1086. }
  1087. return nil
  1088. }
  1089. func paramCheckCreateTrainJob(form auth.CreateModelArtsTrainJobForm) error {
  1090. if !strings.HasSuffix(form.BootFile, ".py") {
  1091. log.Error("the boot file(%s) must be a python file", form.BootFile)
  1092. return errors.New("启动文件必须是python文件")
  1093. }
  1094. if form.WorkServerNumber > 25 || form.WorkServerNumber < 1 {
  1095. log.Error("the WorkServerNumber(%d) must be in (1,25)", form.WorkServerNumber)
  1096. return errors.New("计算节点数必须在1-25之间")
  1097. }
  1098. return nil
  1099. }
  1100. func TrainJobShow(ctx *context.Context) {
  1101. ctx.Data["PageIsCloudBrain"] = true
  1102. var jobID = ctx.Params(":jobid")
  1103. repo := ctx.Repo.Repository
  1104. page := ctx.QueryInt("page")
  1105. if page <= 0 {
  1106. page = 1
  1107. }
  1108. VersionListTasks, VersionListCount, err := models.CloudbrainsVersionList(&models.CloudbrainsOptions{
  1109. ListOptions: models.ListOptions{
  1110. Page: page,
  1111. PageSize: setting.UI.IssuePagingNum,
  1112. },
  1113. RepoID: repo.ID,
  1114. Type: models.TypeCloudBrainTwo,
  1115. JobType: string(models.JobTypeTrain),
  1116. JobID: jobID,
  1117. })
  1118. if err != nil {
  1119. log.Error("GetVersionListTasks(%s) failed:%v", jobID, err.Error())
  1120. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1121. return
  1122. }
  1123. //设置权限
  1124. canNewJob, err := canUserCreateTrainJobVersion(ctx, VersionListTasks[0].UserID)
  1125. if err != nil {
  1126. ctx.ServerError("canNewJob failed", err)
  1127. return
  1128. }
  1129. ctx.Data["canNewJob"] = canNewJob
  1130. //将运行参数转化为epoch_size = 3, device_target = Ascend的格式
  1131. for i, task := range VersionListTasks {
  1132. var parameters models.Parameters
  1133. err := json.Unmarshal([]byte(VersionListTasks[i].Parameters), &parameters)
  1134. if err != nil {
  1135. log.Error("Failed to Unmarshal Parameters: %s (%v)", VersionListTasks[i].Parameters, err)
  1136. trainJobNewDataPrepare(ctx)
  1137. return
  1138. }
  1139. if len(parameters.Parameter) > 0 {
  1140. paramTemp := ""
  1141. for _, Parameter := range parameters.Parameter {
  1142. param := Parameter.Label + " = " + Parameter.Value + "; "
  1143. paramTemp = paramTemp + param
  1144. }
  1145. VersionListTasks[i].Parameters = paramTemp[:len(paramTemp)-2]
  1146. } else {
  1147. VersionListTasks[i].Parameters = ""
  1148. }
  1149. VersionListTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  1150. VersionListTasks[i].CanModify = cloudbrain.CanModifyJob(ctx, &task.Cloudbrain)
  1151. }
  1152. pager := context.NewPagination(VersionListCount, setting.UI.IssuePagingNum, page, 5)
  1153. pager.SetDefaultParams(ctx)
  1154. ctx.Data["Page"] = pager
  1155. ctx.Data["jobID"] = jobID
  1156. ctx.Data["jobName"] = VersionListTasks[0].JobName
  1157. ctx.Data["version_list_task"] = VersionListTasks
  1158. ctx.Data["version_list_count"] = VersionListCount
  1159. ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1160. }
  1161. func TrainJobGetLog(ctx *context.Context) {
  1162. ctx.Data["PageIsTrainJob"] = true
  1163. var jobID = ctx.Params(":jobid")
  1164. var logFileName = ctx.Query("file_name")
  1165. var baseLine = ctx.Query("base_line")
  1166. var order = ctx.Query("order")
  1167. if order != modelarts.OrderDesc && order != modelarts.OrderAsc {
  1168. log.Error("order(%s) check failed", order)
  1169. ctx.HTML(http.StatusBadRequest, tplModelArtsTrainJobShow)
  1170. return
  1171. }
  1172. task, err := models.GetCloudbrainByJobID(jobID)
  1173. if err != nil {
  1174. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  1175. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1176. return
  1177. }
  1178. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), baseLine, logFileName, order, modelarts.Lines)
  1179. if err != nil {
  1180. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  1181. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1182. return
  1183. }
  1184. ctx.Data["log"] = result
  1185. //ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1186. }
  1187. func trainJobGetLog(jobID string) (*models.GetTrainJobLogFileNamesResult, *models.GetTrainJobLogResult, error) {
  1188. task, err := models.GetCloudbrainByJobID(jobID)
  1189. if err != nil {
  1190. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  1191. return nil, nil, err
  1192. }
  1193. resultLogFile, err := modelarts.GetTrainJobLogFileNames(jobID, strconv.FormatInt(task.VersionID, 10))
  1194. if err != nil {
  1195. log.Error("GetTrainJobLogFileNames(%s) failed:%v", jobID, err.Error())
  1196. return nil, nil, err
  1197. }
  1198. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), "", resultLogFile.LogFileList[0], modelarts.OrderDesc, modelarts.Lines)
  1199. if err != nil {
  1200. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  1201. return nil, nil, err
  1202. }
  1203. return resultLogFile, result, err
  1204. }
  1205. func TrainJobDel(ctx *context.Context) {
  1206. var jobID = ctx.Params(":jobid")
  1207. repo := ctx.Repo.Repository
  1208. VersionListTasks, _, err := models.CloudbrainsVersionList(&models.CloudbrainsOptions{
  1209. RepoID: repo.ID,
  1210. Type: models.TypeCloudBrainTwo,
  1211. JobType: string(models.JobTypeTrain),
  1212. JobID: jobID,
  1213. })
  1214. if err != nil {
  1215. ctx.ServerError("get VersionListTasks failed", err)
  1216. return
  1217. }
  1218. //删除modelarts上的任务记录
  1219. _, err = modelarts.DelTrainJob(jobID)
  1220. if err != nil {
  1221. log.Error("DelTrainJob(%s) failed:%v", jobID, err.Error())
  1222. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil)
  1223. return
  1224. }
  1225. //删除数据库Cloudbrain表的记录
  1226. for _, task := range VersionListTasks {
  1227. err = models.DeleteJob(&task.Cloudbrain)
  1228. if err != nil {
  1229. ctx.ServerError("DeleteJob failed", err)
  1230. return
  1231. }
  1232. }
  1233. //删除存储
  1234. if len(VersionListTasks) > 0 {
  1235. DeleteJobStorage(VersionListTasks[0].JobName)
  1236. }
  1237. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  1238. }
  1239. func TrainJobStop(ctx *context.Context) {
  1240. var jobID = ctx.Params(":jobid")
  1241. task, err := models.GetCloudbrainByJobID(jobID)
  1242. if err != nil {
  1243. log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
  1244. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil)
  1245. return
  1246. }
  1247. _, err = modelarts.StopTrainJob(jobID, strconv.FormatInt(task.VersionID, 10))
  1248. if err != nil {
  1249. log.Error("StopTrainJob(%s) failed:%v", task.JobName, err.Error())
  1250. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil)
  1251. return
  1252. }
  1253. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  1254. }
  1255. func canUserCreateTrainJob(uid int64) (bool, error) {
  1256. org, err := models.GetOrgByName(setting.AllowedOrg)
  1257. if err != nil {
  1258. log.Error("get allowed org failed: ", setting.AllowedOrg)
  1259. return false, err
  1260. }
  1261. return org.IsOrgMember(uid)
  1262. }
  1263. func canUserCreateTrainJobVersion(ctx *context.Context, userID int64) (bool, error) {
  1264. if ctx == nil || ctx.User == nil {
  1265. log.Error("user unlogin!")
  1266. return false, nil
  1267. }
  1268. if userID == ctx.User.ID || ctx.User.IsAdmin {
  1269. return true, nil
  1270. } else {
  1271. log.Error("Only user itself and admin can new trainjob!")
  1272. return false, nil
  1273. }
  1274. }
  1275. func TrainJobGetConfigList(ctx *context.Context) {
  1276. ctx.Data["PageIsTrainJob"] = true
  1277. var jobID = ctx.Params(":jobid")
  1278. var logFileName = ctx.Query("file_name")
  1279. var baseLine = ctx.Query("base_line")
  1280. var order = ctx.Query("order")
  1281. if order != modelarts.OrderDesc && order != modelarts.OrderAsc {
  1282. log.Error("order(%s) check failed", order)
  1283. ctx.HTML(http.StatusBadRequest, tplModelArtsTrainJobShow)
  1284. return
  1285. }
  1286. task, err := models.GetCloudbrainByJobID(jobID)
  1287. if err != nil {
  1288. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  1289. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1290. return
  1291. }
  1292. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), baseLine, logFileName, order, modelarts.Lines)
  1293. if err != nil {
  1294. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  1295. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1296. return
  1297. }
  1298. ctx.Data["log"] = result
  1299. //ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1300. }
  1301. func getConfigList(perPage, page int, sortBy, order, searchContent, configType string) (*models.GetConfigListResult, error) {
  1302. var result models.GetConfigListResult
  1303. list, err := modelarts.GetConfigList(perPage, page, sortBy, order, searchContent, configType)
  1304. if err != nil {
  1305. log.Error("GetConfigList failed:", err)
  1306. return &result, err
  1307. }
  1308. for _, config := range list.ParaConfigs {
  1309. paraConfig, err := modelarts.GetParaConfig(config.ConfigName, configType)
  1310. if err != nil {
  1311. log.Error("GetParaConfig failed:", err)
  1312. return &result, err
  1313. }
  1314. config.Result = paraConfig
  1315. }
  1316. return list, nil
  1317. }
  1318. func ModelDownload(ctx *context.Context) {
  1319. var (
  1320. err error
  1321. )
  1322. var jobID = ctx.Params(":jobid")
  1323. versionName := ctx.Query("version_name")
  1324. parentDir := ctx.Query("parent_dir")
  1325. fileName := ctx.Query("file_name")
  1326. log.Info("DownloadSingleModelFile start.")
  1327. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  1328. if err != nil {
  1329. log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
  1330. return
  1331. }
  1332. path := strings.TrimPrefix(path.Join(setting.TrainJobModelPath, task.JobName, setting.OutPutPath, versionName, parentDir, fileName), "/")
  1333. log.Info("Download path is:%s", path)
  1334. url, err := storage.GetObsCreateSignedUrlByBucketAndKey(setting.Bucket, path)
  1335. if err != nil {
  1336. log.Error("GetObsCreateSignedUrl failed: %v", err.Error(), ctx.Data["msgID"])
  1337. ctx.ServerError("GetObsCreateSignedUrl", err)
  1338. return
  1339. }
  1340. http.Redirect(ctx.Resp, ctx.Req.Request, url, http.StatusMovedPermanently)
  1341. }
  1342. func DeleteJobStorage(jobName string) error {
  1343. //delete local
  1344. localJobPath := setting.JobPath + jobName
  1345. err := os.RemoveAll(localJobPath)
  1346. if err != nil {
  1347. log.Error("RemoveAll(%s) failed:%v", localJobPath, err)
  1348. }
  1349. //delete oss
  1350. dirPath := setting.CodePathPrefix + jobName + "/"
  1351. err = storage.ObsRemoveObject(setting.Bucket, dirPath)
  1352. if err != nil {
  1353. log.Error("ObsRemoveObject(%s) failed:%v", localJobPath, err)
  1354. }
  1355. return nil
  1356. }