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 72 kB

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
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
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
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
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 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
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 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
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
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
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
3 years ago
4 years ago
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
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
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
4 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
3 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
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
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
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
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
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
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
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
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
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
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
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
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
3 years ago
4 years ago
4 years ago
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
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
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
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
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
4 years ago
4 years ago
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
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
3 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
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
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
3 years ago
3 years ago
3 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
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
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
3 years ago
4 years ago
4 years ago
3 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
4 years ago
3 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
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
12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198219922002201220222032204220522062207220822092210221122122213221422152216221722182219222022212222222322242225222622272228222922302231223222332234223522362237223822392240224122422243224422452246224722482249225022512252225322542255225622572258
  1. package repo
  2. import (
  3. "archive/zip"
  4. "encoding/json"
  5. "errors"
  6. "io"
  7. "io/ioutil"
  8. "net/http"
  9. "os"
  10. "path"
  11. "strconv"
  12. "strings"
  13. "time"
  14. "unicode/utf8"
  15. "code.gitea.io/gitea/models"
  16. "code.gitea.io/gitea/modules/auth"
  17. "code.gitea.io/gitea/modules/base"
  18. "code.gitea.io/gitea/modules/cloudbrain"
  19. "code.gitea.io/gitea/modules/context"
  20. "code.gitea.io/gitea/modules/git"
  21. "code.gitea.io/gitea/modules/log"
  22. "code.gitea.io/gitea/modules/modelarts"
  23. "code.gitea.io/gitea/modules/obs"
  24. "code.gitea.io/gitea/modules/setting"
  25. "code.gitea.io/gitea/modules/storage"
  26. "code.gitea.io/gitea/modules/util"
  27. )
  28. const (
  29. tplDebugJobIndex base.TplName = "repo/debugjob/index"
  30. tplModelArtsNotebookIndex base.TplName = "repo/modelarts/notebook/index"
  31. tplModelArtsNotebookNew base.TplName = "repo/modelarts/notebook/new"
  32. tplModelArtsNotebookShow base.TplName = "repo/modelarts/notebook/show"
  33. tplModelArtsTrainJobIndex base.TplName = "repo/modelarts/trainjob/index"
  34. tplModelArtsTrainJobNew base.TplName = "repo/modelarts/trainjob/new"
  35. tplModelArtsTrainJobShow base.TplName = "repo/modelarts/trainjob/show"
  36. tplModelArtsTrainJobVersionNew base.TplName = "repo/modelarts/trainjob/version_new"
  37. tplModelArtsInferenceJobIndex base.TplName = "repo/modelarts/inferencejob/index"
  38. tplModelArtsInferenceJobNew base.TplName = "repo/modelarts/inferencejob/new"
  39. tplModelArtsInferenceJobShow base.TplName = "repo/modelarts/inferencejob/show"
  40. )
  41. func DebugJobIndex(ctx *context.Context) {
  42. debugListType := ctx.Query("debugListType")
  43. ctx.Data["ListType"] = debugListType
  44. MustEnableCloudbrain(ctx)
  45. repo := ctx.Repo.Repository
  46. page := ctx.QueryInt("page")
  47. if page <= 0 {
  48. page = 1
  49. }
  50. debugType := modelarts.DebugType
  51. jobTypeNot := false
  52. if debugListType == models.GPUResource {
  53. debugType = models.TypeCloudBrainOne
  54. } else if debugListType == models.NPUResource {
  55. debugType = models.TypeCloudBrainTwo
  56. }
  57. var jobTypes []string
  58. jobTypes = append(jobTypes, string(models.JobTypeSnn4imagenet), string(models.JobTypeBrainScore), string(models.JobTypeDebug))
  59. ciTasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  60. ListOptions: models.ListOptions{
  61. Page: page,
  62. PageSize: setting.UI.IssuePagingNum,
  63. },
  64. RepoID: repo.ID,
  65. Type: debugType,
  66. JobTypeNot: jobTypeNot,
  67. JobTypes: jobTypes,
  68. })
  69. if err != nil {
  70. ctx.ServerError("Get debugjob faild:", err)
  71. return
  72. }
  73. for i, task := range ciTasks {
  74. ciTasks[i].CanDebug = cloudbrain.CanModifyJob(ctx, &task.Cloudbrain)
  75. ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  76. ciTasks[i].Cloudbrain.ComputeResource = task.ComputeResource
  77. }
  78. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  79. pager.AddParam(ctx, "debugListType", "ListType")
  80. ctx.Data["Page"] = pager
  81. ctx.Data["PageIsCloudBrain"] = true
  82. ctx.Data["Tasks"] = ciTasks
  83. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  84. ctx.Data["RepoIsEmpty"] = repo.IsEmpty
  85. ctx.Data["debugListType"] = debugListType
  86. ctx.HTML(200, tplDebugJobIndex)
  87. }
  88. // MustEnableDataset check if repository enable internal cb
  89. func MustEnableModelArts(ctx *context.Context) {
  90. if !ctx.Repo.CanRead(models.UnitTypeCloudBrain) {
  91. ctx.NotFound("MustEnableCloudbrain", nil)
  92. return
  93. }
  94. }
  95. func NotebookNew(ctx *context.Context) {
  96. notebookNewDataPrepare(ctx)
  97. ctx.HTML(200, tplModelArtsNotebookNew)
  98. }
  99. func notebookNewDataPrepare(ctx *context.Context) error {
  100. ctx.Data["PageIsCloudBrain"] = true
  101. t := time.Now()
  102. var displayJobName = jobNamePrefixValid(cutString(ctx.User.Name, 5)) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  103. ctx.Data["display_job_name"] = displayJobName
  104. attachs, err := models.GetModelArtsUserAttachments(ctx.User.ID)
  105. if err != nil {
  106. ctx.ServerError("GetAllUserAttachments failed:", err)
  107. return err
  108. }
  109. ctx.Data["attachments"] = attachs
  110. if modelarts.ImageInfos == nil {
  111. json.Unmarshal([]byte(setting.ImageInfos), &modelarts.ImageInfos)
  112. }
  113. ctx.Data["images"] = modelarts.ImageInfos.ImageInfo
  114. if modelarts.FlavorInfos == nil {
  115. json.Unmarshal([]byte(setting.FlavorInfos), &modelarts.FlavorInfos)
  116. }
  117. ctx.Data["flavors"] = modelarts.FlavorInfos.FlavorInfo
  118. return nil
  119. }
  120. func NotebookCreate(ctx *context.Context, form auth.CreateModelArtsNotebookForm) {
  121. ctx.Data["PageIsNotebook"] = true
  122. jobName := form.JobName
  123. uuid := form.Attachment
  124. description := form.Description
  125. flavor := form.Flavor
  126. count, err := models.GetCloudbrainNotebookCountByUserID(ctx.User.ID)
  127. if err != nil {
  128. log.Error("GetCloudbrainNotebookCountByUserID failed:%v", err, ctx.Data["MsgID"])
  129. cloudBrainNewDataPrepare(ctx)
  130. ctx.RenderWithErr("system error", tplModelArtsNotebookNew, &form)
  131. return
  132. } else {
  133. if count >= 1 {
  134. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  135. cloudBrainNewDataPrepare(ctx)
  136. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsNotebookNew, &form)
  137. return
  138. }
  139. }
  140. _, err = models.GetCloudbrainByName(jobName)
  141. if err == nil {
  142. log.Error("the job name did already exist", ctx.Data["MsgID"])
  143. cloudBrainNewDataPrepare(ctx)
  144. ctx.RenderWithErr("the job name did already exist", tplModelArtsNotebookNew, &form)
  145. return
  146. } else {
  147. if !models.IsErrJobNotExist(err) {
  148. log.Error("system error, %v", err, ctx.Data["MsgID"])
  149. cloudBrainNewDataPrepare(ctx)
  150. ctx.RenderWithErr("system error", tplModelArtsNotebookNew, &form)
  151. return
  152. }
  153. }
  154. err = modelarts.GenerateTask(ctx, jobName, uuid, description, flavor)
  155. if err != nil {
  156. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookNew, &form)
  157. return
  158. }
  159. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=all")
  160. }
  161. func Notebook2Create(ctx *context.Context, form auth.CreateModelArtsNotebookForm) {
  162. ctx.Data["PageIsNotebook"] = true
  163. displayJobName := form.DisplayJobName
  164. jobName := util.ConvertDisplayJobNameToJobName(displayJobName)
  165. uuid := form.Attachment
  166. description := form.Description
  167. flavor := form.Flavor
  168. imageId := form.ImageId
  169. repo := ctx.Repo.Repository
  170. count, err := models.GetCloudbrainNotebookCountByUserID(ctx.User.ID)
  171. if err != nil {
  172. log.Error("GetCloudbrainNotebookCountByUserID failed:%v", err, ctx.Data["MsgID"])
  173. notebookNewDataPrepare(ctx)
  174. ctx.RenderWithErr("system error", tplModelArtsNotebookNew, &form)
  175. return
  176. } else {
  177. if count >= 1 {
  178. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  179. notebookNewDataPrepare(ctx)
  180. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsNotebookNew, &form)
  181. return
  182. }
  183. }
  184. tasks, err := models.GetCloudbrainsByDisplayJobName(repo.ID, string(models.JobTypeDebug), displayJobName)
  185. if err == nil {
  186. for _, task := range tasks {
  187. if strings.EqualFold(task.DisplayJobName, displayJobName) {
  188. log.Error("the job name did already exist", ctx.Data["MsgID"])
  189. notebookNewDataPrepare(ctx)
  190. ctx.RenderWithErr("the job name did already exist", tplModelArtsNotebookNew, &form)
  191. return
  192. }
  193. }
  194. } else {
  195. if !models.IsErrJobNotExist(err) {
  196. log.Error("system error, %v", err, ctx.Data["MsgID"])
  197. notebookNewDataPrepare(ctx)
  198. ctx.RenderWithErr("system error", tplModelArtsNotebookNew, &form)
  199. return
  200. }
  201. }
  202. err = modelarts.GenerateNotebook2(ctx, displayJobName, jobName, uuid, description, flavor, imageId)
  203. if err != nil {
  204. log.Error("GenerateNotebook2 failed, %v", err, ctx.Data["MsgID"])
  205. notebookNewDataPrepare(ctx)
  206. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookNew, &form)
  207. return
  208. }
  209. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=all")
  210. }
  211. func NotebookShow(ctx *context.Context) {
  212. ctx.Data["PageIsCloudBrain"] = true
  213. debugListType := ctx.Query("debugListType")
  214. var ID = ctx.Params(":id")
  215. task, err := models.GetCloudbrainByID(ID)
  216. if err != nil {
  217. ctx.Data["error"] = err.Error()
  218. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil)
  219. return
  220. }
  221. result, err := modelarts.GetNotebook2(task.JobID)
  222. if err != nil {
  223. ctx.Data["error"] = err.Error()
  224. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil)
  225. return
  226. }
  227. if result != nil {
  228. task.Status = result.Status
  229. err = models.UpdateJob(task)
  230. if err != nil {
  231. ctx.Data["error"] = err.Error()
  232. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil)
  233. return
  234. }
  235. result.CreateTime = time.Unix(int64(result.CreateAt/1000), 0).Format("2006-01-02 15:04:05")
  236. result.LatestUpdateTime = time.Unix(int64(result.UpdateAt/1000), 0).Format("2006-01-02 15:04:05")
  237. }
  238. datasetDownloadLink := "-"
  239. if ctx.IsSigned {
  240. if task.Uuid != "" && task.UserID == ctx.User.ID {
  241. attachment, err := models.GetAttachmentByUUID(task.Uuid)
  242. if err == nil {
  243. datasetDownloadLink = attachment.S3DownloadURL()
  244. }
  245. }
  246. }
  247. ctx.Data["datasetDownloadLink"] = datasetDownloadLink
  248. ctx.Data["task"] = task
  249. ctx.Data["ID"] = ID
  250. ctx.Data["jobName"] = task.JobName
  251. ctx.Data["result"] = result
  252. ctx.Data["debugListType"] = debugListType
  253. ctx.HTML(200, tplModelArtsNotebookShow)
  254. }
  255. func NotebookDebug(ctx *context.Context) {
  256. var jobID = ctx.Params(":jobid")
  257. result, err := modelarts.GetJob(jobID)
  258. if err != nil {
  259. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil)
  260. return
  261. }
  262. res, err := modelarts.GetJobToken(jobID)
  263. if err != nil {
  264. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil)
  265. return
  266. }
  267. urls := strings.Split(result.Spec.Annotations.Url, "/")
  268. urlPrefix := result.Spec.Annotations.TargetDomain
  269. for i, url := range urls {
  270. if i > 2 {
  271. urlPrefix += "/" + url
  272. }
  273. }
  274. debugUrl := urlPrefix + "?token=" + res.Token
  275. ctx.Redirect(debugUrl)
  276. }
  277. func NotebookDebug2(ctx *context.Context) {
  278. task := ctx.Cloudbrain
  279. result, err := modelarts.GetNotebook2(task.JobID)
  280. if err != nil {
  281. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil)
  282. return
  283. }
  284. ctx.Redirect(result.Url + "?token=" + result.Token)
  285. }
  286. func NotebookManage(ctx *context.Context) {
  287. var ID = ctx.Params(":id")
  288. var action = ctx.Params(":action")
  289. var resultCode = "0"
  290. var errorMsg = ""
  291. var status = ""
  292. for {
  293. task, err := models.GetCloudbrainByID(ID)
  294. if err != nil {
  295. log.Error("get task(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"])
  296. resultCode = "-1"
  297. errorMsg = "system error"
  298. break
  299. }
  300. if action == models.ActionStop {
  301. if task.Status != string(models.ModelArtsRunning) {
  302. log.Error("the job(%s) is not running", task.JobName, ctx.Data["MsgID"])
  303. resultCode = "-1"
  304. errorMsg = "the job is not running"
  305. break
  306. }
  307. if !ctx.IsSigned || (ctx.User.ID != task.UserID && !ctx.IsUserSiteAdmin() && !ctx.IsUserRepoOwner()) {
  308. log.Error("the user has no right ro stop the job", task.JobName, ctx.Data["MsgID"])
  309. resultCode = "-1"
  310. errorMsg = "you have no right to stop the job"
  311. break
  312. }
  313. } else if action == models.ActionRestart {
  314. ctx.CheckWechatBind()
  315. if ctx.Written() {
  316. return
  317. }
  318. if task.Status != string(models.ModelArtsStopped) && task.Status != string(models.ModelArtsStartFailed) && task.Status != string(models.ModelArtsCreateFailed) {
  319. log.Error("the job(%s) is not stopped", task.JobName, ctx.Data["MsgID"])
  320. resultCode = "-1"
  321. errorMsg = "the job is not stopped"
  322. break
  323. }
  324. if !ctx.IsSigned || (ctx.User.ID != task.UserID && !ctx.IsUserSiteAdmin()) {
  325. log.Error("the user has no right ro restart the job", task.JobName, ctx.Data["MsgID"])
  326. resultCode = "-1"
  327. errorMsg = "you have no right to restart the job"
  328. break
  329. }
  330. count, err := models.GetCloudbrainNotebookCountByUserID(ctx.User.ID)
  331. if err != nil {
  332. log.Error("GetCloudbrainNotebookCountByUserID failed:%v", err, ctx.Data["MsgID"])
  333. resultCode = "-1"
  334. errorMsg = "system error"
  335. break
  336. } else {
  337. if count >= 1 {
  338. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  339. resultCode = "-1"
  340. errorMsg = "you have already a running or waiting task, can not create more"
  341. break
  342. }
  343. }
  344. action = models.ActionStart
  345. } else {
  346. log.Error("the action(%s) is illegal", action, ctx.Data["MsgID"])
  347. resultCode = "-1"
  348. errorMsg = "非法操作"
  349. break
  350. }
  351. param := models.NotebookAction{
  352. Action: action,
  353. }
  354. res, err := modelarts.ManageNotebook2(task.JobID, param)
  355. if err != nil {
  356. log.Error("ManageNotebook2(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"])
  357. resultCode = "-1"
  358. errorMsg = err.Error()
  359. if strings.Contains(err.Error(), modelarts.NotebookNotFound) {
  360. errorMsg = "the job's version is too old and can not be restarted"
  361. }
  362. break
  363. }
  364. task.Status = res.Status
  365. err = models.UpdateJob(task)
  366. if err != nil {
  367. log.Error("UpdateJob(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"])
  368. resultCode = "-1"
  369. errorMsg = "system error"
  370. break
  371. }
  372. status = task.Status
  373. break
  374. }
  375. ctx.JSON(200, map[string]string{
  376. "result_code": resultCode,
  377. "error_msg": errorMsg,
  378. "status": status,
  379. "id": ID,
  380. })
  381. }
  382. func NotebookDel(ctx *context.Context) {
  383. var listType = ctx.Query("debugListType")
  384. task := ctx.Cloudbrain
  385. if task.Status != string(models.ModelArtsCreateFailed) && task.Status != string(models.ModelArtsStartFailed) && task.Status != string(models.ModelArtsStopped) {
  386. log.Error("the job(%s) has not been stopped", task.JobName)
  387. ctx.RenderWithErr("the job has not been stopped", tplDebugJobIndex, nil)
  388. return
  389. }
  390. _, err := modelarts.DelNotebook2(task.JobID)
  391. if err != nil {
  392. log.Error("DelNotebook2(%s) failed:%v", task.JobName, err.Error())
  393. if strings.Contains(err.Error(), modelarts.NotebookNotFound) || strings.Contains(err.Error(), modelarts.NotebookNoPermission) || strings.Contains(err.Error(), modelarts.NotebookInvalid) {
  394. log.Info("old notebook version")
  395. } else {
  396. ctx.RenderWithErr(err.Error(), tplDebugJobIndex, nil)
  397. return
  398. }
  399. }
  400. err = models.DeleteJob(task)
  401. if err != nil {
  402. ctx.RenderWithErr(err.Error(), tplDebugJobIndex, nil)
  403. return
  404. }
  405. var isAdminPage = ctx.Query("isadminpage")
  406. if ctx.IsUserSiteAdmin() && isAdminPage == "true" {
  407. ctx.Redirect(setting.AppSubURL + "/admin" + "/cloudbrains")
  408. } else {
  409. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=" + listType)
  410. }
  411. }
  412. func TrainJobIndex(ctx *context.Context) {
  413. MustEnableModelArts(ctx)
  414. repo := ctx.Repo.Repository
  415. page := ctx.QueryInt("page")
  416. if page <= 0 {
  417. page = 1
  418. }
  419. var jobTypes []string
  420. jobTypes = append(jobTypes, string(models.JobTypeTrain))
  421. tasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  422. ListOptions: models.ListOptions{
  423. Page: page,
  424. PageSize: setting.UI.IssuePagingNum,
  425. },
  426. RepoID: repo.ID,
  427. Type: models.TypeCloudBrainTwo,
  428. JobTypeNot: false,
  429. JobTypes: jobTypes,
  430. IsLatestVersion: modelarts.IsLatestVersion,
  431. })
  432. if err != nil {
  433. ctx.ServerError("Cloudbrain", err)
  434. return
  435. }
  436. for i, task := range tasks {
  437. tasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  438. tasks[i].CanModify = cloudbrain.CanModifyJob(ctx, &task.Cloudbrain)
  439. tasks[i].ComputeResource = models.NPUResource
  440. }
  441. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  442. pager.SetDefaultParams(ctx)
  443. ctx.Data["Page"] = pager
  444. ctx.Data["PageIsCloudBrain"] = true
  445. ctx.Data["Tasks"] = tasks
  446. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  447. ctx.Data["RepoIsEmpty"] = repo.IsEmpty
  448. ctx.HTML(200, tplModelArtsTrainJobIndex)
  449. }
  450. func TrainJobNew(ctx *context.Context) {
  451. err := trainJobNewDataPrepare(ctx)
  452. if err != nil {
  453. ctx.ServerError("get new train-job info failed", err)
  454. return
  455. }
  456. ctx.HTML(200, tplModelArtsTrainJobNew)
  457. }
  458. func trainJobNewDataPrepare(ctx *context.Context) error {
  459. ctx.Data["PageIsCloudBrain"] = true
  460. //can, err := canUserCreateTrainJob(ctx.User.ID)
  461. //if err != nil {
  462. // ctx.ServerError("canUserCreateTrainJob", err)
  463. // return
  464. //}
  465. //
  466. //if !can {
  467. // log.Error("the user can not create train-job")
  468. // ctx.ServerError("the user can not create train-job", fmt.Errorf("the user can not create train-job"))
  469. // return
  470. //}
  471. t := time.Now()
  472. var displayJobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  473. ctx.Data["display_job_name"] = displayJobName
  474. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  475. if err != nil {
  476. ctx.ServerError("GetAllUserAttachments failed:", err)
  477. return err
  478. }
  479. ctx.Data["attachments"] = attachs
  480. var resourcePools modelarts.ResourcePool
  481. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  482. ctx.ServerError("json.Unmarshal failed:", err)
  483. return err
  484. }
  485. ctx.Data["resource_pools"] = resourcePools.Info
  486. var engines modelarts.Engine
  487. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  488. ctx.ServerError("json.Unmarshal failed:", err)
  489. return err
  490. }
  491. ctx.Data["engines"] = engines.Info
  492. var versionInfos modelarts.VersionInfo
  493. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  494. ctx.ServerError("json.Unmarshal failed:", err)
  495. return err
  496. }
  497. ctx.Data["engine_versions"] = versionInfos.Version
  498. var flavorInfos modelarts.Flavor
  499. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  500. ctx.ServerError("json.Unmarshal failed:", err)
  501. return err
  502. }
  503. ctx.Data["flavor_infos"] = flavorInfos.Info
  504. ctx.Data["params"] = ""
  505. ctx.Data["branchName"] = ctx.Repo.BranchName
  506. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  507. if err != nil {
  508. ctx.ServerError("getConfigList failed:", err)
  509. return err
  510. }
  511. ctx.Data["config_list"] = configList.ParaConfigs
  512. return nil
  513. }
  514. func trainJobErrorNewDataPrepare(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) error {
  515. ctx.Data["PageIsCloudBrain"] = true
  516. //can, err := canUserCreateTrainJob(ctx.User.ID)
  517. //if err != nil {
  518. // ctx.ServerError("canUserCreateTrainJob", err)
  519. // return
  520. //}
  521. //
  522. //if !can {
  523. // log.Error("the user can not create train-job")
  524. // ctx.ServerError("the user can not create train-job", fmt.Errorf("the user can not create train-job"))
  525. // return
  526. //}
  527. t := time.Now()
  528. var displayJobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  529. ctx.Data["display_job_name"] = displayJobName
  530. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  531. if err != nil {
  532. ctx.ServerError("GetAllUserAttachments failed:", err)
  533. return err
  534. }
  535. ctx.Data["attachments"] = attachs
  536. var resourcePools modelarts.ResourcePool
  537. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  538. ctx.ServerError("json.Unmarshal failed:", err)
  539. return err
  540. }
  541. ctx.Data["resource_pools"] = resourcePools.Info
  542. var engines modelarts.Engine
  543. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  544. ctx.ServerError("json.Unmarshal failed:", err)
  545. return err
  546. }
  547. ctx.Data["engines"] = engines.Info
  548. var versionInfos modelarts.VersionInfo
  549. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  550. ctx.ServerError("json.Unmarshal failed:", err)
  551. return err
  552. }
  553. ctx.Data["engine_versions"] = versionInfos.Version
  554. var flavorInfos modelarts.Flavor
  555. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  556. ctx.ServerError("json.Unmarshal failed:", err)
  557. return err
  558. }
  559. ctx.Data["flavor_infos"] = flavorInfos.Info
  560. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  561. if err != nil {
  562. ctx.ServerError("getConfigList failed:", err)
  563. return err
  564. }
  565. var Parameters modelarts.Parameters
  566. if err = json.Unmarshal([]byte(form.Params), &Parameters); err != nil {
  567. ctx.ServerError("json.Unmarshal failed:", err)
  568. return err
  569. }
  570. ctx.Data["params"] = Parameters.Parameter
  571. ctx.Data["config_list"] = configList.ParaConfigs
  572. ctx.Data["bootFile"] = form.BootFile
  573. ctx.Data["uuid"] = form.Attachment
  574. ctx.Data["branch_name"] = form.BranchName
  575. return nil
  576. }
  577. func TrainJobNewVersion(ctx *context.Context) {
  578. err := trainJobNewVersionDataPrepare(ctx)
  579. if err != nil {
  580. ctx.ServerError("get new train-job info failed", err)
  581. return
  582. }
  583. ctx.HTML(200, tplModelArtsTrainJobVersionNew)
  584. }
  585. func trainJobNewVersionDataPrepare(ctx *context.Context) error {
  586. ctx.Data["PageIsCloudBrain"] = true
  587. var jobID = ctx.Params(":jobid")
  588. var versionName = ctx.Query("version_name")
  589. // canNewJob, err := canUserCreateTrainJobVersion(ctx, jobID, versionName)
  590. // if err != nil {
  591. // ctx.ServerError("canNewJob can info failed", err)
  592. // return err
  593. // }
  594. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  595. if err != nil {
  596. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  597. return err
  598. }
  599. ctx.Data["display_job_name"] = task.DisplayJobName
  600. ctx.Data["job_name"] = task.JobName
  601. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  602. if err != nil {
  603. ctx.ServerError("GetAllUserAttachments failed:", err)
  604. return err
  605. }
  606. ctx.Data["attachments"] = attachs
  607. var resourcePools modelarts.ResourcePool
  608. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  609. ctx.ServerError("json.Unmarshal failed:", err)
  610. return err
  611. }
  612. ctx.Data["resource_pools"] = resourcePools.Info
  613. var engines modelarts.Engine
  614. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  615. ctx.ServerError("json.Unmarshal failed:", err)
  616. return err
  617. }
  618. ctx.Data["engines"] = engines.Info
  619. var versionInfos modelarts.VersionInfo
  620. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  621. ctx.ServerError("json.Unmarshal failed:", err)
  622. return err
  623. }
  624. ctx.Data["engine_versions"] = versionInfos.Version
  625. var flavorInfos modelarts.Flavor
  626. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  627. ctx.ServerError("json.Unmarshal failed:", err)
  628. return err
  629. }
  630. ctx.Data["flavor_infos"] = flavorInfos.Info
  631. var Parameters modelarts.Parameters
  632. if err = json.Unmarshal([]byte(task.Parameters), &Parameters); err != nil {
  633. ctx.ServerError("json.Unmarshal failed:", err)
  634. return err
  635. }
  636. ctx.Data["params"] = Parameters.Parameter
  637. branches, _, err := ctx.Repo.GitRepo.GetBranches(0, 0)
  638. if err != nil {
  639. ctx.ServerError("GetBranches error:", err)
  640. return err
  641. }
  642. ctx.Data["branches"] = branches
  643. ctx.Data["branch_name"] = task.BranchName
  644. ctx.Data["description"] = task.Description
  645. ctx.Data["boot_file"] = task.BootFile
  646. ctx.Data["dataset_name"] = task.DatasetName
  647. ctx.Data["work_server_number"] = task.WorkServerNumber
  648. ctx.Data["flavor_name"] = task.FlavorName
  649. ctx.Data["engine_name"] = task.EngineName
  650. ctx.Data["uuid"] = task.Uuid
  651. ctx.Data["flavor_code"] = task.FlavorCode
  652. ctx.Data["engine_id"] = task.EngineID
  653. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  654. if err != nil {
  655. ctx.ServerError("getConfigList failed:", err)
  656. return err
  657. }
  658. ctx.Data["config_list"] = configList.ParaConfigs
  659. return nil
  660. }
  661. func versionErrorDataPrepare(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) error {
  662. ctx.Data["PageIsCloudBrain"] = true
  663. var jobID = ctx.Params(":jobid")
  664. // var versionName = ctx.Params(":version-name")
  665. var versionName = ctx.Query("version_name")
  666. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  667. if err != nil {
  668. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  669. return err
  670. }
  671. t := time.Now()
  672. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  673. ctx.Data["job_name"] = task.JobName
  674. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  675. if err != nil {
  676. ctx.ServerError("GetAllUserAttachments failed:", err)
  677. return err
  678. }
  679. ctx.Data["attachments"] = attachs
  680. var resourcePools modelarts.ResourcePool
  681. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  682. ctx.ServerError("json.Unmarshal failed:", err)
  683. return err
  684. }
  685. ctx.Data["resource_pools"] = resourcePools.Info
  686. var engines modelarts.Engine
  687. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  688. ctx.ServerError("json.Unmarshal failed:", err)
  689. return err
  690. }
  691. ctx.Data["engines"] = engines.Info
  692. var versionInfos modelarts.VersionInfo
  693. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  694. ctx.ServerError("json.Unmarshal failed:", err)
  695. return err
  696. }
  697. ctx.Data["engine_versions"] = versionInfos.Version
  698. var flavorInfos modelarts.Flavor
  699. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  700. ctx.ServerError("json.Unmarshal failed:", err)
  701. return err
  702. }
  703. ctx.Data["flavor_infos"] = flavorInfos.Info
  704. var Parameters modelarts.Parameters
  705. if err = json.Unmarshal([]byte(form.Params), &Parameters); err != nil {
  706. ctx.ServerError("json.Unmarshal failed:", err)
  707. return err
  708. }
  709. ctx.Data["params"] = Parameters.Parameter
  710. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  711. ctx.Data["train_url"] = outputObsPath
  712. branches, _, err := ctx.Repo.GitRepo.GetBranches(0, 0)
  713. if err != nil {
  714. ctx.ServerError("GetBranches error:", err)
  715. return err
  716. }
  717. ctx.Data["branches"] = branches
  718. ctx.Data["description"] = form.Description
  719. ctx.Data["dataset_name"] = task.DatasetName
  720. ctx.Data["work_server_number"] = form.WorkServerNumber
  721. ctx.Data["flavor_name"] = form.FlavorName
  722. ctx.Data["engine_name"] = form.EngineName
  723. ctx.Data["flavor_code"] = task.FlavorCode
  724. ctx.Data["engine_id"] = task.EngineID
  725. ctx.Data["version_name"] = form.VersionName
  726. ctx.Data["bootFile"] = form.BootFile
  727. ctx.Data["uuid"] = form.Attachment
  728. ctx.Data["branch_name"] = form.BranchName
  729. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  730. if err != nil {
  731. ctx.ServerError("getConfigList failed:", err)
  732. return err
  733. }
  734. ctx.Data["config_list"] = configList.ParaConfigs
  735. return nil
  736. }
  737. func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) {
  738. ctx.Data["PageIsTrainJob"] = true
  739. VersionOutputPath := modelarts.GetOutputPathByCount(modelarts.TotalVersionCount)
  740. displayJobName := form.DisplayJobName
  741. jobName := util.ConvertDisplayJobNameToJobName(displayJobName)
  742. uuid := form.Attachment
  743. description := form.Description
  744. workServerNumber := form.WorkServerNumber
  745. engineID := form.EngineID
  746. bootFile := form.BootFile
  747. flavorCode := form.Flavor
  748. params := form.Params
  749. poolID := form.PoolID
  750. isSaveParam := form.IsSaveParam
  751. repo := ctx.Repo.Repository
  752. codeLocalPath := setting.JobPath + jobName + modelarts.CodePath
  753. codeObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.CodePath
  754. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath + VersionOutputPath + "/"
  755. logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath + VersionOutputPath + "/"
  756. dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/"
  757. branch_name := form.BranchName
  758. isLatestVersion := modelarts.IsLatestVersion
  759. FlavorName := form.FlavorName
  760. VersionCount := modelarts.VersionCount
  761. EngineName := form.EngineName
  762. count, err := models.GetCloudbrainTrainJobCountByUserID(ctx.User.ID)
  763. if err != nil {
  764. log.Error("GetCloudbrainTrainJobCountByUserID failed:%v", err, ctx.Data["MsgID"])
  765. trainJobErrorNewDataPrepare(ctx, form)
  766. ctx.RenderWithErr("system error", tplModelArtsTrainJobNew, &form)
  767. return
  768. } else {
  769. if count >= 1 {
  770. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  771. trainJobErrorNewDataPrepare(ctx, form)
  772. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsTrainJobNew, &form)
  773. return
  774. }
  775. }
  776. if err := paramCheckCreateTrainJob(form); err != nil {
  777. log.Error("paramCheckCreateTrainJob failed:(%v)", err)
  778. trainJobErrorNewDataPrepare(ctx, form)
  779. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobNew, &form)
  780. return
  781. }
  782. //Determine whether the task name of the task in the project is duplicated
  783. tasks, err := models.GetCloudbrainsByDisplayJobName(repo.ID, string(models.JobTypeTrain), displayJobName)
  784. if err == nil {
  785. if len(tasks) != 0 {
  786. log.Error("the job name did already exist", ctx.Data["MsgID"])
  787. trainJobErrorNewDataPrepare(ctx, form)
  788. ctx.RenderWithErr("the job name did already exist", tplModelArtsTrainJobNew, &form)
  789. return
  790. }
  791. } else {
  792. if !models.IsErrJobNotExist(err) {
  793. log.Error("system error, %v", err, ctx.Data["MsgID"])
  794. trainJobErrorNewDataPrepare(ctx, form)
  795. ctx.RenderWithErr("system error", tplModelArtsTrainJobNew, &form)
  796. return
  797. }
  798. }
  799. //todo: del the codeLocalPath
  800. _, err = ioutil.ReadDir(codeLocalPath)
  801. if err == nil {
  802. os.RemoveAll(codeLocalPath)
  803. }
  804. gitRepo, _ := git.OpenRepository(repo.RepoPath())
  805. commitID, _ := gitRepo.GetBranchCommitID(branch_name)
  806. if err := git.Clone(repo.RepoPath(), codeLocalPath, git.CloneRepoOptions{
  807. Branch: branch_name,
  808. }); err != nil {
  809. log.Error("Create task failed, server timed out: %s (%v)", repo.FullName(), err)
  810. trainJobErrorNewDataPrepare(ctx, form)
  811. ctx.RenderWithErr("Create task failed, server timed out", tplModelArtsTrainJobNew, &form)
  812. return
  813. }
  814. //todo: upload code (send to file_server todo this work?)
  815. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.OutputPath + VersionOutputPath + "/"); err != nil {
  816. log.Error("Failed to obsMkdir_output: %s (%v)", repo.FullName(), err)
  817. trainJobErrorNewDataPrepare(ctx, form)
  818. ctx.RenderWithErr("Failed to obsMkdir_output", tplModelArtsTrainJobNew, &form)
  819. return
  820. }
  821. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.LogPath + VersionOutputPath + "/"); err != nil {
  822. log.Error("Failed to obsMkdir_log: %s (%v)", repo.FullName(), err)
  823. trainJobErrorNewDataPrepare(ctx, form)
  824. ctx.RenderWithErr("Failed to obsMkdir_log", tplModelArtsTrainJobNew, &form)
  825. return
  826. }
  827. // parentDir := VersionOutputPath + "/"
  828. if err := uploadCodeToObs(codeLocalPath, jobName, ""); err != nil {
  829. // if err := uploadCodeToObs(codeLocalPath, jobName, parentDir); err != nil {
  830. log.Error("Failed to uploadCodeToObs: %s (%v)", repo.FullName(), err)
  831. trainJobErrorNewDataPrepare(ctx, form)
  832. ctx.RenderWithErr("Failed to uploadCodeToObs", tplModelArtsTrainJobNew, &form)
  833. return
  834. }
  835. //todo: del local code?
  836. var parameters models.Parameters
  837. param := make([]models.Parameter, 0)
  838. param = append(param, models.Parameter{
  839. Label: modelarts.TrainUrl,
  840. Value: outputObsPath,
  841. }, models.Parameter{
  842. Label: modelarts.DataUrl,
  843. Value: dataPath,
  844. })
  845. if len(params) != 0 {
  846. err := json.Unmarshal([]byte(params), &parameters)
  847. if err != nil {
  848. log.Error("Failed to Unmarshal params: %s (%v)", params, err)
  849. trainJobErrorNewDataPrepare(ctx, form)
  850. ctx.RenderWithErr("运行参数错误", tplModelArtsTrainJobNew, &form)
  851. return
  852. }
  853. for _, parameter := range parameters.Parameter {
  854. if parameter.Label != modelarts.TrainUrl && parameter.Label != modelarts.DataUrl {
  855. param = append(param, models.Parameter{
  856. Label: parameter.Label,
  857. Value: parameter.Value,
  858. })
  859. }
  860. }
  861. }
  862. //save param config
  863. if isSaveParam == "on" {
  864. if form.ParameterTemplateName == "" {
  865. log.Error("ParameterTemplateName is empty")
  866. trainJobNewDataPrepare(ctx)
  867. ctx.RenderWithErr("保存作业参数时,作业参数名称不能为空", tplModelArtsTrainJobNew, &form)
  868. return
  869. }
  870. _, err := modelarts.CreateTrainJobConfig(models.CreateConfigParams{
  871. ConfigName: form.ParameterTemplateName,
  872. Description: form.PrameterDescription,
  873. DataUrl: dataPath,
  874. AppUrl: codeObsPath,
  875. BootFileUrl: codeObsPath + bootFile,
  876. TrainUrl: outputObsPath,
  877. Flavor: models.Flavor{
  878. Code: flavorCode,
  879. },
  880. WorkServerNum: workServerNumber,
  881. EngineID: int64(engineID),
  882. LogUrl: logObsPath,
  883. PoolID: poolID,
  884. Parameter: param,
  885. })
  886. if err != nil {
  887. log.Error("Failed to CreateTrainJobConfig: %v", err)
  888. trainJobErrorNewDataPrepare(ctx, form)
  889. ctx.RenderWithErr("保存作业参数失败:"+err.Error(), tplModelArtsTrainJobNew, &form)
  890. return
  891. }
  892. }
  893. req := &modelarts.GenerateTrainJobReq{
  894. JobName: jobName,
  895. DisplayJobName: displayJobName,
  896. DataUrl: dataPath,
  897. Description: description,
  898. CodeObsPath: codeObsPath,
  899. BootFileUrl: codeObsPath + bootFile,
  900. BootFile: bootFile,
  901. TrainUrl: outputObsPath,
  902. FlavorCode: flavorCode,
  903. WorkServerNumber: workServerNumber,
  904. EngineID: int64(engineID),
  905. LogUrl: logObsPath,
  906. PoolID: poolID,
  907. Uuid: uuid,
  908. Parameters: parameters.Parameter,
  909. CommitID: commitID,
  910. IsLatestVersion: isLatestVersion,
  911. BranchName: branch_name,
  912. Params: form.Params,
  913. FlavorName: FlavorName,
  914. EngineName: EngineName,
  915. VersionCount: VersionCount,
  916. TotalVersionCount: modelarts.TotalVersionCount,
  917. }
  918. //将params转换Parameters.Parameter,出错时返回给前端
  919. var Parameters modelarts.Parameters
  920. if err := json.Unmarshal([]byte(params), &Parameters); err != nil {
  921. ctx.ServerError("json.Unmarshal failed:", err)
  922. return
  923. }
  924. err = modelarts.GenerateTrainJob(ctx, req)
  925. if err != nil {
  926. log.Error("GenerateTrainJob failed:%v", err.Error())
  927. trainJobErrorNewDataPrepare(ctx, form)
  928. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobNew, &form)
  929. return
  930. }
  931. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  932. }
  933. func TrainJobCreateVersion(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) {
  934. ctx.Data["PageIsTrainJob"] = true
  935. var jobID = ctx.Params(":jobid")
  936. count, err := models.GetCloudbrainTrainJobCountByUserID(ctx.User.ID)
  937. if err != nil {
  938. log.Error("GetCloudbrainTrainJobCountByUserID failed:%v", err, ctx.Data["MsgID"])
  939. versionErrorDataPrepare(ctx, form)
  940. ctx.RenderWithErr("system error", tplModelArtsTrainJobVersionNew, &form)
  941. return
  942. } else {
  943. if count >= 1 {
  944. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  945. versionErrorDataPrepare(ctx, form)
  946. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsTrainJobVersionNew, &form)
  947. return
  948. }
  949. }
  950. latestTask, err := models.GetCloudbrainByJobIDAndIsLatestVersion(jobID, modelarts.IsLatestVersion)
  951. if err != nil {
  952. ctx.ServerError("GetCloudbrainByJobIDAndIsLatestVersion faild:", err)
  953. return
  954. }
  955. VersionOutputPath := modelarts.GetOutputPathByCount(latestTask.TotalVersionCount + 1)
  956. displayJobName := form.DisplayJobName
  957. jobName := form.JobName
  958. uuid := form.Attachment
  959. description := form.Description
  960. workServerNumber := form.WorkServerNumber
  961. engineID := form.EngineID
  962. bootFile := form.BootFile
  963. flavorCode := form.Flavor
  964. params := form.Params
  965. poolID := form.PoolID
  966. isSaveParam := form.IsSaveParam
  967. repo := ctx.Repo.Repository
  968. codeLocalPath := setting.JobPath + jobName + modelarts.CodePath
  969. codeObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.CodePath + VersionOutputPath + "/"
  970. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath + VersionOutputPath + "/"
  971. logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath + VersionOutputPath + "/"
  972. dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/"
  973. branch_name := form.BranchName
  974. PreVersionName := form.VersionName
  975. FlavorName := form.FlavorName
  976. EngineName := form.EngineName
  977. isLatestVersion := modelarts.IsLatestVersion
  978. canNewJob, _ := canUserCreateTrainJobVersion(ctx, latestTask.UserID)
  979. if !canNewJob {
  980. ctx.RenderWithErr("user cann't new trainjob", tplModelArtsTrainJobVersionNew, &form)
  981. return
  982. }
  983. if err := paramCheckCreateTrainJob(form); err != nil {
  984. log.Error("paramCheckCreateTrainJob failed:(%v)", err)
  985. versionErrorDataPrepare(ctx, form)
  986. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  987. return
  988. }
  989. //todo: del the codeLocalPath
  990. _, err = ioutil.ReadDir(codeLocalPath)
  991. if err == nil {
  992. os.RemoveAll(codeLocalPath)
  993. }
  994. gitRepo, _ := git.OpenRepository(repo.RepoPath())
  995. commitID, _ := gitRepo.GetBranchCommitID(branch_name)
  996. if err := git.Clone(repo.RepoPath(), codeLocalPath, git.CloneRepoOptions{
  997. Branch: branch_name,
  998. }); err != nil {
  999. log.Error("Failed git clone repo to local(!: %s (%v)", repo.FullName(), err)
  1000. versionErrorDataPrepare(ctx, form)
  1001. ctx.RenderWithErr("Failed git clone repo to local!", tplModelArtsTrainJobVersionNew, &form)
  1002. return
  1003. }
  1004. //todo: upload code (send to file_server todo this work?)
  1005. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.OutputPath + VersionOutputPath + "/"); err != nil {
  1006. log.Error("Failed to obsMkdir_output: %s (%v)", repo.FullName(), err)
  1007. versionErrorDataPrepare(ctx, form)
  1008. ctx.RenderWithErr("Failed to obsMkdir_output", tplModelArtsTrainJobVersionNew, &form)
  1009. return
  1010. }
  1011. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.LogPath + VersionOutputPath + "/"); err != nil {
  1012. log.Error("Failed to obsMkdir_log: %s (%v)", repo.FullName(), err)
  1013. versionErrorDataPrepare(ctx, form)
  1014. ctx.RenderWithErr("Failed to obsMkdir_log", tplModelArtsTrainJobVersionNew, &form)
  1015. return
  1016. }
  1017. parentDir := VersionOutputPath + "/"
  1018. // parentDir := ""
  1019. // if err := uploadCodeToObs(codeLocalPath, jobName, ""); err != nil {
  1020. if err := uploadCodeToObs(codeLocalPath, jobName, parentDir); err != nil {
  1021. log.Error("Failed to uploadCodeToObs: %s (%v)", repo.FullName(), err)
  1022. versionErrorDataPrepare(ctx, form)
  1023. ctx.RenderWithErr("Failed to uploadCodeToObs", tplModelArtsTrainJobVersionNew, &form)
  1024. return
  1025. }
  1026. //todo: del local code?
  1027. var parameters models.Parameters
  1028. param := make([]models.Parameter, 0)
  1029. param = append(param, models.Parameter{
  1030. Label: modelarts.TrainUrl,
  1031. Value: outputObsPath,
  1032. }, models.Parameter{
  1033. Label: modelarts.DataUrl,
  1034. Value: dataPath,
  1035. })
  1036. if len(params) != 0 {
  1037. err := json.Unmarshal([]byte(params), &parameters)
  1038. if err != nil {
  1039. log.Error("Failed to Unmarshal params: %s (%v)", params, err)
  1040. versionErrorDataPrepare(ctx, form)
  1041. ctx.RenderWithErr("运行参数错误", tplModelArtsTrainJobVersionNew, &form)
  1042. return
  1043. }
  1044. for _, parameter := range parameters.Parameter {
  1045. if parameter.Label != modelarts.TrainUrl && parameter.Label != modelarts.DataUrl {
  1046. param = append(param, models.Parameter{
  1047. Label: parameter.Label,
  1048. Value: parameter.Value,
  1049. })
  1050. }
  1051. }
  1052. }
  1053. //save param config
  1054. if isSaveParam == "on" {
  1055. if form.ParameterTemplateName == "" {
  1056. log.Error("ParameterTemplateName is empty")
  1057. versionErrorDataPrepare(ctx, form)
  1058. ctx.RenderWithErr("保存作业参数时,作业参数名称不能为空", tplModelArtsTrainJobVersionNew, &form)
  1059. return
  1060. }
  1061. _, err := modelarts.CreateTrainJobConfig(models.CreateConfigParams{
  1062. ConfigName: form.ParameterTemplateName,
  1063. Description: form.PrameterDescription,
  1064. DataUrl: dataPath,
  1065. AppUrl: codeObsPath,
  1066. BootFileUrl: codeObsPath + bootFile,
  1067. TrainUrl: outputObsPath,
  1068. Flavor: models.Flavor{
  1069. Code: flavorCode,
  1070. },
  1071. WorkServerNum: workServerNumber,
  1072. EngineID: int64(engineID),
  1073. LogUrl: logObsPath,
  1074. PoolID: poolID,
  1075. Parameter: parameters.Parameter,
  1076. })
  1077. if err != nil {
  1078. log.Error("Failed to CreateTrainJobConfig: %v", err)
  1079. versionErrorDataPrepare(ctx, form)
  1080. ctx.RenderWithErr("保存作业参数失败:"+err.Error(), tplModelArtsTrainJobVersionNew, &form)
  1081. return
  1082. }
  1083. }
  1084. if err != nil {
  1085. log.Error("getFlavorNameByEngineID(%s) failed:%v", engineID, err.Error())
  1086. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  1087. return
  1088. }
  1089. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, PreVersionName)
  1090. if err != nil {
  1091. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  1092. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  1093. return
  1094. }
  1095. req := &modelarts.GenerateTrainJobReq{
  1096. JobName: jobName,
  1097. DisplayJobName: displayJobName,
  1098. DataUrl: dataPath,
  1099. Description: description,
  1100. CodeObsPath: codeObsPath,
  1101. BootFileUrl: codeObsPath + bootFile,
  1102. BootFile: bootFile,
  1103. TrainUrl: outputObsPath,
  1104. FlavorCode: flavorCode,
  1105. WorkServerNumber: workServerNumber,
  1106. IsLatestVersion: isLatestVersion,
  1107. EngineID: int64(engineID),
  1108. LogUrl: logObsPath,
  1109. PoolID: poolID,
  1110. Uuid: uuid,
  1111. Params: form.Params,
  1112. Parameters: parameters.Parameter,
  1113. PreVersionId: task.VersionID,
  1114. CommitID: commitID,
  1115. BranchName: branch_name,
  1116. FlavorName: FlavorName,
  1117. EngineName: EngineName,
  1118. PreVersionName: PreVersionName,
  1119. TotalVersionCount: latestTask.TotalVersionCount + 1,
  1120. }
  1121. err = modelarts.GenerateTrainJobVersion(ctx, req, jobID)
  1122. if err != nil {
  1123. log.Error("GenerateTrainJob failed:%v", err.Error())
  1124. versionErrorDataPrepare(ctx, form)
  1125. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  1126. return
  1127. }
  1128. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job/" + jobID)
  1129. // ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1130. }
  1131. // readDir reads the directory named by dirname and returns
  1132. // a list of directory entries sorted by filename.
  1133. func readDir(dirname string) ([]os.FileInfo, error) {
  1134. f, err := os.Open(dirname)
  1135. if err != nil {
  1136. return nil, err
  1137. }
  1138. list, err := f.Readdir(100)
  1139. f.Close()
  1140. if err != nil {
  1141. //todo: can not upload empty folder
  1142. if err == io.EOF {
  1143. return nil, nil
  1144. }
  1145. return nil, err
  1146. }
  1147. //sort.Slice(list, func(i, j int) bool { return list[i].Name() < list[j].Name() })
  1148. return list, nil
  1149. }
  1150. func uploadCodeToObs(codePath, jobName, parentDir string) error {
  1151. files, err := readDir(codePath)
  1152. if err != nil {
  1153. log.Error("readDir(%s) failed: %s", codePath, err.Error())
  1154. return err
  1155. }
  1156. for _, file := range files {
  1157. if file.IsDir() {
  1158. input := &obs.PutObjectInput{}
  1159. input.Bucket = setting.Bucket
  1160. input.Key = parentDir + file.Name() + "/"
  1161. _, err = storage.ObsCli.PutObject(input)
  1162. if err != nil {
  1163. log.Error("PutObject(%s) failed: %s", input.Key, err.Error())
  1164. return err
  1165. }
  1166. if err = uploadCodeToObs(codePath+file.Name()+"/", jobName, parentDir+file.Name()+"/"); err != nil {
  1167. log.Error("uploadCodeToObs(%s) failed: %s", file.Name(), err.Error())
  1168. return err
  1169. }
  1170. } else {
  1171. input := &obs.PutFileInput{}
  1172. input.Bucket = setting.Bucket
  1173. input.Key = setting.CodePathPrefix + jobName + "/code/" + parentDir + file.Name()
  1174. input.SourceFile = codePath + file.Name()
  1175. _, err = storage.ObsCli.PutFile(input)
  1176. if err != nil {
  1177. log.Error("PutFile(%s) failed: %s", input.SourceFile, err.Error())
  1178. return err
  1179. }
  1180. }
  1181. }
  1182. return nil
  1183. }
  1184. func obsMkdir(dir string) error {
  1185. input := &obs.PutObjectInput{}
  1186. input.Bucket = setting.Bucket
  1187. input.Key = dir
  1188. _, err := storage.ObsCli.PutObject(input)
  1189. if err != nil {
  1190. log.Error("PutObject(%s) failed: %s", input.Key, err.Error())
  1191. return err
  1192. }
  1193. return nil
  1194. }
  1195. func paramCheckCreateTrainJob(form auth.CreateModelArtsTrainJobForm) error {
  1196. if !strings.HasSuffix(form.BootFile, ".py") {
  1197. log.Error("the boot file(%s) must be a python file", form.BootFile)
  1198. return errors.New("启动文件必须是python文件")
  1199. }
  1200. if form.WorkServerNumber > 25 || form.WorkServerNumber < 1 {
  1201. log.Error("the WorkServerNumber(%d) must be in (1,25)", form.WorkServerNumber)
  1202. return errors.New("计算节点数必须在1-25之间")
  1203. }
  1204. if form.BranchName == "" {
  1205. log.Error("the branch must not be null!", form.BranchName)
  1206. return errors.New("代码分支不能为空!")
  1207. }
  1208. return nil
  1209. }
  1210. func paramCheckCreateInferenceJob(form auth.CreateModelArtsInferenceJobForm) error {
  1211. if !strings.HasSuffix(form.BootFile, ".py") {
  1212. log.Error("the boot file(%s) must be a python file", form.BootFile)
  1213. return errors.New("启动文件必须是python文件")
  1214. }
  1215. if form.WorkServerNumber > 25 || form.WorkServerNumber < 1 {
  1216. log.Error("the WorkServerNumber(%d) must be in (1,25)", form.WorkServerNumber)
  1217. return errors.New("计算节点数必须在1-25之间")
  1218. }
  1219. if form.ModelName == "" {
  1220. log.Error("the ModelName(%d) must not be nil", form.ModelName)
  1221. return errors.New("模型名称不能为空")
  1222. }
  1223. if form.ModelVersion == "" {
  1224. log.Error("the ModelVersion(%d) must not be nil", form.ModelVersion)
  1225. return errors.New("模型版本不能为空")
  1226. }
  1227. if form.CkptName == "" {
  1228. log.Error("the CkptName(%d) must not be nil", form.CkptName)
  1229. return errors.New("权重文件不能为空")
  1230. }
  1231. if form.BranchName == "" {
  1232. log.Error("the Branch(%d) must not be nil", form.BranchName)
  1233. return errors.New("分支名不能为空")
  1234. }
  1235. if utf8.RuneCountInString(form.Description) > 255 {
  1236. log.Error("the Description length(%d) must not more than 255", form.Description)
  1237. return errors.New("描述字符不能超过255个字符")
  1238. }
  1239. return nil
  1240. }
  1241. func TrainJobShow(ctx *context.Context) {
  1242. ctx.Data["PageIsCloudBrain"] = true
  1243. var jobID = ctx.Params(":jobid")
  1244. repo := ctx.Repo.Repository
  1245. page := ctx.QueryInt("page")
  1246. if page <= 0 {
  1247. page = 1
  1248. }
  1249. var jobTypes []string
  1250. jobTypes = append(jobTypes, string(models.JobTypeTrain))
  1251. VersionListTasks, VersionListCount, err := models.CloudbrainsVersionList(&models.CloudbrainsOptions{
  1252. ListOptions: models.ListOptions{
  1253. Page: page,
  1254. PageSize: setting.UI.IssuePagingNum,
  1255. },
  1256. RepoID: repo.ID,
  1257. Type: models.TypeCloudBrainTwo,
  1258. JobTypes: jobTypes,
  1259. JobID: jobID,
  1260. })
  1261. if err != nil {
  1262. log.Error("GetVersionListTasks(%s) failed:%v", jobID, err.Error())
  1263. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1264. return
  1265. }
  1266. //设置权限
  1267. canNewJob, err := canUserCreateTrainJobVersion(ctx, VersionListTasks[0].UserID)
  1268. if err != nil {
  1269. ctx.ServerError("canNewJob failed", err)
  1270. return
  1271. }
  1272. ctx.Data["canNewJob"] = canNewJob
  1273. //将运行参数转化为epoch_size = 3, device_target = Ascend的格式
  1274. for i, task := range VersionListTasks {
  1275. var parameters models.Parameters
  1276. err := json.Unmarshal([]byte(VersionListTasks[i].Parameters), &parameters)
  1277. if err != nil {
  1278. log.Error("Failed to Unmarshal Parameters: %s (%v)", VersionListTasks[i].Parameters, err)
  1279. trainJobNewDataPrepare(ctx)
  1280. return
  1281. }
  1282. if len(parameters.Parameter) > 0 {
  1283. paramTemp := ""
  1284. for _, Parameter := range parameters.Parameter {
  1285. param := Parameter.Label + " = " + Parameter.Value + "; "
  1286. paramTemp = paramTemp + param
  1287. }
  1288. VersionListTasks[i].Parameters = paramTemp[:len(paramTemp)-2]
  1289. } else {
  1290. VersionListTasks[i].Parameters = ""
  1291. }
  1292. VersionListTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  1293. VersionListTasks[i].CanModify = cloudbrain.CanModifyJob(ctx, &task.Cloudbrain)
  1294. }
  1295. pager := context.NewPagination(VersionListCount, setting.UI.IssuePagingNum, page, 5)
  1296. pager.SetDefaultParams(ctx)
  1297. ctx.Data["Page"] = pager
  1298. ctx.Data["jobID"] = jobID
  1299. ctx.Data["displayJobName"] = VersionListTasks[0].DisplayJobName
  1300. ctx.Data["version_list_task"] = VersionListTasks
  1301. ctx.Data["version_list_count"] = VersionListCount
  1302. ctx.Data["canDownload"] = cloudbrain.CanDeleteJob(ctx, &VersionListTasks[0].Cloudbrain)
  1303. ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1304. }
  1305. func TrainJobGetLog(ctx *context.Context) {
  1306. ctx.Data["PageIsTrainJob"] = true
  1307. var jobID = ctx.Params(":jobid")
  1308. var logFileName = ctx.Query("file_name")
  1309. var baseLine = ctx.Query("base_line")
  1310. var order = ctx.Query("order")
  1311. if order != modelarts.OrderDesc && order != modelarts.OrderAsc {
  1312. log.Error("order(%s) check failed", order)
  1313. ctx.HTML(http.StatusBadRequest, tplModelArtsTrainJobShow)
  1314. return
  1315. }
  1316. task, err := models.GetCloudbrainByJobID(jobID)
  1317. if err != nil {
  1318. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  1319. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1320. return
  1321. }
  1322. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), baseLine, logFileName, order, modelarts.Lines)
  1323. if err != nil {
  1324. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  1325. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1326. return
  1327. }
  1328. ctx.Data["log"] = result
  1329. //ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1330. }
  1331. func trainJobGetLog(jobID string) (*models.GetTrainJobLogFileNamesResult, *models.GetTrainJobLogResult, error) {
  1332. task, err := models.GetCloudbrainByJobID(jobID)
  1333. if err != nil {
  1334. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  1335. return nil, nil, err
  1336. }
  1337. resultLogFile, err := modelarts.GetTrainJobLogFileNames(jobID, strconv.FormatInt(task.VersionID, 10))
  1338. if err != nil {
  1339. log.Error("GetTrainJobLogFileNames(%s) failed:%v", jobID, err.Error())
  1340. return nil, nil, err
  1341. }
  1342. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), "", resultLogFile.LogFileList[0], modelarts.OrderDesc, modelarts.Lines)
  1343. if err != nil {
  1344. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  1345. return nil, nil, err
  1346. }
  1347. return resultLogFile, result, err
  1348. }
  1349. func TrainJobDel(ctx *context.Context) {
  1350. var jobID = ctx.Params(":jobid")
  1351. repo := ctx.Repo.Repository
  1352. var jobTypes []string
  1353. jobTypes = append(jobTypes, string(models.JobTypeTrain))
  1354. VersionListTasks, _, err := models.CloudbrainsVersionList(&models.CloudbrainsOptions{
  1355. RepoID: repo.ID,
  1356. Type: models.TypeCloudBrainTwo,
  1357. JobTypes: jobTypes,
  1358. JobID: jobID,
  1359. })
  1360. if err != nil {
  1361. ctx.ServerError("get VersionListTasks failed", err)
  1362. return
  1363. }
  1364. //删除modelarts上的任务记录
  1365. _, err = modelarts.DelTrainJob(jobID)
  1366. if err != nil {
  1367. log.Error("DelTrainJob(%s) failed:%v", jobID, err.Error())
  1368. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil)
  1369. return
  1370. }
  1371. //删除数据库Cloudbrain表的记录
  1372. for _, task := range VersionListTasks {
  1373. err = models.DeleteJob(&task.Cloudbrain)
  1374. if err != nil {
  1375. ctx.ServerError("DeleteJob failed", err)
  1376. return
  1377. }
  1378. }
  1379. //删除存储
  1380. if len(VersionListTasks) > 0 {
  1381. DeleteJobStorage(VersionListTasks[0].JobName)
  1382. }
  1383. var isAdminPage = ctx.Query("isadminpage")
  1384. if ctx.IsUserSiteAdmin() && isAdminPage == "true" {
  1385. ctx.Redirect(setting.AppSubURL + "/admin" + "/cloudbrains")
  1386. } else {
  1387. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  1388. }
  1389. }
  1390. func TrainJobStop(ctx *context.Context) {
  1391. var jobID = ctx.Params(":jobid")
  1392. task := ctx.Cloudbrain
  1393. _, err := modelarts.StopTrainJob(jobID, strconv.FormatInt(task.VersionID, 10))
  1394. if err != nil {
  1395. log.Error("StopTrainJob(%s) failed:%v", task.JobName, err.Error())
  1396. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil)
  1397. return
  1398. }
  1399. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  1400. }
  1401. func canUserCreateTrainJob(uid int64) (bool, error) {
  1402. org, err := models.GetOrgByName(setting.AllowedOrg)
  1403. if err != nil {
  1404. log.Error("get allowed org failed: ", setting.AllowedOrg)
  1405. return false, err
  1406. }
  1407. return org.IsOrgMember(uid)
  1408. }
  1409. func canUserCreateTrainJobVersion(ctx *context.Context, userID int64) (bool, error) {
  1410. if ctx == nil || ctx.User == nil {
  1411. log.Error("user unlogin!")
  1412. return false, nil
  1413. }
  1414. if userID == ctx.User.ID || ctx.User.IsAdmin {
  1415. return true, nil
  1416. } else {
  1417. log.Error("Only user itself and admin can new trainjob!")
  1418. return false, nil
  1419. }
  1420. }
  1421. func TrainJobGetConfigList(ctx *context.Context) {
  1422. ctx.Data["PageIsTrainJob"] = true
  1423. var jobID = ctx.Params(":jobid")
  1424. var logFileName = ctx.Query("file_name")
  1425. var baseLine = ctx.Query("base_line")
  1426. var order = ctx.Query("order")
  1427. if order != modelarts.OrderDesc && order != modelarts.OrderAsc {
  1428. log.Error("order(%s) check failed", order)
  1429. ctx.HTML(http.StatusBadRequest, tplModelArtsTrainJobShow)
  1430. return
  1431. }
  1432. task, err := models.GetCloudbrainByJobID(jobID)
  1433. if err != nil {
  1434. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  1435. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1436. return
  1437. }
  1438. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), baseLine, logFileName, order, modelarts.Lines)
  1439. if err != nil {
  1440. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  1441. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1442. return
  1443. }
  1444. ctx.Data["log"] = result
  1445. //ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1446. }
  1447. func getConfigList(perPage, page int, sortBy, order, searchContent, configType string) (*models.GetConfigListResult, error) {
  1448. var result models.GetConfigListResult
  1449. list, err := modelarts.GetConfigList(perPage, page, sortBy, order, searchContent, configType)
  1450. if err != nil {
  1451. log.Error("GetConfigList failed:", err)
  1452. return &result, err
  1453. }
  1454. for _, config := range list.ParaConfigs {
  1455. paraConfig, err := modelarts.GetParaConfig(config.ConfigName, configType)
  1456. if err != nil {
  1457. log.Error("GetParaConfig failed:", err)
  1458. return &result, err
  1459. }
  1460. config.Result = paraConfig
  1461. }
  1462. return list, nil
  1463. }
  1464. func InferenceJobCreate(ctx *context.Context, form auth.CreateModelArtsInferenceJobForm) {
  1465. ctx.Data["PageIsTrainJob"] = true
  1466. VersionOutputPath := modelarts.GetOutputPathByCount(modelarts.TotalVersionCount)
  1467. displayJobName := form.DisplayJobName
  1468. jobName := util.ConvertDisplayJobNameToJobName(displayJobName)
  1469. uuid := form.Attachment
  1470. description := form.Description
  1471. workServerNumber := form.WorkServerNumber
  1472. engineID := form.EngineID
  1473. bootFile := form.BootFile
  1474. flavorCode := form.Flavor
  1475. params := form.Params
  1476. poolID := form.PoolID
  1477. repo := ctx.Repo.Repository
  1478. codeLocalPath := setting.JobPath + jobName + modelarts.CodePath
  1479. codeObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.CodePath
  1480. resultObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.ResultPath + VersionOutputPath + "/"
  1481. logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath + VersionOutputPath + "/"
  1482. dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/"
  1483. branch_name := form.BranchName
  1484. FlavorName := form.FlavorName
  1485. EngineName := form.EngineName
  1486. LabelName := form.LabelName
  1487. isLatestVersion := modelarts.IsLatestVersion
  1488. VersionCount := modelarts.VersionCount
  1489. trainUrl := form.TrainUrl
  1490. modelName := form.ModelName
  1491. modelVersion := form.ModelVersion
  1492. ckptName := form.CkptName
  1493. ckptUrl := form.TrainUrl + form.CkptName
  1494. count, err := models.GetCloudbrainInferenceJobCountByUserID(ctx.User.ID)
  1495. if err != nil {
  1496. log.Error("GetCloudbrainInferenceJobCountByUserID failed:%v", err, ctx.Data["MsgID"])
  1497. inferenceJobErrorNewDataPrepare(ctx, form)
  1498. ctx.RenderWithErr("system error", tplModelArtsInferenceJobNew, &form)
  1499. return
  1500. } else {
  1501. if count >= 1 {
  1502. log.Error("the user already has running or waiting inference task", ctx.Data["MsgID"])
  1503. inferenceJobErrorNewDataPrepare(ctx, form)
  1504. ctx.RenderWithErr("you have already a running or waiting inference task, can not create more", tplModelArtsInferenceJobNew, &form)
  1505. return
  1506. }
  1507. }
  1508. if err := paramCheckCreateInferenceJob(form); err != nil {
  1509. log.Error("paramCheckCreateInferenceJob failed:(%v)", err)
  1510. inferenceJobErrorNewDataPrepare(ctx, form)
  1511. ctx.RenderWithErr(err.Error(), tplModelArtsInferenceJobNew, &form)
  1512. return
  1513. }
  1514. //Determine whether the task name of the task in the project is duplicated
  1515. tasks, err := models.GetCloudbrainsByDisplayJobName(repo.ID, string(models.JobTypeInference), displayJobName)
  1516. if err == nil {
  1517. if len(tasks) != 0 {
  1518. log.Error("the job name did already exist", ctx.Data["MsgID"])
  1519. inferenceJobErrorNewDataPrepare(ctx, form)
  1520. ctx.RenderWithErr("the job name did already exist", tplModelArtsInferenceJobNew, &form)
  1521. return
  1522. }
  1523. } else {
  1524. if !models.IsErrJobNotExist(err) {
  1525. log.Error("system error, %v", err, ctx.Data["MsgID"])
  1526. inferenceJobErrorNewDataPrepare(ctx, form)
  1527. ctx.RenderWithErr("system error", tplModelArtsInferenceJobNew, &form)
  1528. return
  1529. }
  1530. }
  1531. //todo: del the codeLocalPath
  1532. _, err = ioutil.ReadDir(codeLocalPath)
  1533. if err == nil {
  1534. os.RemoveAll(codeLocalPath)
  1535. }
  1536. gitRepo, _ := git.OpenRepository(repo.RepoPath())
  1537. commitID, _ := gitRepo.GetBranchCommitID(branch_name)
  1538. if err := git.Clone(repo.RepoPath(), codeLocalPath, git.CloneRepoOptions{
  1539. Branch: branch_name,
  1540. }); err != nil {
  1541. log.Error("Create task failed, server timed out: %s (%v)", repo.FullName(), err)
  1542. inferenceJobErrorNewDataPrepare(ctx, form)
  1543. ctx.RenderWithErr("Create task failed, server timed out", tplModelArtsInferenceJobNew, &form)
  1544. return
  1545. }
  1546. //todo: upload code (send to file_server todo this work?)
  1547. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.ResultPath + VersionOutputPath + "/"); err != nil {
  1548. log.Error("Failed to obsMkdir_result: %s (%v)", repo.FullName(), err)
  1549. inferenceJobErrorNewDataPrepare(ctx, form)
  1550. ctx.RenderWithErr("Failed to obsMkdir_result", tplModelArtsInferenceJobNew, &form)
  1551. return
  1552. }
  1553. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.LogPath + VersionOutputPath + "/"); err != nil {
  1554. log.Error("Failed to obsMkdir_log: %s (%v)", repo.FullName(), err)
  1555. inferenceJobErrorNewDataPrepare(ctx, form)
  1556. ctx.RenderWithErr("Failed to obsMkdir_log", tplModelArtsInferenceJobNew, &form)
  1557. return
  1558. }
  1559. if err := uploadCodeToObs(codeLocalPath, jobName, ""); err != nil {
  1560. log.Error("Failed to uploadCodeToObs: %s (%v)", repo.FullName(), err)
  1561. inferenceJobErrorNewDataPrepare(ctx, form)
  1562. ctx.RenderWithErr("Failed to uploadCodeToObs", tplModelArtsInferenceJobNew, &form)
  1563. return
  1564. }
  1565. //todo: del local code?
  1566. var parameters models.Parameters
  1567. param := make([]models.Parameter, 0)
  1568. param = append(param, models.Parameter{
  1569. Label: modelarts.ResultUrl,
  1570. Value: "s3:/" + resultObsPath,
  1571. }, models.Parameter{
  1572. Label: modelarts.CkptUrl,
  1573. Value: "s3:/" + ckptUrl,
  1574. })
  1575. if len(params) != 0 {
  1576. err := json.Unmarshal([]byte(params), &parameters)
  1577. if err != nil {
  1578. log.Error("Failed to Unmarshal params: %s (%v)", params, err)
  1579. inferenceJobErrorNewDataPrepare(ctx, form)
  1580. ctx.RenderWithErr("运行参数错误", tplModelArtsInferenceJobNew, &form)
  1581. return
  1582. }
  1583. for _, parameter := range parameters.Parameter {
  1584. if parameter.Label != modelarts.TrainUrl && parameter.Label != modelarts.DataUrl {
  1585. param = append(param, models.Parameter{
  1586. Label: parameter.Label,
  1587. Value: parameter.Value,
  1588. })
  1589. }
  1590. }
  1591. }
  1592. req := &modelarts.GenerateInferenceJobReq{
  1593. JobName: jobName,
  1594. DisplayJobName: displayJobName,
  1595. DataUrl: dataPath,
  1596. Description: description,
  1597. CodeObsPath: codeObsPath,
  1598. BootFileUrl: codeObsPath + bootFile,
  1599. BootFile: bootFile,
  1600. TrainUrl: trainUrl,
  1601. FlavorCode: flavorCode,
  1602. WorkServerNumber: workServerNumber,
  1603. EngineID: int64(engineID),
  1604. LogUrl: logObsPath,
  1605. PoolID: poolID,
  1606. Uuid: uuid,
  1607. Parameters: param, //modelarts train parameters
  1608. CommitID: commitID,
  1609. BranchName: branch_name,
  1610. Params: form.Params,
  1611. FlavorName: FlavorName,
  1612. EngineName: EngineName,
  1613. LabelName: LabelName,
  1614. IsLatestVersion: isLatestVersion,
  1615. VersionCount: VersionCount,
  1616. TotalVersionCount: modelarts.TotalVersionCount,
  1617. ModelName: modelName,
  1618. ModelVersion: modelVersion,
  1619. CkptName: ckptName,
  1620. ResultUrl: resultObsPath,
  1621. }
  1622. err = modelarts.GenerateInferenceJob(ctx, req)
  1623. if err != nil {
  1624. log.Error("GenerateTrainJob failed:%v", err.Error())
  1625. inferenceJobErrorNewDataPrepare(ctx, form)
  1626. ctx.RenderWithErr(err.Error(), tplModelArtsInferenceJobNew, &form)
  1627. return
  1628. }
  1629. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/inference-job")
  1630. }
  1631. func InferenceJobIndex(ctx *context.Context) {
  1632. MustEnableModelArts(ctx)
  1633. repo := ctx.Repo.Repository
  1634. page := ctx.QueryInt("page")
  1635. if page <= 0 {
  1636. page = 1
  1637. }
  1638. var jobTypes []string
  1639. jobTypes = append(jobTypes, string(models.JobTypeInference))
  1640. tasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  1641. ListOptions: models.ListOptions{
  1642. Page: page,
  1643. PageSize: setting.UI.IssuePagingNum,
  1644. },
  1645. RepoID: repo.ID,
  1646. Type: models.TypeCloudBrainTwo,
  1647. JobTypes: jobTypes,
  1648. })
  1649. if err != nil {
  1650. ctx.ServerError("Cloudbrain", err)
  1651. return
  1652. }
  1653. for i, task := range tasks {
  1654. tasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  1655. tasks[i].CanModify = cloudbrain.CanModifyJob(ctx, &task.Cloudbrain)
  1656. tasks[i].ComputeResource = models.NPUResource
  1657. }
  1658. repoId := ctx.Repo.Repository.ID
  1659. Type := -1
  1660. _, model_count, _ := models.QueryModel(&models.AiModelQueryOptions{
  1661. ListOptions: models.ListOptions{
  1662. Page: 1,
  1663. PageSize: 2,
  1664. },
  1665. RepoID: repoId,
  1666. Type: Type,
  1667. New: MODEL_LATEST,
  1668. })
  1669. ctx.Data["MODEL_COUNT"] = model_count
  1670. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  1671. pager.SetDefaultParams(ctx)
  1672. ctx.Data["Page"] = pager
  1673. ctx.Data["PageIsCloudBrain"] = true
  1674. ctx.Data["Tasks"] = tasks
  1675. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  1676. ctx.Data["RepoIsEmpty"] = repo.IsEmpty
  1677. ctx.HTML(200, tplModelArtsInferenceJobIndex)
  1678. }
  1679. func InferenceJobNew(ctx *context.Context) {
  1680. err := inferenceJobNewDataPrepare(ctx)
  1681. if err != nil {
  1682. ctx.ServerError("get new inference-job info failed", err)
  1683. return
  1684. }
  1685. ctx.HTML(200, tplModelArtsInferenceJobNew)
  1686. }
  1687. func inferenceJobNewDataPrepare(ctx *context.Context) error {
  1688. ctx.Data["PageIsCloudBrain"] = true
  1689. t := time.Now()
  1690. var displayJobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  1691. ctx.Data["display_job_name"] = displayJobName
  1692. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  1693. if err != nil {
  1694. ctx.ServerError("GetAllUserAttachments failed:", err)
  1695. return err
  1696. }
  1697. ctx.Data["attachments"] = attachs
  1698. var resourcePools modelarts.ResourcePool
  1699. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  1700. ctx.ServerError("json.Unmarshal failed:", err)
  1701. return err
  1702. }
  1703. ctx.Data["resource_pools"] = resourcePools.Info
  1704. var engines modelarts.Engine
  1705. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  1706. ctx.ServerError("json.Unmarshal failed:", err)
  1707. return err
  1708. }
  1709. ctx.Data["engines"] = engines.Info
  1710. var versionInfos modelarts.VersionInfo
  1711. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  1712. ctx.ServerError("json.Unmarshal failed:", err)
  1713. return err
  1714. }
  1715. ctx.Data["engine_versions"] = versionInfos.Version
  1716. var flavorInfos modelarts.Flavor
  1717. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  1718. ctx.ServerError("json.Unmarshal failed:", err)
  1719. return err
  1720. }
  1721. ctx.Data["flavor_infos"] = flavorInfos.Info
  1722. ctx.Data["params"] = ""
  1723. ctx.Data["branchName"] = ctx.Repo.BranchName
  1724. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  1725. if err != nil {
  1726. ctx.ServerError("getConfigList failed:", err)
  1727. return err
  1728. }
  1729. ctx.Data["config_list"] = configList.ParaConfigs
  1730. repoId := ctx.Repo.Repository.ID
  1731. Type := -1
  1732. _, model_count, _ := models.QueryModel(&models.AiModelQueryOptions{
  1733. ListOptions: models.ListOptions{
  1734. Page: 1,
  1735. PageSize: 2,
  1736. },
  1737. RepoID: repoId,
  1738. Type: Type,
  1739. New: MODEL_LATEST,
  1740. })
  1741. ctx.Data["MODEL_COUNT"] = model_count
  1742. return nil
  1743. }
  1744. func inferenceJobErrorNewDataPrepare(ctx *context.Context, form auth.CreateModelArtsInferenceJobForm) error {
  1745. ctx.Data["PageIsCloudBrain"] = true
  1746. t := time.Now()
  1747. var jobName = "inference" + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  1748. ctx.Data["job_name"] = jobName
  1749. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  1750. if err != nil {
  1751. ctx.ServerError("GetAllUserAttachments failed:", err)
  1752. return err
  1753. }
  1754. ctx.Data["attachments"] = attachs
  1755. var resourcePools modelarts.ResourcePool
  1756. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  1757. ctx.ServerError("json.Unmarshal failed:", err)
  1758. return err
  1759. }
  1760. ctx.Data["resource_pools"] = resourcePools.Info
  1761. var engines modelarts.Engine
  1762. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  1763. ctx.ServerError("json.Unmarshal failed:", err)
  1764. return err
  1765. }
  1766. ctx.Data["engines"] = engines.Info
  1767. var versionInfos modelarts.VersionInfo
  1768. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  1769. ctx.ServerError("json.Unmarshal failed:", err)
  1770. return err
  1771. }
  1772. ctx.Data["engine_versions"] = versionInfos.Version
  1773. var flavorInfos modelarts.Flavor
  1774. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  1775. ctx.ServerError("json.Unmarshal failed:", err)
  1776. return err
  1777. }
  1778. ctx.Data["flavor_infos"] = flavorInfos.Info
  1779. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  1780. if err != nil {
  1781. ctx.ServerError("getConfigList failed:", err)
  1782. return err
  1783. }
  1784. var Parameters modelarts.Parameters
  1785. if err = json.Unmarshal([]byte(form.Params), &Parameters); err != nil {
  1786. ctx.ServerError("json.Unmarshal failed:", err)
  1787. return err
  1788. }
  1789. ctx.Data["params"] = Parameters.Parameter
  1790. ctx.Data["config_list"] = configList.ParaConfigs
  1791. ctx.Data["bootFile"] = form.BootFile
  1792. ctx.Data["uuid"] = form.Attachment
  1793. ctx.Data["branch_name"] = form.BranchName
  1794. ctx.Data["model_name"] = form.ModelName
  1795. ctx.Data["model_version"] = form.ModelVersion
  1796. ctx.Data["ckpt_name"] = form.CkptName
  1797. ctx.Data["train_url"] = form.TrainUrl
  1798. return nil
  1799. }
  1800. func InferenceJobShow(ctx *context.Context) {
  1801. ctx.Data["PageIsCloudBrain"] = true
  1802. var jobID = ctx.Params(":jobid")
  1803. page := ctx.QueryInt("page")
  1804. if page <= 0 {
  1805. page = 1
  1806. }
  1807. task, err := models.GetCloudbrainByJobID(jobID)
  1808. if err != nil {
  1809. log.Error("GetInferenceTask(%s) failed:%v", jobID, err.Error())
  1810. ctx.RenderWithErr(err.Error(), tplModelArtsInferenceJobShow, nil)
  1811. return
  1812. }
  1813. //设置权限
  1814. canNewJob, err := canUserCreateTrainJobVersion(ctx, task.UserID)
  1815. if err != nil {
  1816. ctx.ServerError("canNewJob failed", err)
  1817. return
  1818. }
  1819. ctx.Data["canNewJob"] = canNewJob
  1820. //将运行参数转化为epoch_size = 3, device_target = Ascend的格式
  1821. var parameters models.Parameters
  1822. err = json.Unmarshal([]byte(task.Parameters), &parameters)
  1823. if err != nil {
  1824. log.Error("Failed to Unmarshal Parameters: %s (%v)", task.Parameters, err)
  1825. trainJobNewDataPrepare(ctx)
  1826. return
  1827. }
  1828. if len(parameters.Parameter) > 0 {
  1829. paramTemp := ""
  1830. for _, Parameter := range parameters.Parameter {
  1831. param := Parameter.Label + " = " + Parameter.Value + "; "
  1832. paramTemp = paramTemp + param
  1833. }
  1834. task.Parameters = paramTemp[:len(paramTemp)-2]
  1835. } else {
  1836. task.Parameters = ""
  1837. }
  1838. LabelName := strings.Fields(task.LabelName)
  1839. ctx.Data["labelName"] = LabelName
  1840. ctx.Data["jobID"] = jobID
  1841. ctx.Data["jobName"] = task.JobName
  1842. ctx.Data["displayJobName"] = task.DisplayJobName
  1843. ctx.Data["task"] = task
  1844. ctx.Data["canDownload"] = cloudbrain.CanDeleteJob(ctx, task)
  1845. tempUids := []int64{}
  1846. tempUids = append(tempUids, task.UserID)
  1847. JobCreater, err := models.GetUserNamesByIDs(tempUids)
  1848. if err != nil {
  1849. log.Error("GetUserNamesByIDs (WhitelistUserIDs): %v", err)
  1850. }
  1851. ctx.Data["userName"] = JobCreater[0]
  1852. ctx.HTML(http.StatusOK, tplModelArtsInferenceJobShow)
  1853. }
  1854. func ModelDownload(ctx *context.Context) {
  1855. var (
  1856. err error
  1857. )
  1858. var jobID = ctx.Params(":jobid")
  1859. versionName := ctx.Query("version_name")
  1860. parentDir := ctx.Query("parent_dir")
  1861. fileName := ctx.Query("file_name")
  1862. log.Info("DownloadSingleModelFile start.")
  1863. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  1864. if err != nil {
  1865. log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
  1866. return
  1867. }
  1868. path := strings.TrimPrefix(path.Join(setting.TrainJobModelPath, task.JobName, setting.OutPutPath, versionName, parentDir, fileName), "/")
  1869. log.Info("Download path is:%s", path)
  1870. url, err := storage.GetObsCreateSignedUrlByBucketAndKey(setting.Bucket, path)
  1871. if err != nil {
  1872. log.Error("GetObsCreateSignedUrl failed: %v", err.Error(), ctx.Data["msgID"])
  1873. ctx.ServerError("GetObsCreateSignedUrl", err)
  1874. return
  1875. }
  1876. http.Redirect(ctx.Resp, ctx.Req.Request, url, http.StatusMovedPermanently)
  1877. }
  1878. func ResultDownload(ctx *context.Context) {
  1879. var (
  1880. err error
  1881. )
  1882. versionName := ctx.Query("version_name")
  1883. parentDir := ctx.Query("parent_dir")
  1884. fileName := ctx.Query("file_name")
  1885. log.Info("DownloadResult start.")
  1886. task := ctx.Cloudbrain
  1887. if err != nil {
  1888. ctx.Data["error"] = err.Error()
  1889. }
  1890. path := strings.TrimPrefix(path.Join(setting.TrainJobModelPath, task.JobName, "result/", versionName, parentDir, fileName), "/")
  1891. log.Info("Download path is:%s", path)
  1892. url, err := storage.GetObsCreateSignedUrlByBucketAndKey(setting.Bucket, path)
  1893. if err != nil {
  1894. log.Error("GetObsCreateSignedUrl failed: %v", err.Error(), ctx.Data["msgID"])
  1895. ctx.ServerError("GetObsCreateSignedUrl", err)
  1896. return
  1897. }
  1898. http.Redirect(ctx.Resp, ctx.Req.Request, url, http.StatusMovedPermanently)
  1899. }
  1900. func DeleteJobStorage(jobName string) error {
  1901. //delete local
  1902. localJobPath := setting.JobPath + jobName
  1903. err := os.RemoveAll(localJobPath)
  1904. if err != nil {
  1905. log.Error("RemoveAll(%s) failed:%v", localJobPath, err)
  1906. }
  1907. //delete oss
  1908. dirPath := setting.CodePathPrefix + jobName + "/"
  1909. err = storage.ObsRemoveObject(setting.Bucket, dirPath)
  1910. if err != nil {
  1911. log.Error("ObsRemoveObject(%s) failed:%v", localJobPath, err)
  1912. }
  1913. return nil
  1914. }
  1915. func DownloadMultiResultFile(ctx *context.Context) {
  1916. var jobID = ctx.Params(":jobid")
  1917. var versionName = ctx.Query("version_name")
  1918. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  1919. if err != nil {
  1920. log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
  1921. return
  1922. }
  1923. // if !isCanDeleteOrDownload(ctx, task) {
  1924. // ctx.ServerError("no right.", errors.New(ctx.Tr("repo.model_noright")))
  1925. // return
  1926. // }
  1927. // path := Model_prefix + models.AttachmentRelativePath(id) + "/"
  1928. path := strings.TrimPrefix(path.Join(setting.TrainJobModelPath, task.JobName, "result/", versionName), "/") + "/"
  1929. allFile, err := storage.GetAllObjectByBucketAndPrefix(setting.Bucket, path)
  1930. if err == nil {
  1931. //count++
  1932. // models.ModifyModelDownloadCount(id)
  1933. returnFileName := task.DisplayJobName + ".zip"
  1934. ctx.Resp.Header().Set("Content-Disposition", "attachment; filename="+returnFileName)
  1935. ctx.Resp.Header().Set("Content-Type", "application/octet-stream")
  1936. w := zip.NewWriter(ctx.Resp)
  1937. defer w.Close()
  1938. for _, oneFile := range allFile {
  1939. if oneFile.IsDir {
  1940. log.Info("zip dir name:" + oneFile.FileName)
  1941. } else {
  1942. log.Info("zip file name:" + oneFile.FileName)
  1943. fDest, err := w.Create(oneFile.FileName)
  1944. if err != nil {
  1945. log.Info("create zip entry error, download file failed: %s\n", err.Error())
  1946. ctx.ServerError("download file failed:", err)
  1947. return
  1948. }
  1949. body, err := storage.ObsDownloadAFile(setting.Bucket, path+oneFile.FileName)
  1950. if err != nil {
  1951. log.Info("download file failed: %s\n", err.Error())
  1952. ctx.ServerError("download file failed:", err)
  1953. return
  1954. } else {
  1955. defer body.Close()
  1956. p := make([]byte, 1024)
  1957. var readErr error
  1958. var readCount int
  1959. // 读取对象内容
  1960. for {
  1961. readCount, readErr = body.Read(p)
  1962. if readCount > 0 {
  1963. fDest.Write(p[:readCount])
  1964. }
  1965. if readErr != nil {
  1966. break
  1967. }
  1968. }
  1969. }
  1970. }
  1971. }
  1972. } else {
  1973. log.Info("error,msg=" + err.Error())
  1974. ctx.ServerError("no file to download.", err)
  1975. }
  1976. }
  1977. func SetJobCount(ctx *context.Context) {
  1978. repoId := ctx.Repo.Repository.ID
  1979. _, jobCount, err := models.Cloudbrains(&models.CloudbrainsOptions{
  1980. RepoID: repoId,
  1981. Type: modelarts.DebugType,
  1982. })
  1983. if err != nil {
  1984. ctx.ServerError("Get job faild:", err)
  1985. return
  1986. }
  1987. ctx.Data["jobCount"] = jobCount
  1988. }