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.

cloudbrain.go 43 kB

4 years ago
5 years ago
4 years ago
5 years ago
5 years ago
3 years ago
5 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
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
4 years ago
4 years ago
3 years ago
3 years ago
3 years ago
4 years ago
3 years ago
3 years ago
4 years ago
4 years ago
4 years ago
5 years ago
3 years ago
4 years ago
5 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
5 years ago
5 years ago
4 years ago
5 years ago
4 years ago
4 years ago
5 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
3 years ago
4 years ago
3 years ago
5 years ago
4 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
4 years ago
4 years ago
3 years ago
3 years ago
3 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
5 years ago
5 years ago
3 years ago
3 years ago
3 years ago
5 years ago
3 years ago
5 years ago
4 years ago
3 years ago
4 years ago
4 years ago
5 years ago
4 years ago
5 years ago
3 years ago
3 years ago
3 years ago
5 years ago
3 years ago
3 years ago
5 years ago
5 years ago
5 years ago
3 years ago
5 years ago
5 years ago
3 years ago
5 years ago
3 years ago
5 years ago
5 years ago
5 years ago
4 years ago
4 years ago
5 years ago
3 years ago
4 years ago
4 years ago
5 years ago
3 years ago
4 years ago
5 years ago
4 years ago
4 years ago
4 years ago
5 years ago
4 years ago
4 years ago
4 years ago
4 years ago
5 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
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
5 years ago
5 years ago
4 years ago
4 years ago
5 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
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
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385
  1. package repo
  2. import (
  3. "bufio"
  4. "encoding/json"
  5. "errors"
  6. "fmt"
  7. "io"
  8. "net/http"
  9. "os"
  10. "regexp"
  11. "sort"
  12. "strconv"
  13. "strings"
  14. "time"
  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/setting"
  24. "code.gitea.io/gitea/modules/storage"
  25. "code.gitea.io/gitea/modules/util"
  26. )
  27. const (
  28. tplCloudBrainNew base.TplName = "repo/cloudbrain/new"
  29. tplCloudBrainShow base.TplName = "repo/cloudbrain/show"
  30. tplCloudBrainShowModels base.TplName = "repo/cloudbrain/models/index"
  31. tplCloudBrainBenchmarkIndex base.TplName = "repo/cloudbrain/benchmark/index"
  32. tplCloudBrainBenchmarkNew base.TplName = "repo/cloudbrain/benchmark/new"
  33. tplCloudBrainBenchmarkShow base.TplName = "repo/cloudbrain/benchmark/show"
  34. )
  35. var (
  36. gpuInfos *models.GpuInfos
  37. categories *models.Categories
  38. benchmarkTypes *models.BenchmarkTypes
  39. benchmarkGpuInfos *models.GpuInfos
  40. benchmarkResourceSpecs *models.ResourceSpecs
  41. )
  42. var jobNamePattern = regexp.MustCompile(`^[a-z0-9][a-z0-9-_]{1,34}[a-z0-9-]$`)
  43. // MustEnableDataset check if repository enable internal cb
  44. func MustEnableCloudbrain(ctx *context.Context) {
  45. if !ctx.Repo.CanRead(models.UnitTypeCloudBrain) {
  46. ctx.NotFound("MustEnableCloudbrain", nil)
  47. return
  48. }
  49. }
  50. func cutString(str string, lens int) string {
  51. if len(str) < lens {
  52. return str
  53. }
  54. return str[:lens]
  55. }
  56. func jobNamePrefixValid(s string) string {
  57. lowStr := strings.ToLower(s)
  58. re := regexp.MustCompile(`[^a-z0-9_\\-]+`)
  59. removeSpecial := re.ReplaceAllString(lowStr, "")
  60. re = regexp.MustCompile(`^[_\\-]+`)
  61. return re.ReplaceAllString(removeSpecial, "")
  62. }
  63. func cloudBrainNewDataPrepare(ctx *context.Context) error {
  64. ctx.Data["PageIsCloudBrain"] = true
  65. t := time.Now()
  66. var jobName = jobNamePrefixValid(cutString(ctx.User.Name, 5)) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  67. ctx.Data["job_name"] = jobName
  68. result, err := cloudbrain.GetImages()
  69. if err != nil {
  70. ctx.Data["error"] = err.Error()
  71. log.Error("cloudbrain.GetImages failed:", err.Error(), ctx.Data["MsgID"])
  72. }
  73. for i, payload := range result.Payload.ImageInfo {
  74. if strings.HasPrefix(result.Payload.ImageInfo[i].Place, "192.168") {
  75. result.Payload.ImageInfo[i].PlaceView = payload.Place[strings.Index(payload.Place, "/"):len(payload.Place)]
  76. } else {
  77. result.Payload.ImageInfo[i].PlaceView = payload.Place
  78. }
  79. }
  80. ctx.Data["images"] = result.Payload.ImageInfo
  81. resultPublic, err := cloudbrain.GetPublicImages()
  82. if err != nil {
  83. ctx.Data["error"] = err.Error()
  84. log.Error("cloudbrain.GetPublicImages failed:", err.Error(), ctx.Data["MsgID"])
  85. }
  86. for i, payload := range resultPublic.Payload.ImageInfo {
  87. if strings.HasPrefix(resultPublic.Payload.ImageInfo[i].Place, "192.168") {
  88. resultPublic.Payload.ImageInfo[i].PlaceView = payload.Place[strings.Index(payload.Place, "/"):len(payload.Place)]
  89. } else {
  90. resultPublic.Payload.ImageInfo[i].PlaceView = payload.Place
  91. }
  92. }
  93. ctx.Data["public_images"] = resultPublic.Payload.ImageInfo
  94. attachs, err := models.GetAllUserAttachments(ctx.User.ID)
  95. if err != nil {
  96. log.Error("GetAllUserAttachments failed: %v", err, ctx.Data["MsgID"])
  97. return err
  98. }
  99. ctx.Data["attachments"] = attachs
  100. ctx.Data["command"] = cloudbrain.Command
  101. ctx.Data["code_path"] = cloudbrain.CodeMountPath
  102. ctx.Data["dataset_path"] = cloudbrain.DataSetMountPath
  103. ctx.Data["model_path"] = cloudbrain.ModelMountPath
  104. ctx.Data["benchmark_path"] = cloudbrain.BenchMarkMountPath
  105. ctx.Data["is_benchmark_enabled"] = setting.IsBenchmarkEnabled
  106. if categories == nil {
  107. json.Unmarshal([]byte(setting.BenchmarkCategory), &categories)
  108. }
  109. ctx.Data["benchmark_categories"] = categories.Category
  110. if benchmarkTypes == nil {
  111. if err := json.Unmarshal([]byte(setting.BenchmarkTypes), &benchmarkTypes); err != nil {
  112. log.Error("json.Unmarshal BenchmarkTypes(%s) failed:%v", setting.BenchmarkTypes, err, ctx.Data["MsgID"])
  113. }
  114. }
  115. ctx.Data["benchmark_types"] = benchmarkTypes.BenchmarkType
  116. if gpuInfos == nil {
  117. json.Unmarshal([]byte(setting.GpuTypes), &gpuInfos)
  118. }
  119. ctx.Data["gpu_types"] = gpuInfos.GpuInfo
  120. if benchmarkGpuInfos == nil {
  121. json.Unmarshal([]byte(setting.BenchmarkGpuTypes), &benchmarkGpuInfos)
  122. }
  123. ctx.Data["benchmark_gpu_types"] = benchmarkGpuInfos.GpuInfo
  124. if benchmarkResourceSpecs == nil {
  125. json.Unmarshal([]byte(setting.BenchmarkResourceSpecs), &benchmarkResourceSpecs)
  126. }
  127. ctx.Data["benchmark_resource_specs"] = benchmarkResourceSpecs.ResourceSpec
  128. if cloudbrain.ResourceSpecs == nil {
  129. json.Unmarshal([]byte(setting.ResourceSpecs), &cloudbrain.ResourceSpecs)
  130. }
  131. ctx.Data["resource_specs"] = cloudbrain.ResourceSpecs.ResourceSpec
  132. ctx.Data["snn4imagenet_path"] = cloudbrain.Snn4imagenetMountPath
  133. ctx.Data["is_snn4imagenet_enabled"] = setting.IsSnn4imagenetEnabled
  134. ctx.Data["brainscore_path"] = cloudbrain.BrainScoreMountPath
  135. ctx.Data["is_brainscore_enabled"] = setting.IsBrainScoreEnabled
  136. return nil
  137. }
  138. func CloudBrainNew(ctx *context.Context) {
  139. err := cloudBrainNewDataPrepare(ctx)
  140. if err != nil {
  141. ctx.ServerError("get new cloudbrain info failed", err)
  142. return
  143. }
  144. ctx.HTML(200, tplCloudBrainNew)
  145. }
  146. func CloudBrainCreate(ctx *context.Context, form auth.CreateCloudBrainForm) {
  147. ctx.Data["PageIsCloudBrain"] = true
  148. jobName := form.JobName
  149. image := form.Image
  150. uuid := form.Attachment
  151. jobType := form.JobType
  152. command := cloudbrain.Command
  153. gpuQueue := form.GpuType
  154. codePath := setting.JobPath + jobName + cloudbrain.CodeMountPath
  155. resourceSpecId := form.ResourceSpecId
  156. if !jobNamePattern.MatchString(jobName) {
  157. ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_jobname_err"), tplCloudBrainNew, &form)
  158. return
  159. }
  160. if jobType != string(models.JobTypeBenchmark) && jobType != string(models.JobTypeDebug) && jobType != string(models.JobTypeSnn4imagenet) && jobType != string(models.JobTypeBrainScore) {
  161. log.Error("jobtype error:", jobType, ctx.Data["MsgID"])
  162. cloudBrainNewDataPrepare(ctx)
  163. ctx.RenderWithErr("jobtype error", tplCloudBrainNew, &form)
  164. return
  165. }
  166. count, err := models.GetCloudbrainCountByUserID(ctx.User.ID, jobType)
  167. if err != nil {
  168. log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])
  169. cloudBrainNewDataPrepare(ctx)
  170. ctx.RenderWithErr("system error", tplCloudBrainNew, &form)
  171. return
  172. } else {
  173. if count >= 1 {
  174. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  175. cloudBrainNewDataPrepare(ctx)
  176. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplCloudBrainNew, &form)
  177. return
  178. }
  179. }
  180. _, err = models.GetCloudbrainByName(jobName)
  181. if err == nil {
  182. log.Error("the job name did already exist", ctx.Data["MsgID"])
  183. cloudBrainNewDataPrepare(ctx)
  184. ctx.RenderWithErr("the job name did already exist", tplCloudBrainNew, &form)
  185. return
  186. } else {
  187. if !models.IsErrJobNotExist(err) {
  188. log.Error("system error, %v", err, ctx.Data["MsgID"])
  189. cloudBrainNewDataPrepare(ctx)
  190. ctx.RenderWithErr("system error", tplCloudBrainNew, &form)
  191. return
  192. }
  193. }
  194. repo := ctx.Repo.Repository
  195. downloadCode(repo, codePath)
  196. uploadCodeToMinio(codePath+"/", jobName, cloudbrain.CodeMountPath+"/")
  197. modelPath := setting.JobPath + jobName + cloudbrain.ModelMountPath + "/"
  198. mkModelPath(modelPath)
  199. uploadCodeToMinio(modelPath, jobName, cloudbrain.ModelMountPath+"/")
  200. benchmarkPath := setting.JobPath + jobName + cloudbrain.BenchMarkMountPath
  201. if setting.IsBenchmarkEnabled && jobType == string(models.JobTypeBenchmark) {
  202. var gpuType string
  203. for _, gpuInfo := range gpuInfos.GpuInfo {
  204. if gpuInfo.Queue == gpuQueue {
  205. gpuType = gpuInfo.Value
  206. }
  207. }
  208. downloadRateCode(repo, jobName, setting.BenchmarkOwner, setting.BenchmarkName, benchmarkPath, form.BenchmarkCategory, gpuType)
  209. uploadCodeToMinio(benchmarkPath+"/", jobName, cloudbrain.BenchMarkMountPath+"/")
  210. }
  211. snn4imagenetPath := setting.JobPath + jobName + cloudbrain.Snn4imagenetMountPath
  212. if setting.IsSnn4imagenetEnabled && jobType == string(models.JobTypeSnn4imagenet) {
  213. downloadRateCode(repo, jobName, setting.Snn4imagenetOwner, setting.Snn4imagenetName, snn4imagenetPath, "", "")
  214. uploadCodeToMinio(snn4imagenetPath+"/", jobName, cloudbrain.Snn4imagenetMountPath+"/")
  215. }
  216. brainScorePath := setting.JobPath + jobName + cloudbrain.BrainScoreMountPath
  217. if setting.IsBrainScoreEnabled && jobType == string(models.JobTypeBrainScore) {
  218. downloadRateCode(repo, jobName, setting.BrainScoreOwner, setting.BrainScoreName, brainScorePath, "", "")
  219. uploadCodeToMinio(brainScorePath+"/", jobName, cloudbrain.BrainScoreMountPath+"/")
  220. }
  221. err = cloudbrain.GenerateTask(ctx, jobName, image, command, uuid, storage.GetMinioPath(jobName, cloudbrain.CodeMountPath+"/"),
  222. storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/"),
  223. storage.GetMinioPath(jobName, cloudbrain.BenchMarkMountPath+"/"), storage.GetMinioPath(jobName, cloudbrain.Snn4imagenetMountPath+"/"),
  224. storage.GetMinioPath(jobName, cloudbrain.BrainScoreMountPath+"/"), jobType, gpuQueue, form.Description,
  225. 0, 0, resourceSpecId)
  226. if err != nil {
  227. cloudBrainNewDataPrepare(ctx)
  228. ctx.RenderWithErr(err.Error(), tplCloudBrainNew, &form)
  229. return
  230. }
  231. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=all")
  232. }
  233. func CloudBrainRestart(ctx *context.Context) {
  234. var jobID = ctx.Params(":jobid")
  235. var resultCode = "0"
  236. var errorMsg = ""
  237. var status = string(models.JobWaiting)
  238. task := ctx.Cloudbrain
  239. for {
  240. if task.Status != string(models.JobStopped) && task.Status != string(models.JobSucceeded) && task.Status != string(models.JobFailed) {
  241. log.Error("the job(%s) is not stopped", task.JobName, ctx.Data["MsgID"])
  242. resultCode = "-1"
  243. errorMsg = "the job is not stopped"
  244. break
  245. }
  246. if task.Image == "" || task.GpuQueue == "" || task.Type != models.TypeCloudBrainOne {
  247. log.Error("the job(%s) version is too old", task.JobName, ctx.Data["MsgID"])
  248. resultCode = "-1"
  249. errorMsg = "the job's version is too old and can not be restarted"
  250. break
  251. }
  252. if !ctx.IsSigned || (ctx.User.ID != task.UserID && !ctx.IsUserSiteAdmin()) {
  253. log.Error("the user has no right ro restart the job", task.JobName, ctx.Data["MsgID"])
  254. resultCode = "-1"
  255. errorMsg = "you have no right to restart the job"
  256. break
  257. }
  258. count, err := models.GetCloudbrainCountByUserID(ctx.User.ID, string(models.JobTypeDebug))
  259. if err != nil {
  260. log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])
  261. resultCode = "-1"
  262. errorMsg = "system error"
  263. break
  264. } else {
  265. if count >= 1 {
  266. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  267. resultCode = "-1"
  268. errorMsg = "you have already a running or waiting task, can not create more"
  269. break
  270. }
  271. }
  272. err = cloudbrain.RestartTask(ctx, task, &jobID)
  273. if err != nil {
  274. log.Error("RestartTask failed:%v", err.Error(), ctx.Data["MsgID"])
  275. resultCode = "-1"
  276. errorMsg = "system error"
  277. break
  278. }
  279. break
  280. }
  281. ctx.JSON(200, map[string]string{
  282. "result_code": resultCode,
  283. "error_msg": errorMsg,
  284. "status": status,
  285. "job_id": jobID,
  286. })
  287. }
  288. func CloudBrainBenchMarkShow(ctx *context.Context) {
  289. if benchmarkTypes == nil {
  290. if err := json.Unmarshal([]byte(setting.BenchmarkTypes), &benchmarkTypes); err != nil {
  291. log.Error("json.Unmarshal BenchmarkTypes(%s) failed:%v", setting.BenchmarkTypes, err, ctx.Data["MsgID"])
  292. ctx.ServerError(err.Error(), err)
  293. return
  294. }
  295. }
  296. cloudBrainShow(ctx, tplCloudBrainBenchmarkShow)
  297. }
  298. func CloudBrainShow(ctx *context.Context) {
  299. cloudBrainShow(ctx, tplCloudBrainShow)
  300. }
  301. func cloudBrainShow(ctx *context.Context, tpName base.TplName) {
  302. ctx.Data["PageIsCloudBrain"] = true
  303. var jobName = ctx.Params(":jobname")
  304. debugListType := ctx.Query("debugListType")
  305. task, err := models.GetCloudbrainByName(jobName)
  306. if err != nil {
  307. ctx.Data["error"] = err.Error()
  308. }
  309. result, err := cloudbrain.GetJob(task.JobID)
  310. if err != nil {
  311. ctx.Data["error"] = err.Error()
  312. }
  313. if result != nil {
  314. jobRes, _ := models.ConvertToJobResultPayload(result.Payload)
  315. jobRes.Resource.Memory = strings.ReplaceAll(jobRes.Resource.Memory, "Mi", "MB")
  316. spec := "GPU数:" + strconv.Itoa(jobRes.Resource.NvidiaComGpu) + ",CPU数:" + strconv.Itoa(jobRes.Resource.CPU) + ",内存(MB):" + jobRes.Resource.Memory
  317. ctx.Data["resource_spec"] = spec
  318. taskRoles := jobRes.TaskRoles
  319. if jobRes.JobStatus.State != string(models.JobFailed) {
  320. taskRes, _ := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
  321. ctx.Data["taskRes"] = taskRes
  322. task.Status = taskRes.TaskStatuses[0].State
  323. task.ContainerID = taskRes.TaskStatuses[0].ContainerID
  324. task.ContainerIp = taskRes.TaskStatuses[0].ContainerIP
  325. err = models.UpdateJob(task)
  326. if err != nil {
  327. ctx.Data["error"] = err.Error()
  328. }
  329. } else {
  330. task.Status = jobRes.JobStatus.State
  331. taskRes := models.TaskPod{TaskStatuses: []models.TaskStatuses{
  332. {
  333. State: jobRes.JobStatus.State,
  334. },
  335. }}
  336. ctx.Data["taskRes"] = taskRes
  337. jobRes.JobStatus.StartTime = time.Unix(int64(task.CreatedUnix), 0).Format("2006-01-02 15:04:05")
  338. jobRes.JobStatus.EndTime = time.Unix(int64(task.UpdatedUnix), 0).Format("2006-01-02 15:04:05")
  339. }
  340. ctx.Data["result"] = jobRes
  341. } else {
  342. log.Info("error:" + err.Error())
  343. }
  344. user, err := models.GetUserByID(task.UserID)
  345. if err == nil {
  346. task.User = user
  347. }
  348. var duration int64
  349. if task.Status == string(models.JobRunning) {
  350. duration = time.Now().Unix() - int64(task.CreatedUnix)
  351. } else {
  352. duration = int64(task.UpdatedUnix) - int64(task.CreatedUnix)
  353. }
  354. if task.BenchmarkTypeID > 0 {
  355. for _, benchmarkType := range benchmarkTypes.BenchmarkType {
  356. if task.BenchmarkTypeID == benchmarkType.Id {
  357. ctx.Data["BenchmarkTypeName"] = benchmarkType.First
  358. for _, benchmarkChildType := range benchmarkType.Second {
  359. if task.BenchmarkChildTypeID == benchmarkChildType.Id {
  360. ctx.Data["BenchmarkChildTypeName"] = benchmarkChildType.Value
  361. break
  362. }
  363. }
  364. break
  365. }
  366. }
  367. }
  368. ctx.Data["duration"] = util.AddZero(duration/3600000) + ":" + util.AddZero(duration%3600000/60000) + ":" + util.AddZero(duration%60000/1000)
  369. ctx.Data["task"] = task
  370. // ctx.Data["jobID"] = task.JobID
  371. ctx.Data["jobName"] = task.JobName
  372. version_list_task := make([]*models.Cloudbrain, 0)
  373. version_list_task = append(version_list_task, task)
  374. ctx.Data["version_list_task"] = version_list_task
  375. ctx.Data["debugListType"] = debugListType
  376. ctx.HTML(200, tpName)
  377. }
  378. func CloudBrainDebug(ctx *context.Context) {
  379. debugUrl := setting.DebugServerHost + "jpylab_" + ctx.Cloudbrain.JobID + "_" + ctx.Cloudbrain.SubTaskName
  380. ctx.Redirect(debugUrl)
  381. }
  382. func CloudBrainCommitImage(ctx *context.Context, form auth.CommitImageCloudBrainForm) {
  383. err := cloudbrain.CommitImage(ctx.Cloudbrain.JobID, models.CommitImageParams{
  384. Ip: ctx.Cloudbrain.ContainerIp,
  385. TaskContainerId: ctx.Cloudbrain.ContainerID,
  386. ImageDescription: form.Description,
  387. ImageTag: form.Tag,
  388. })
  389. if err != nil {
  390. log.Error("CommitImage(%s) failed:%v", ctx.Cloudbrain.JobName, err.Error(), ctx.Data["msgID"])
  391. ctx.JSON(200, map[string]string{
  392. "result_code": "-1",
  393. "error_msg": "CommitImage failed",
  394. })
  395. return
  396. }
  397. ctx.JSON(200, map[string]string{
  398. "result_code": "0",
  399. "error_msg": "",
  400. })
  401. }
  402. func CloudBrainStop(ctx *context.Context) {
  403. var jobID = ctx.Params(":jobid")
  404. var resultCode = "0"
  405. var errorMsg = ""
  406. var status = ""
  407. task := ctx.Cloudbrain
  408. for {
  409. if task.Status == string(models.JobStopped) || task.Status == string(models.JobFailed) || task.Status == string(models.JobSucceeded) {
  410. log.Error("the job(%s) has been stopped", task.JobName, ctx.Data["msgID"])
  411. resultCode = "-1"
  412. errorMsg = "system error"
  413. break
  414. }
  415. err := cloudbrain.StopJob(jobID)
  416. if err != nil {
  417. log.Error("StopJob(%s) failed:%v", task.JobName, err, ctx.Data["msgID"])
  418. resultCode = "-1"
  419. errorMsg = "system error"
  420. break
  421. }
  422. task.Status = string(models.JobStopped)
  423. err = models.UpdateJob(task)
  424. if err != nil {
  425. log.Error("UpdateJob(%s) failed:%v", task.JobName, err, ctx.Data["msgID"])
  426. resultCode = "-1"
  427. errorMsg = "system error"
  428. break
  429. }
  430. status = task.Status
  431. break
  432. }
  433. ctx.JSON(200, map[string]string{
  434. "result_code": resultCode,
  435. "error_msg": errorMsg,
  436. "status": status,
  437. "job_id": jobID,
  438. })
  439. }
  440. func StopJobsByUserID(userID int64) {
  441. cloudBrains, err := models.GetCloudbrainsNeededStopByUserID(userID)
  442. if err != nil {
  443. log.Warn("Failed to get cloudBrain info", err)
  444. return
  445. }
  446. StopJobs(cloudBrains)
  447. }
  448. func StopJobsByRepoID(repoID int64) {
  449. cloudBrains, err := models.GetCloudbrainsNeededStopByRepoID(repoID)
  450. if err != nil {
  451. log.Warn("Failed to get cloudBrain info", err)
  452. return
  453. }
  454. StopJobs(cloudBrains)
  455. }
  456. /**
  457. */
  458. func StopJobs(cloudBrains []*models.Cloudbrain) {
  459. for _, taskInfo := range cloudBrains {
  460. if taskInfo.Type == models.TypeCloudBrainOne {
  461. err := retry(3, time.Second*30, func() error {
  462. return cloudbrain.StopJob(taskInfo.JobID)
  463. })
  464. logErrorAndUpdateJobStatus(err, taskInfo)
  465. } else {
  466. if taskInfo.JobType == string(models.JobTypeTrain) {
  467. err := retry(3, time.Second*30, func() error {
  468. _, err := modelarts.StopTrainJob(taskInfo.JobID, strconv.FormatInt(taskInfo.VersionID, 10))
  469. return err
  470. })
  471. logErrorAndUpdateJobStatus(err, taskInfo)
  472. } else {
  473. param := models.NotebookAction{
  474. Action: models.ActionStop,
  475. }
  476. err := retry(3, time.Second*30, func() error {
  477. _, err := modelarts.ManageNotebook(taskInfo.JobID, param)
  478. return err
  479. })
  480. logErrorAndUpdateJobStatus(err, taskInfo)
  481. }
  482. }
  483. }
  484. }
  485. func retry(attempts int, sleep time.Duration, f func() error) (err error) {
  486. for i := 0; i < attempts; i++ {
  487. if i > 0 {
  488. log.Warn("retrying after error:", err)
  489. time.Sleep(sleep)
  490. }
  491. err = f()
  492. if err == nil {
  493. return nil
  494. }
  495. }
  496. return fmt.Errorf("after %d attempts, last error: %s", attempts, err)
  497. }
  498. func logErrorAndUpdateJobStatus(err error, taskInfo *models.Cloudbrain) {
  499. if err != nil {
  500. log.Warn("Failed to stop cloudBrain job:"+taskInfo.JobID, err)
  501. } else {
  502. taskInfo.Status = string(models.JobStopped)
  503. err = models.UpdateJob(taskInfo)
  504. if err != nil {
  505. log.Warn("UpdateJob failed", err)
  506. }
  507. }
  508. }
  509. func CloudBrainDel(ctx *context.Context) {
  510. var listType = ctx.Query("debugListType")
  511. if err := deleteCloudbrainJob(ctx); err != nil {
  512. log.Error("deleteCloudbrainJob failed: %v", err, ctx.Data["msgID"])
  513. ctx.ServerError(err.Error(), err)
  514. return
  515. }
  516. var isAdminPage = ctx.Query("isadminpage")
  517. if ctx.IsUserSiteAdmin() && isAdminPage == "true" {
  518. ctx.Redirect(setting.AppSubURL + "/admin" + "/cloudbrains")
  519. } else {
  520. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=" + listType)
  521. }
  522. }
  523. func deleteCloudbrainJob(ctx *context.Context) error {
  524. task := ctx.Cloudbrain
  525. if task.Status != string(models.JobStopped) && task.Status != string(models.JobFailed) && task.Status != string(models.JobSucceeded) {
  526. log.Error("the job(%s) has not been stopped", task.JobName, ctx.Data["msgID"])
  527. return errors.New("the job has not been stopped")
  528. }
  529. err := models.DeleteJob(task)
  530. if err != nil {
  531. log.Error("DeleteJob failed: %v", err, ctx.Data["msgID"])
  532. return err
  533. }
  534. deleteJobStorage(task.JobName, models.TypeCloudBrainOne)
  535. return nil
  536. }
  537. func CloudBrainShowModels(ctx *context.Context) {
  538. ctx.Data["PageIsCloudBrain"] = true
  539. jobID := ctx.Params(":jobid")
  540. parentDir := ctx.Query("parentDir")
  541. dirArray := strings.Split(parentDir, "/")
  542. task, err := models.GetCloudbrainByJobID(jobID)
  543. if err != nil {
  544. log.Error("no such job!", ctx.Data["msgID"])
  545. ctx.ServerError("no such job:", err)
  546. return
  547. }
  548. //get dirs
  549. dirs, err := GetModelDirs(task.JobName, parentDir)
  550. if err != nil {
  551. log.Error("GetModelDirs failed:%v", err.Error(), ctx.Data["msgID"])
  552. ctx.ServerError("GetModelDirs failed:", err)
  553. return
  554. }
  555. var fileInfos []storage.FileInfo
  556. err = json.Unmarshal([]byte(dirs), &fileInfos)
  557. if err != nil {
  558. log.Error("json.Unmarshal failed:%v", err.Error(), ctx.Data["msgID"])
  559. ctx.ServerError("json.Unmarshal failed:", err)
  560. return
  561. }
  562. for i, fileInfo := range fileInfos {
  563. temp, _ := time.Parse("2006-01-02 15:04:05", fileInfo.ModTime)
  564. fileInfos[i].ModTime = temp.Local().Format("2006-01-02 15:04:05")
  565. }
  566. sort.Slice(fileInfos, func(i, j int) bool {
  567. return fileInfos[i].ModTime > fileInfos[j].ModTime
  568. })
  569. ctx.Data["Path"] = dirArray
  570. ctx.Data["Dirs"] = fileInfos
  571. ctx.Data["task"] = task
  572. ctx.Data["JobID"] = jobID
  573. ctx.HTML(200, tplCloudBrainShowModels)
  574. }
  575. func GetPublicImages(ctx *context.Context) {
  576. getImages(ctx, cloudbrain.Public)
  577. }
  578. func GetCustomImages(ctx *context.Context) {
  579. getImages(ctx, cloudbrain.Custom)
  580. }
  581. func getImages(ctx *context.Context, imageType string) {
  582. log.Info("Get images begin")
  583. page := ctx.QueryInt("page")
  584. size := ctx.QueryInt("size")
  585. name := ctx.Query("name")
  586. getImagesResult, err := cloudbrain.GetImagesPageable(page, size, imageType, name)
  587. if err != nil {
  588. log.Error("Can not get images:%v", err)
  589. ctx.JSON(http.StatusOK, models.GetImagesPayload{
  590. Count: 0,
  591. TotalPages: 0,
  592. ImageInfo: []*models.ImageInfo{},
  593. })
  594. } else {
  595. ctx.JSON(http.StatusOK, getImagesResult.Payload)
  596. }
  597. log.Info("Get images end")
  598. }
  599. func GetModelDirs(jobName string, parentDir string) (string, error) {
  600. var req string
  601. modelActualPath := storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/")
  602. if parentDir == "" {
  603. req = "baseDir=" + modelActualPath
  604. } else {
  605. req = "baseDir=" + modelActualPath + "&parentDir=" + parentDir
  606. }
  607. return getDirs(req)
  608. }
  609. func CloudBrainDownloadModel(ctx *context.Context) {
  610. parentDir := ctx.Query("parentDir")
  611. fileName := ctx.Query("fileName")
  612. jobName := ctx.Query("jobName")
  613. filePath := "jobs/" + jobName + "/model/" + parentDir
  614. url, err := storage.Attachments.PresignedGetURL(filePath, fileName)
  615. if err != nil {
  616. log.Error("PresignedGetURL failed: %v", err.Error(), ctx.Data["msgID"])
  617. ctx.ServerError("PresignedGetURL", err)
  618. return
  619. }
  620. http.Redirect(ctx.Resp, ctx.Req.Request, url, http.StatusMovedPermanently)
  621. }
  622. func GetRate(ctx *context.Context) {
  623. isObjectDetcionAll := ctx.QueryBool("isObjectDetcionAll")
  624. if isObjectDetcionAll {
  625. ctx.Redirect(setting.BenchmarkServerHost + "?username=admin")
  626. return
  627. }
  628. var jobID = ctx.Params(":jobid")
  629. job, err := models.GetCloudbrainByJobID(jobID)
  630. if err != nil {
  631. ctx.ServerError("GetCloudbrainByJobID failed", err)
  632. return
  633. }
  634. if job.JobType == string(models.JobTypeBenchmark) {
  635. log.Info("url=" + setting.BenchmarkServerHost + "?username=" + ctx.User.Name)
  636. ctx.Redirect(setting.BenchmarkServerHost + "?username=" + ctx.User.Name)
  637. } else if job.JobType == string(models.JobTypeSnn4imagenet) {
  638. ctx.Redirect(setting.Snn4imagenetServerHost)
  639. } else if job.JobType == string(models.JobTypeBrainScore) {
  640. ctx.Redirect(setting.BrainScoreServerHost)
  641. } else {
  642. log.Error("JobType error:%s", job.JobType, ctx.Data["msgID"])
  643. }
  644. }
  645. func downloadCode(repo *models.Repository, codePath string) error {
  646. if err := git.Clone(repo.RepoPath(), codePath, git.CloneRepoOptions{}); err != nil {
  647. log.Error("Failed to clone repository: %s (%v)", repo.FullName(), err)
  648. return err
  649. }
  650. configFile, err := os.OpenFile(codePath+"/.git/config", os.O_RDWR, 0666)
  651. if err != nil {
  652. log.Error("open file(%s) failed:%v", codePath+"/,git/config", err)
  653. return err
  654. }
  655. defer configFile.Close()
  656. pos := int64(0)
  657. reader := bufio.NewReader(configFile)
  658. for {
  659. line, err := reader.ReadString('\n')
  660. if err != nil {
  661. if err == io.EOF {
  662. log.Error("not find the remote-url")
  663. return nil
  664. } else {
  665. log.Error("read error: %v", err)
  666. return err
  667. }
  668. }
  669. if strings.Contains(line, "url") && strings.Contains(line, ".git") {
  670. originUrl := "\turl = " + repo.CloneLink().HTTPS + "\n"
  671. if len(line) > len(originUrl) {
  672. originUrl += strings.Repeat(" ", len(line)-len(originUrl))
  673. }
  674. bytes := []byte(originUrl)
  675. _, err := configFile.WriteAt(bytes, pos)
  676. if err != nil {
  677. log.Error("WriteAt failed:%v", err)
  678. return err
  679. }
  680. break
  681. }
  682. pos += int64(len(line))
  683. }
  684. return nil
  685. }
  686. func downloadRateCode(repo *models.Repository, taskName, rateOwnerName, rateRepoName, codePath, benchmarkCategory, gpuType string) error {
  687. err := os.MkdirAll(codePath, os.ModePerm)
  688. if err != nil {
  689. log.Error("mkdir codePath failed", err.Error())
  690. return err
  691. }
  692. repoExt, err := models.GetRepositoryByOwnerAndName(rateOwnerName, rateRepoName)
  693. if err != nil {
  694. log.Error("GetRepositoryByOwnerAndName(%s) failed", rateRepoName, err.Error())
  695. return err
  696. }
  697. if err := git.Clone(repoExt.RepoPath(), codePath, git.CloneRepoOptions{}); err != nil {
  698. log.Error("Failed to clone repository: %s (%v)", repoExt.FullName(), err)
  699. return err
  700. }
  701. fileName := codePath + cloudbrain.TaskInfoName
  702. f, err := os.OpenFile(fileName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, os.ModePerm)
  703. if err != nil {
  704. log.Error("OpenFile failed", err.Error())
  705. return err
  706. }
  707. defer f.Close()
  708. data, err := json.Marshal(models.TaskInfo{
  709. Username: repo.Owner.Name,
  710. TaskName: taskName,
  711. CodeName: repo.Name,
  712. BenchmarkCategory: strings.Split(benchmarkCategory, ","),
  713. CodeLink: strings.TrimSuffix(repo.CloneLink().HTTPS, ".git"),
  714. GpuType: gpuType,
  715. })
  716. if err != nil {
  717. log.Error("json.Marshal failed", err.Error())
  718. return err
  719. }
  720. _, err = f.Write(data)
  721. if err != nil {
  722. log.Error("WriteString failed", err.Error())
  723. return err
  724. }
  725. return nil
  726. }
  727. func uploadCodeToMinio(codePath, jobName, parentDir string) error {
  728. files, err := readDir(codePath)
  729. if err != nil {
  730. log.Error("readDir(%s) failed: %s", codePath, err.Error())
  731. return err
  732. }
  733. for _, file := range files {
  734. if file.IsDir() {
  735. if err = uploadCodeToMinio(codePath+file.Name()+"/", jobName, parentDir+file.Name()+"/"); err != nil {
  736. log.Error("uploadCodeToMinio(%s) failed: %s", file.Name(), err.Error())
  737. return err
  738. }
  739. } else {
  740. destObject := setting.CBCodePathPrefix + jobName + parentDir + file.Name()
  741. sourceFile := codePath + file.Name()
  742. err = storage.Attachments.UploadObject(destObject, sourceFile)
  743. if err != nil {
  744. log.Error("UploadObject(%s) failed: %s", file.Name(), err.Error())
  745. return err
  746. }
  747. }
  748. }
  749. return nil
  750. }
  751. func mkModelPath(modelPath string) error {
  752. err := os.MkdirAll(modelPath, os.ModePerm)
  753. if err != nil {
  754. log.Error("MkdirAll(%s) failed:%v", modelPath, err)
  755. return err
  756. }
  757. fileName := modelPath + "README"
  758. f, err := os.OpenFile(fileName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, os.ModePerm)
  759. if err != nil {
  760. log.Error("OpenFile failed", err.Error())
  761. return err
  762. }
  763. defer f.Close()
  764. _, err = f.WriteString("You can put the model file into this directory and download it by the web page.")
  765. if err != nil {
  766. log.Error("WriteString failed", err.Error())
  767. return err
  768. }
  769. return nil
  770. }
  771. func deleteJobStorage(jobName string, cloudbrainType int) error {
  772. //delete local
  773. localJobPath := setting.JobPath + jobName
  774. err := os.RemoveAll(localJobPath)
  775. if err != nil {
  776. log.Error("RemoveAll(%s) failed:%v", localJobPath, err)
  777. }
  778. //delete oss
  779. if cloudbrainType == models.TypeCloudBrainOne {
  780. dirPath := setting.CBCodePathPrefix + jobName + "/"
  781. err = storage.Attachments.DeleteDir(dirPath)
  782. if err != nil {
  783. log.Error("DeleteDir(%s) failed:%v", localJobPath, err)
  784. }
  785. } else if cloudbrainType == models.TypeCloudBrainTwo {
  786. //dirPath := setting.CodePathPrefix + jobName + "/"
  787. //err = storage.ObsRemoveObject(setting.Bucket, dirPath)
  788. //if err != nil {
  789. // log.Error("ObsRemoveObject(%s) failed:%v", localJobPath, err)
  790. //}
  791. log.Info("no need to delete")
  792. } else {
  793. log.Error("cloudbrainType(%d) error", cloudbrainType)
  794. }
  795. return nil
  796. }
  797. func SyncCloudbrainStatus() {
  798. cloudBrains, err := models.GetCloudBrainUnStoppedJob()
  799. if err != nil {
  800. log.Error("GetCloudBrainUnStoppedJob failed:", err.Error())
  801. return
  802. }
  803. for _, task := range cloudBrains {
  804. if task.Type == models.TypeCloudBrainOne {
  805. result, err := cloudbrain.GetJob(task.JobID)
  806. if err != nil {
  807. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  808. continue
  809. }
  810. if result != nil {
  811. jobRes, _ := models.ConvertToJobResultPayload(result.Payload)
  812. taskRoles := jobRes.TaskRoles
  813. taskRes, _ := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
  814. task.Status = taskRes.TaskStatuses[0].State
  815. if task.Status != string(models.JobWaiting) {
  816. task.Duration = time.Now().Unix() - taskRes.TaskStatuses[0].StartAt.Unix()
  817. err = models.UpdateJob(task)
  818. if err != nil {
  819. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  820. }
  821. var maxDuration int64
  822. if task.JobType == string(models.JobTypeBenchmark) {
  823. maxDuration = setting.BenchmarkMaxDuration
  824. } else {
  825. maxDuration = setting.MaxDuration
  826. }
  827. if task.Duration >= maxDuration {
  828. log.Info("begin to stop job(%s), because of the duration", task.JobName)
  829. err = cloudbrain.StopJob(task.JobID)
  830. if err != nil {
  831. log.Error("StopJob(%s) failed:%v", task.JobName, err)
  832. continue
  833. }
  834. task.Status = string(models.JobStopped)
  835. err = models.UpdateJob(task)
  836. if err != nil {
  837. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  838. }
  839. }
  840. }
  841. }
  842. } else if task.Type == models.TypeCloudBrainTwo {
  843. if task.JobType == string(models.JobTypeDebug) {
  844. //result, err := modelarts.GetJob(task.JobID)
  845. result, err := modelarts.GetNotebook2(task.JobID)
  846. if err != nil {
  847. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  848. continue
  849. }
  850. if result != nil {
  851. task.Status = result.Status
  852. err = models.UpdateJob(task)
  853. if err != nil {
  854. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  855. continue
  856. }
  857. }
  858. } else if task.JobType == string(models.JobTypeTrain) {
  859. result, err := modelarts.GetTrainJob(task.JobID, strconv.FormatInt(task.VersionID, 10))
  860. if err != nil {
  861. log.Error("GetTrainJob(%s) failed:%v", task.JobName, err)
  862. continue
  863. }
  864. if result != nil {
  865. task.Status = modelarts.TransTrainJobStatus(result.IntStatus)
  866. task.Duration = result.Duration
  867. task.TrainJobDuration = result.TrainJobDuration
  868. if result.Duration != 0 {
  869. task.TrainJobDuration = util.AddZero(result.Duration/3600000) + ":" + util.AddZero(result.Duration%3600000/60000) + ":" + util.AddZero(result.Duration%60000/1000)
  870. } else {
  871. task.TrainJobDuration = "00:00:00"
  872. }
  873. err = models.UpdateJob(task)
  874. if err != nil {
  875. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  876. continue
  877. }
  878. }
  879. } else {
  880. log.Error("task.JobType(%s) is error:%s", task.JobName, task.JobType)
  881. }
  882. } else {
  883. log.Error("task.Type(%s) is error:%d", task.JobName, task.Type)
  884. }
  885. }
  886. return
  887. }
  888. func CloudBrainBenchmarkIndex(ctx *context.Context) {
  889. MustEnableCloudbrain(ctx)
  890. repo := ctx.Repo.Repository
  891. page := ctx.QueryInt("page")
  892. if page <= 0 {
  893. page = 1
  894. }
  895. var jobTypes []string
  896. jobTypes = append(jobTypes, string(models.JobTypeBenchmark))
  897. ciTasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  898. ListOptions: models.ListOptions{
  899. Page: page,
  900. PageSize: setting.UI.IssuePagingNum,
  901. },
  902. RepoID: repo.ID,
  903. Type: models.TypeCloudBrainOne,
  904. JobTypes: jobTypes,
  905. })
  906. if err != nil {
  907. ctx.ServerError("Get debugjob faild:", err)
  908. return
  909. }
  910. if benchmarkTypes == nil {
  911. if err := json.Unmarshal([]byte(setting.BenchmarkTypes), &benchmarkTypes); err != nil {
  912. ctx.ServerError("Get BenchmarkTypes faild:", err)
  913. return
  914. }
  915. }
  916. for i, task := range ciTasks {
  917. ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  918. ciTasks[i].Cloudbrain.ComputeResource = task.ComputeResource
  919. var duration int64
  920. if task.Status == string(models.JobRunning) {
  921. duration = time.Now().Unix() - int64(task.Cloudbrain.CreatedUnix)
  922. } else {
  923. duration = int64(task.Cloudbrain.UpdatedUnix) - int64(task.Cloudbrain.CreatedUnix)
  924. }
  925. ciTasks[i].TrainJobDuration = util.AddZero(duration/3600000) + ":" + util.AddZero(duration%3600000/60000) + ":" + util.AddZero(duration%60000/1000)
  926. ciTasks[i].BenchmarkTypeName = ""
  927. if task.BenchmarkTypeID > 0 {
  928. for _, benchmarkType := range benchmarkTypes.BenchmarkType {
  929. if task.BenchmarkTypeID == benchmarkType.Id {
  930. ciTasks[i].BenchmarkTypeRankLink = benchmarkType.RankLink
  931. ciTasks[i].BenchmarkTypeName = benchmarkType.First
  932. break
  933. }
  934. }
  935. }
  936. }
  937. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  938. ctx.Data["Page"] = pager
  939. ctx.Data["PageIsCloudBrain"] = true
  940. ctx.Data["Tasks"] = ciTasks
  941. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  942. ctx.Data["RepoIsEmpty"] = repo.IsEmpty
  943. ctx.HTML(200, tplCloudBrainBenchmarkIndex)
  944. }
  945. func GetChildTypes(ctx *context.Context) {
  946. benchmarkTypeID := ctx.QueryInt("benchmark_type_id")
  947. re := make(map[string]interface{})
  948. for {
  949. if benchmarkTypes == nil {
  950. if err := json.Unmarshal([]byte(setting.BenchmarkTypes), &benchmarkTypes); err != nil {
  951. log.Error("json.Unmarshal BenchmarkTypes(%s) failed:%v", setting.BenchmarkTypes, err, ctx.Data["MsgID"])
  952. re["errMsg"] = "system error"
  953. break
  954. }
  955. }
  956. var isExist bool
  957. for _, benchmarkType := range benchmarkTypes.BenchmarkType {
  958. if benchmarkTypeID == benchmarkType.Id {
  959. isExist = true
  960. re["child_types"] = benchmarkType.Second
  961. re["result_code"] = "0"
  962. break
  963. }
  964. }
  965. if !isExist {
  966. re["result_code"] = "1"
  967. log.Error("no such benchmark_type_id", ctx.Data["MsgID"])
  968. re["errMsg"] = "system error"
  969. break
  970. }
  971. break
  972. }
  973. ctx.JSON(200, re)
  974. }
  975. func CloudBrainBenchmarkNew(ctx *context.Context) {
  976. ctx.Data["description"] = ""
  977. ctx.Data["benchmarkTypeID"] = -1
  978. ctx.Data["benchmark_child_types_id_hidden"] = -1
  979. err := cloudBrainNewDataPrepare(ctx)
  980. if err != nil {
  981. ctx.ServerError("get new cloudbrain info failed", err)
  982. return
  983. }
  984. ctx.HTML(200, tplCloudBrainBenchmarkNew)
  985. }
  986. func getBenchmarkAttachment(benchmarkTypeID, benchmarkChildTypeID int) (*models.BenchmarkDataset, error) {
  987. var childInfo *models.BenchmarkDataset
  988. if benchmarkTypes == nil {
  989. if err := json.Unmarshal([]byte(setting.BenchmarkTypes), &benchmarkTypes); err != nil {
  990. log.Error("json.Unmarshal BenchmarkTypes(%s) failed:%v", setting.BenchmarkTypes, err)
  991. return childInfo, err
  992. }
  993. }
  994. var isExist bool
  995. for _, benchmarkType := range benchmarkTypes.BenchmarkType {
  996. if benchmarkType.Id == benchmarkTypeID {
  997. for _, childType := range benchmarkType.Second {
  998. if childType.Id == benchmarkChildTypeID {
  999. childInfo = childType
  1000. isExist = true
  1001. break
  1002. }
  1003. }
  1004. break
  1005. }
  1006. }
  1007. if !isExist {
  1008. log.Error("no such benchmark_type_id&benchmark_child_type_id")
  1009. return childInfo, errors.New("no such benchmark_type_id&benchmark_child_type_id")
  1010. }
  1011. return childInfo, nil
  1012. }
  1013. func getBenchmarkGpuQueue(gpuQueue string) (string, error) {
  1014. queue := ""
  1015. if benchmarkGpuInfos == nil {
  1016. if err := json.Unmarshal([]byte(setting.BenchmarkGpuTypes), &benchmarkGpuInfos); err != nil {
  1017. log.Error("json.Unmarshal BenchmarkGpuTypes(%s) failed:%v", setting.BenchmarkGpuTypes, err)
  1018. return queue, err
  1019. }
  1020. }
  1021. var isExist bool
  1022. for _, gpuInfo := range benchmarkGpuInfos.GpuInfo {
  1023. if gpuQueue == gpuInfo.Queue {
  1024. isExist = true
  1025. queue = gpuQueue
  1026. break
  1027. }
  1028. }
  1029. if !isExist {
  1030. log.Error("no such gpuQueue, %s", gpuQueue)
  1031. return queue, errors.New("no such gpuQueue")
  1032. }
  1033. return queue, nil
  1034. }
  1035. func getBenchmarkResourceSpec(resourceSpecID int) (int, error) {
  1036. var id int
  1037. if benchmarkResourceSpecs == nil {
  1038. if err := json.Unmarshal([]byte(setting.BenchmarkResourceSpecs), &benchmarkResourceSpecs); err != nil {
  1039. log.Error("json.Unmarshal BenchmarkResourceSpecs(%s) failed:%v", setting.BenchmarkResourceSpecs, err)
  1040. return id, err
  1041. }
  1042. }
  1043. var isExist bool
  1044. for _, resourceSpec := range benchmarkResourceSpecs.ResourceSpec {
  1045. if resourceSpecID == resourceSpec.Id {
  1046. isExist = true
  1047. id = resourceSpecID
  1048. break
  1049. }
  1050. }
  1051. if !isExist {
  1052. log.Error("no such resourceSpecID, %d", resourceSpecID)
  1053. return id, errors.New("no such resourceSpec")
  1054. }
  1055. return id, nil
  1056. }
  1057. func CloudBrainBenchmarkCreate(ctx *context.Context, form auth.CreateCloudBrainForm) {
  1058. ctx.Data["PageIsCloudBrain"] = true
  1059. jobName := form.JobName
  1060. image := form.Image
  1061. gpuQueue := form.GpuType
  1062. command := cloudbrain.CommandBenchmark
  1063. codePath := setting.JobPath + jobName + cloudbrain.CodeMountPath
  1064. resourceSpecId := cloudbrain.BenchMarkResourceID
  1065. benchmarkTypeID := form.BenchmarkTypeID
  1066. benchmarkChildTypeID := form.BenchmarkChildTypeID
  1067. ctx.Data["description"] = form.Description
  1068. ctx.Data["benchmarkTypeID"] = benchmarkTypeID
  1069. ctx.Data["benchmark_child_types_id_hidden"] = benchmarkChildTypeID
  1070. if !jobNamePattern.MatchString(jobName) {
  1071. cloudBrainNewDataPrepare(ctx)
  1072. ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_jobname_err"), tplCloudBrainBenchmarkNew, &form)
  1073. return
  1074. }
  1075. childInfo, err := getBenchmarkAttachment(benchmarkTypeID, benchmarkChildTypeID)
  1076. if err != nil {
  1077. log.Error("getBenchmarkAttachment failed:%v", err, ctx.Data["MsgID"])
  1078. cloudBrainNewDataPrepare(ctx)
  1079. ctx.RenderWithErr("benchmark type error", tplCloudBrainBenchmarkNew, &form)
  1080. return
  1081. }
  1082. _, err = getBenchmarkGpuQueue(gpuQueue)
  1083. if err != nil {
  1084. log.Error("getBenchmarkGpuQueue failed:%v", err, ctx.Data["MsgID"])
  1085. cloudBrainNewDataPrepare(ctx)
  1086. ctx.RenderWithErr("gpu queue error", tplCloudBrainBenchmarkNew, &form)
  1087. return
  1088. }
  1089. _, err = getBenchmarkResourceSpec(resourceSpecId)
  1090. if err != nil {
  1091. log.Error("getBenchmarkResourceSpec failed:%v", err, ctx.Data["MsgID"])
  1092. cloudBrainNewDataPrepare(ctx)
  1093. ctx.RenderWithErr("resource spec error", tplCloudBrainBenchmarkNew, &form)
  1094. return
  1095. }
  1096. count, err := models.GetCloudbrainCountByUserID(ctx.User.ID, string(models.JobTypeBenchmark))
  1097. if err != nil {
  1098. log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])
  1099. cloudBrainNewDataPrepare(ctx)
  1100. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1101. return
  1102. } else {
  1103. if count >= 1 {
  1104. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  1105. cloudBrainNewDataPrepare(ctx)
  1106. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplCloudBrainBenchmarkNew, &form)
  1107. return
  1108. }
  1109. }
  1110. _, err = models.GetCloudbrainByName(jobName)
  1111. if err == nil {
  1112. log.Error("the job name did already exist", ctx.Data["MsgID"])
  1113. cloudBrainNewDataPrepare(ctx)
  1114. ctx.RenderWithErr("the job name did already exist", tplCloudBrainBenchmarkNew, &form)
  1115. return
  1116. } else {
  1117. if !models.IsErrJobNotExist(err) {
  1118. log.Error("GetCloudbrainByName failed, %v", err, ctx.Data["MsgID"])
  1119. cloudBrainNewDataPrepare(ctx)
  1120. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1121. return
  1122. }
  1123. }
  1124. repo := ctx.Repo.Repository
  1125. os.RemoveAll(codePath)
  1126. if err := downloadCode(repo, codePath); err != nil {
  1127. log.Error("downloadCode failed, %v", err, ctx.Data["MsgID"])
  1128. cloudBrainNewDataPrepare(ctx)
  1129. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1130. return
  1131. }
  1132. if _, err := os.Stat(codePath + "/train.py"); err != nil {
  1133. if os.IsNotExist(err) {
  1134. // file does not exist
  1135. log.Error("train.py does not exist, %v", err, ctx.Data["MsgID"])
  1136. cloudBrainNewDataPrepare(ctx)
  1137. ctx.RenderWithErr("train.py does not exist", tplCloudBrainBenchmarkNew, &form)
  1138. } else {
  1139. log.Error("Stat failed, %v", err, ctx.Data["MsgID"])
  1140. cloudBrainNewDataPrepare(ctx)
  1141. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1142. }
  1143. return
  1144. } else if _, err := os.Stat(codePath + "/test.py"); err != nil {
  1145. if os.IsNotExist(err) {
  1146. // file does not exist
  1147. log.Error("test.py does not exist, %v", err, ctx.Data["MsgID"])
  1148. cloudBrainNewDataPrepare(ctx)
  1149. ctx.RenderWithErr("test.py does not exist", tplCloudBrainBenchmarkNew, &form)
  1150. } else {
  1151. log.Error("Stat failed, %v", err, ctx.Data["MsgID"])
  1152. cloudBrainNewDataPrepare(ctx)
  1153. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1154. }
  1155. return
  1156. }
  1157. if err := uploadCodeToMinio(codePath+"/", jobName, cloudbrain.CodeMountPath+"/"); err != nil {
  1158. log.Error("uploadCodeToMinio failed, %v", err, ctx.Data["MsgID"])
  1159. cloudBrainNewDataPrepare(ctx)
  1160. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1161. return
  1162. }
  1163. benchmarkPath := setting.JobPath + jobName + cloudbrain.BenchMarkMountPath
  1164. var gpuType string
  1165. for _, gpuInfo := range gpuInfos.GpuInfo {
  1166. if gpuInfo.Queue == gpuQueue {
  1167. gpuType = gpuInfo.Value
  1168. }
  1169. }
  1170. if err := downloadRateCode(repo, jobName, childInfo.Owner, childInfo.RepoName, benchmarkPath, form.BenchmarkCategory, gpuType); err != nil {
  1171. log.Error("downloadRateCode failed, %v", err, ctx.Data["MsgID"])
  1172. //cloudBrainNewDataPrepare(ctx)
  1173. //ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1174. //return
  1175. }
  1176. if err := uploadCodeToMinio(benchmarkPath+"/", jobName, cloudbrain.BenchMarkMountPath+"/"); err != nil {
  1177. log.Error("uploadCodeToMinio failed, %v", err, ctx.Data["MsgID"])
  1178. //cloudBrainNewDataPrepare(ctx)
  1179. //ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1180. //return
  1181. }
  1182. err = cloudbrain.GenerateTask(ctx, jobName, image, command, childInfo.Attachment, storage.GetMinioPath(jobName, cloudbrain.CodeMountPath+"/"),
  1183. storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/"),
  1184. storage.GetMinioPath(jobName, cloudbrain.BenchMarkMountPath+"/"), storage.GetMinioPath(jobName, cloudbrain.Snn4imagenetMountPath+"/"),
  1185. storage.GetMinioPath(jobName, cloudbrain.BrainScoreMountPath+"/"), string(models.JobTypeBenchmark), gpuQueue, form.Description,
  1186. benchmarkTypeID, benchmarkChildTypeID, resourceSpecId)
  1187. if err != nil {
  1188. cloudBrainNewDataPrepare(ctx)
  1189. ctx.RenderWithErr(err.Error(), tplCloudBrainBenchmarkNew, &form)
  1190. return
  1191. }
  1192. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/cloudbrain/benchmark")
  1193. }
  1194. func BenchmarkDel(ctx *context.Context) {
  1195. if err := deleteCloudbrainJob(ctx); err != nil {
  1196. log.Error("deleteCloudbrainJob failed: %v", err, ctx.Data["msgID"])
  1197. ctx.ServerError(err.Error(), err)
  1198. return
  1199. }
  1200. var isAdminPage = ctx.Query("isadminpage")
  1201. if ctx.IsUserSiteAdmin() && isAdminPage == "true" {
  1202. ctx.Redirect(setting.AppSubURL + "/admin" + "/cloudbrains")
  1203. } else {
  1204. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/cloudbrain/benchmark")
  1205. }
  1206. }