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 42 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
3 years ago
5 years ago
3 years ago
4 years ago
5 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
5 years ago
5 years ago
3 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
3 years ago
3 years ago
3 years ago
3 years ago
4 years ago
3 years ago
5 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
5 years ago
5 years ago
5 years ago
5 years ago
4 years ago
3 years ago
4 years ago
5 years ago
4 years ago
5 years ago
3 years ago
3 years ago
5 years ago
5 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
3 years ago
3 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
3 years ago
3 years ago
3 years ago
5 years ago
3 years ago
3 years ago
3 years ago
3 years ago
5 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
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
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
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
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
12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364
  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. openiJobName := form.JobName
  149. jobName := util.ConvertToApiJobName(openiJobName)
  150. image := form.Image
  151. uuid := form.Attachment
  152. jobType := form.JobType
  153. command := cloudbrain.Command
  154. gpuQueue := form.GpuType
  155. codePath := setting.JobPath + jobName + cloudbrain.CodeMountPath
  156. resourceSpecId := form.ResourceSpecId
  157. //检查任务名称是否重复
  158. var jobTypes []string
  159. jobTypes = append(jobTypes, string(models.JobTypeDebug))
  160. tasks, _, err := models.Cloudbrains(&models.CloudbrainsOptions{
  161. ListOptions: models.ListOptions{
  162. PageSize: setting.UI.IssuePagingNum,
  163. },
  164. RepoID: ctx.Repo.Repository.ID,
  165. Type: -1,
  166. JobTypeNot: false,
  167. JobTypes: jobTypes,
  168. })
  169. if err == nil {
  170. for _, task := range tasks {
  171. if strings.EqualFold(task.JobName, openiJobName) {
  172. log.Error("the job name did already exist", ctx.Data["MsgID"])
  173. cloudBrainNewDataPrepare(ctx)
  174. ctx.RenderWithErr("任务名称已经被使用!", tplCloudBrainNew, &form)
  175. return
  176. }
  177. }
  178. } else {
  179. if !models.IsErrJobNotExist(err) {
  180. log.Error("system error, %v", err, ctx.Data["MsgID"])
  181. cloudBrainNewDataPrepare(ctx)
  182. ctx.RenderWithErr("system error", tplCloudBrainNew, &form)
  183. return
  184. }
  185. }
  186. // if !jobNamePattern.MatchString(jobName) {
  187. // ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_jobname_err"), tplCloudBrainNew, &form)
  188. // return
  189. // }
  190. if jobType != string(models.JobTypeBenchmark) && jobType != string(models.JobTypeDebug) && jobType != string(models.JobTypeSnn4imagenet) && jobType != string(models.JobTypeBrainScore) {
  191. log.Error("jobtype error:", jobType, ctx.Data["MsgID"])
  192. cloudBrainNewDataPrepare(ctx)
  193. ctx.RenderWithErr("jobtype error", tplCloudBrainNew, &form)
  194. return
  195. }
  196. count, err := models.GetCloudbrainCountByUserID(ctx.User.ID, jobType)
  197. if err != nil {
  198. log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])
  199. cloudBrainNewDataPrepare(ctx)
  200. ctx.RenderWithErr("system error", tplCloudBrainNew, &form)
  201. return
  202. } else {
  203. if count >= 1 {
  204. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  205. cloudBrainNewDataPrepare(ctx)
  206. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplCloudBrainNew, &form)
  207. return
  208. }
  209. }
  210. repo := ctx.Repo.Repository
  211. downloadCode(repo, codePath)
  212. uploadCodeToMinio(codePath+"/", jobName, cloudbrain.CodeMountPath+"/")
  213. modelPath := setting.JobPath + jobName + cloudbrain.ModelMountPath + "/"
  214. mkModelPath(modelPath)
  215. uploadCodeToMinio(modelPath, jobName, cloudbrain.ModelMountPath+"/")
  216. benchmarkPath := setting.JobPath + jobName + cloudbrain.BenchMarkMountPath
  217. if setting.IsBenchmarkEnabled && jobType == string(models.JobTypeBenchmark) {
  218. var gpuType string
  219. for _, gpuInfo := range gpuInfos.GpuInfo {
  220. if gpuInfo.Queue == gpuQueue {
  221. gpuType = gpuInfo.Value
  222. }
  223. }
  224. downloadRateCode(repo, jobName, setting.BenchmarkOwner, setting.BenchmarkName, benchmarkPath, form.BenchmarkCategory, gpuType)
  225. uploadCodeToMinio(benchmarkPath+"/", jobName, cloudbrain.BenchMarkMountPath+"/")
  226. }
  227. snn4imagenetPath := setting.JobPath + jobName + cloudbrain.Snn4imagenetMountPath
  228. if setting.IsSnn4imagenetEnabled && jobType == string(models.JobTypeSnn4imagenet) {
  229. downloadRateCode(repo, jobName, setting.Snn4imagenetOwner, setting.Snn4imagenetName, snn4imagenetPath, "", "")
  230. uploadCodeToMinio(snn4imagenetPath+"/", jobName, cloudbrain.Snn4imagenetMountPath+"/")
  231. }
  232. brainScorePath := setting.JobPath + jobName + cloudbrain.BrainScoreMountPath
  233. if setting.IsBrainScoreEnabled && jobType == string(models.JobTypeBrainScore) {
  234. downloadRateCode(repo, jobName, setting.BrainScoreOwner, setting.BrainScoreName, brainScorePath, "", "")
  235. uploadCodeToMinio(brainScorePath+"/", jobName, cloudbrain.BrainScoreMountPath+"/")
  236. }
  237. err = cloudbrain.GenerateTask(ctx, openiJobName, jobName, image, command, uuid, storage.GetMinioPath(jobName, cloudbrain.CodeMountPath+"/"),
  238. storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/"),
  239. storage.GetMinioPath(jobName, cloudbrain.BenchMarkMountPath+"/"), storage.GetMinioPath(jobName, cloudbrain.Snn4imagenetMountPath+"/"),
  240. storage.GetMinioPath(jobName, cloudbrain.BrainScoreMountPath+"/"), jobType, gpuQueue, form.Description,
  241. 0, 0, resourceSpecId)
  242. if err != nil {
  243. cloudBrainNewDataPrepare(ctx)
  244. ctx.RenderWithErr(err.Error(), tplCloudBrainNew, &form)
  245. return
  246. }
  247. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=all")
  248. }
  249. func CloudBrainRestart(ctx *context.Context) {
  250. var jobID = ctx.Params(":jobid")
  251. var resultCode = "0"
  252. var errorMsg = ""
  253. var status = string(models.JobWaiting)
  254. task := ctx.Cloudbrain
  255. for {
  256. if task.Status != string(models.JobStopped) && task.Status != string(models.JobSucceeded) && task.Status != string(models.JobFailed) {
  257. log.Error("the job(%s) is not stopped", task.JobName, ctx.Data["MsgID"])
  258. resultCode = "-1"
  259. errorMsg = "the job is not stopped"
  260. break
  261. }
  262. if task.Image == "" || task.GpuQueue == "" || task.Type != models.TypeCloudBrainOne {
  263. log.Error("the job(%s) version is too old", task.JobName, ctx.Data["MsgID"])
  264. resultCode = "-1"
  265. errorMsg = "the job's version is too old and can not be restarted"
  266. break
  267. }
  268. if !ctx.IsSigned || (ctx.User.ID != task.UserID && !ctx.IsUserSiteAdmin()) {
  269. log.Error("the user has no right ro restart the job", task.JobName, ctx.Data["MsgID"])
  270. resultCode = "-1"
  271. errorMsg = "you have no right to restart the job"
  272. break
  273. }
  274. count, err := models.GetCloudbrainCountByUserID(ctx.User.ID, string(models.JobTypeDebug))
  275. if err != nil {
  276. log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])
  277. resultCode = "-1"
  278. errorMsg = "system error"
  279. break
  280. } else {
  281. if count >= 1 {
  282. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  283. resultCode = "-1"
  284. errorMsg = "you have already a running or waiting task, can not create more"
  285. break
  286. }
  287. }
  288. err = cloudbrain.RestartTask(ctx, task, &jobID)
  289. if err != nil {
  290. log.Error("RestartTask failed:%v", err.Error(), ctx.Data["MsgID"])
  291. resultCode = "-1"
  292. errorMsg = "system error"
  293. break
  294. }
  295. break
  296. }
  297. ctx.JSON(200, map[string]string{
  298. "result_code": resultCode,
  299. "error_msg": errorMsg,
  300. "status": status,
  301. "job_id": jobID,
  302. })
  303. }
  304. func CloudBrainBenchMarkShow(ctx *context.Context) {
  305. if benchmarkTypes == nil {
  306. if err := json.Unmarshal([]byte(setting.BenchmarkTypes), &benchmarkTypes); err != nil {
  307. log.Error("json.Unmarshal BenchmarkTypes(%s) failed:%v", setting.BenchmarkTypes, err, ctx.Data["MsgID"])
  308. }
  309. }
  310. cloudBrainShow(ctx, tplCloudBrainBenchmarkShow)
  311. }
  312. func CloudBrainShow(ctx *context.Context) {
  313. cloudBrainShow(ctx, tplCloudBrainShow)
  314. }
  315. func cloudBrainShow(ctx *context.Context, tpName base.TplName) {
  316. ctx.Data["PageIsCloudBrain"] = true
  317. var jobID = ctx.Params(":jobid")
  318. task, err := models.GetCloudbrainByJobID(jobID)
  319. if err != nil {
  320. ctx.Data["error"] = err.Error()
  321. }
  322. result, err := cloudbrain.GetJob(jobID)
  323. if err != nil {
  324. ctx.Data["error"] = err.Error()
  325. }
  326. if result != nil {
  327. jobRes, _ := models.ConvertToJobResultPayload(result.Payload)
  328. jobRes.Resource.Memory = strings.ReplaceAll(jobRes.Resource.Memory, "Mi", "MB")
  329. spec := "GPU数:" + strconv.Itoa(jobRes.Resource.NvidiaComGpu) + ",CPU数:" + strconv.Itoa(jobRes.Resource.CPU) + ",内存(MB):" + jobRes.Resource.Memory
  330. ctx.Data["resource_spec"] = spec
  331. taskRoles := jobRes.TaskRoles
  332. if jobRes.JobStatus.State != string(models.JobFailed) {
  333. taskRes, _ := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
  334. ctx.Data["taskRes"] = taskRes
  335. task.Status = taskRes.TaskStatuses[0].State
  336. task.ContainerID = taskRes.TaskStatuses[0].ContainerID
  337. task.ContainerIp = taskRes.TaskStatuses[0].ContainerIP
  338. err = models.UpdateJob(task)
  339. if err != nil {
  340. ctx.Data["error"] = err.Error()
  341. }
  342. } else {
  343. task.Status = jobRes.JobStatus.State
  344. taskRes := models.TaskPod{TaskStatuses: []models.TaskStatuses{
  345. {
  346. State: jobRes.JobStatus.State,
  347. },
  348. }}
  349. ctx.Data["taskRes"] = taskRes
  350. jobRes.JobStatus.StartTime = time.Unix(int64(task.CreatedUnix), 0).Format("2006-01-02 15:04:05")
  351. jobRes.JobStatus.EndTime = time.Unix(int64(task.UpdatedUnix), 0).Format("2006-01-02 15:04:05")
  352. }
  353. ctx.Data["result"] = jobRes
  354. } else {
  355. log.Info("error:" + err.Error())
  356. }
  357. user, err := models.GetUserByID(task.UserID)
  358. if err == nil {
  359. task.User = user
  360. }
  361. var duration int64
  362. if task.Status == string(models.JobRunning) {
  363. duration = time.Now().Unix() - int64(task.CreatedUnix)
  364. } else {
  365. duration = int64(task.UpdatedUnix) - int64(task.CreatedUnix)
  366. }
  367. if benchmarkTypes != nil {
  368. for _, benchmarkType := range benchmarkTypes.BenchmarkType {
  369. if task.BenchmarkTypeID == benchmarkType.Id {
  370. ctx.Data["BenchmarkTypeName"] = benchmarkType.First
  371. for _, benchmarkChildType := range benchmarkType.Second {
  372. if task.BenchmarkChildTypeID == benchmarkChildType.Id {
  373. ctx.Data["BenchmarkChildTypeName"] = benchmarkChildType.Value
  374. break
  375. }
  376. }
  377. break
  378. }
  379. }
  380. }
  381. ctx.Data["duration"] = util.AddZero(duration/3600000) + ":" + util.AddZero(duration%3600000/60000) + ":" + util.AddZero(duration%60000/1000)
  382. ctx.Data["task"] = task
  383. ctx.Data["jobID"] = jobID
  384. ctx.Data["jobName"] = task.JobName
  385. version_list_task := make([]*models.Cloudbrain, 0)
  386. version_list_task = append(version_list_task, task)
  387. ctx.Data["version_list_task"] = version_list_task
  388. ctx.HTML(200, tpName)
  389. }
  390. func CloudBrainDebug(ctx *context.Context) {
  391. debugUrl := setting.DebugServerHost + "jpylab_" + ctx.Cloudbrain.JobID + "_" + ctx.Cloudbrain.SubTaskName
  392. ctx.Redirect(debugUrl)
  393. }
  394. func CloudBrainCommitImage(ctx *context.Context, form auth.CommitImageCloudBrainForm) {
  395. err := cloudbrain.CommitImage(ctx.Cloudbrain.JobID, models.CommitImageParams{
  396. Ip: ctx.Cloudbrain.ContainerIp,
  397. TaskContainerId: ctx.Cloudbrain.ContainerID,
  398. ImageDescription: form.Description,
  399. ImageTag: form.Tag,
  400. })
  401. if err != nil {
  402. log.Error("CommitImage(%s) failed:%v", ctx.Cloudbrain.JobName, err.Error(), ctx.Data["msgID"])
  403. ctx.JSON(200, map[string]string{
  404. "result_code": "-1",
  405. "error_msg": "CommitImage failed",
  406. })
  407. return
  408. }
  409. ctx.JSON(200, map[string]string{
  410. "result_code": "0",
  411. "error_msg": "",
  412. })
  413. }
  414. func CloudBrainStop(ctx *context.Context) {
  415. var jobID = ctx.Params(":jobid")
  416. var resultCode = "0"
  417. var errorMsg = ""
  418. var status = ""
  419. task := ctx.Cloudbrain
  420. for {
  421. if task.Status == string(models.JobStopped) || task.Status == string(models.JobFailed) || task.Status == string(models.JobSucceeded) {
  422. log.Error("the job(%s) has been stopped", task.JobName, ctx.Data["msgID"])
  423. resultCode = "-1"
  424. errorMsg = "system error"
  425. break
  426. }
  427. err := cloudbrain.StopJob(jobID)
  428. if err != nil {
  429. log.Error("StopJob(%s) failed:%v", task.JobName, err, ctx.Data["msgID"])
  430. resultCode = "-1"
  431. errorMsg = "system error"
  432. break
  433. }
  434. task.Status = string(models.JobStopped)
  435. err = models.UpdateJob(task)
  436. if err != nil {
  437. log.Error("UpdateJob(%s) failed:%v", task.JobName, err, ctx.Data["msgID"])
  438. resultCode = "-1"
  439. errorMsg = "system error"
  440. break
  441. }
  442. status = task.Status
  443. break
  444. }
  445. ctx.JSON(200, map[string]string{
  446. "result_code": resultCode,
  447. "error_msg": errorMsg,
  448. "status": status,
  449. "job_id": jobID,
  450. })
  451. }
  452. func StopJobsByUserID(userID int64) {
  453. cloudBrains, err := models.GetCloudbrainsNeededStopByUserID(userID)
  454. if err != nil {
  455. log.Warn("Failed to get cloudBrain info", err)
  456. return
  457. }
  458. StopJobs(cloudBrains)
  459. }
  460. func StopJobsByRepoID(repoID int64) {
  461. cloudBrains, err := models.GetCloudbrainsNeededStopByRepoID(repoID)
  462. if err != nil {
  463. log.Warn("Failed to get cloudBrain info", err)
  464. return
  465. }
  466. StopJobs(cloudBrains)
  467. }
  468. /**
  469. */
  470. func StopJobs(cloudBrains []*models.Cloudbrain) {
  471. for _, taskInfo := range cloudBrains {
  472. if taskInfo.Type == models.TypeCloudBrainOne {
  473. err := retry(3, time.Second*30, func() error {
  474. return cloudbrain.StopJob(taskInfo.JobID)
  475. })
  476. logErrorAndUpdateJobStatus(err, taskInfo)
  477. } else {
  478. if taskInfo.JobType == string(models.JobTypeTrain) {
  479. err := retry(3, time.Second*30, func() error {
  480. _, err := modelarts.StopTrainJob(taskInfo.JobID, strconv.FormatInt(taskInfo.VersionID, 10))
  481. return err
  482. })
  483. logErrorAndUpdateJobStatus(err, taskInfo)
  484. } else {
  485. param := models.NotebookAction{
  486. Action: models.ActionStop,
  487. }
  488. err := retry(3, time.Second*30, func() error {
  489. _, err := modelarts.ManageNotebook(taskInfo.JobID, param)
  490. return err
  491. })
  492. logErrorAndUpdateJobStatus(err, taskInfo)
  493. }
  494. }
  495. }
  496. }
  497. func retry(attempts int, sleep time.Duration, f func() error) (err error) {
  498. for i := 0; i < attempts; i++ {
  499. if i > 0 {
  500. log.Warn("retrying after error:", err)
  501. time.Sleep(sleep)
  502. }
  503. err = f()
  504. if err == nil {
  505. return nil
  506. }
  507. }
  508. return fmt.Errorf("after %d attempts, last error: %s", attempts, err)
  509. }
  510. func logErrorAndUpdateJobStatus(err error, taskInfo *models.Cloudbrain) {
  511. if err != nil {
  512. log.Warn("Failed to stop cloudBrain job:"+taskInfo.JobID, err)
  513. } else {
  514. taskInfo.Status = string(models.JobStopped)
  515. err = models.UpdateJob(taskInfo)
  516. if err != nil {
  517. log.Warn("UpdateJob failed", err)
  518. }
  519. }
  520. }
  521. func CloudBrainDel(ctx *context.Context) {
  522. if err := deleteCloudbrainJob(ctx); err != nil {
  523. log.Error("deleteCloudbrainJob failed: %v", err, ctx.Data["msgID"])
  524. ctx.ServerError(err.Error(), err)
  525. return
  526. }
  527. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=all")
  528. }
  529. func deleteCloudbrainJob(ctx *context.Context) error {
  530. task := ctx.Cloudbrain
  531. if task.Status != string(models.JobStopped) && task.Status != string(models.JobFailed) && task.Status != string(models.JobSucceeded) {
  532. log.Error("the job(%s) has not been stopped", task.JobName, ctx.Data["msgID"])
  533. return errors.New("the job has not been stopped")
  534. }
  535. err := models.DeleteJob(task)
  536. if err != nil {
  537. log.Error("DeleteJob failed: %v", err, ctx.Data["msgID"])
  538. return err
  539. }
  540. deleteJobStorage(task.JobName, models.TypeCloudBrainOne)
  541. return nil
  542. }
  543. func CloudBrainShowModels(ctx *context.Context) {
  544. ctx.Data["PageIsCloudBrain"] = true
  545. jobID := ctx.Params(":jobid")
  546. parentDir := ctx.Query("parentDir")
  547. dirArray := strings.Split(parentDir, "/")
  548. task, err := models.GetCloudbrainByJobID(jobID)
  549. if err != nil {
  550. log.Error("no such job!", ctx.Data["msgID"])
  551. ctx.ServerError("no such job:", err)
  552. return
  553. }
  554. //get dirs
  555. dirs, err := GetModelDirs(task.JobName, parentDir)
  556. if err != nil {
  557. log.Error("GetModelDirs failed:%v", err.Error(), ctx.Data["msgID"])
  558. ctx.ServerError("GetModelDirs failed:", err)
  559. return
  560. }
  561. var fileInfos []storage.FileInfo
  562. err = json.Unmarshal([]byte(dirs), &fileInfos)
  563. if err != nil {
  564. log.Error("json.Unmarshal failed:%v", err.Error(), ctx.Data["msgID"])
  565. ctx.ServerError("json.Unmarshal failed:", err)
  566. return
  567. }
  568. for i, fileInfo := range fileInfos {
  569. temp, _ := time.Parse("2006-01-02 15:04:05", fileInfo.ModTime)
  570. fileInfos[i].ModTime = temp.Local().Format("2006-01-02 15:04:05")
  571. }
  572. sort.Slice(fileInfos, func(i, j int) bool {
  573. return fileInfos[i].ModTime > fileInfos[j].ModTime
  574. })
  575. ctx.Data["Path"] = dirArray
  576. ctx.Data["Dirs"] = fileInfos
  577. ctx.Data["task"] = task
  578. ctx.Data["JobID"] = jobID
  579. ctx.HTML(200, tplCloudBrainShowModels)
  580. }
  581. func GetPublicImages(ctx *context.Context) {
  582. getImages(ctx, cloudbrain.Public)
  583. }
  584. func GetCustomImages(ctx *context.Context) {
  585. getImages(ctx, cloudbrain.Custom)
  586. }
  587. func getImages(ctx *context.Context, imageType string) {
  588. log.Info("Get images begin")
  589. page := ctx.QueryInt("page")
  590. size := ctx.QueryInt("size")
  591. name := ctx.Query("name")
  592. getImagesResult, err := cloudbrain.GetImagesPageable(page, size, imageType, name)
  593. if err != nil {
  594. log.Error("Can not get images:%v", err)
  595. ctx.JSON(http.StatusOK, models.GetImagesPayload{
  596. Count: 0,
  597. TotalPages: 0,
  598. ImageInfo: []*models.ImageInfo{},
  599. })
  600. } else {
  601. ctx.JSON(http.StatusOK, getImagesResult.Payload)
  602. }
  603. log.Info("Get images end")
  604. }
  605. func GetModelDirs(jobName string, parentDir string) (string, error) {
  606. var req string
  607. modelActualPath := storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/")
  608. if parentDir == "" {
  609. req = "baseDir=" + modelActualPath
  610. } else {
  611. req = "baseDir=" + modelActualPath + "&parentDir=" + parentDir
  612. }
  613. return getDirs(req)
  614. }
  615. func CloudBrainDownloadModel(ctx *context.Context) {
  616. parentDir := ctx.Query("parentDir")
  617. fileName := ctx.Query("fileName")
  618. jobName := ctx.Query("jobName")
  619. filePath := "jobs/" + jobName + "/model/" + parentDir
  620. url, err := storage.Attachments.PresignedGetURL(filePath, fileName)
  621. if err != nil {
  622. log.Error("PresignedGetURL failed: %v", err.Error(), ctx.Data["msgID"])
  623. ctx.ServerError("PresignedGetURL", err)
  624. return
  625. }
  626. http.Redirect(ctx.Resp, ctx.Req.Request, url, http.StatusMovedPermanently)
  627. }
  628. func GetRate(ctx *context.Context) {
  629. isObjectDetcionAll := ctx.QueryBool("isObjectDetcionAll")
  630. if isObjectDetcionAll {
  631. ctx.Redirect(setting.BenchmarkServerHost + "?username=admin")
  632. return
  633. }
  634. var jobID = ctx.Params(":jobid")
  635. job, err := models.GetCloudbrainByJobID(jobID)
  636. if err != nil {
  637. ctx.ServerError("GetCloudbrainByJobID failed", err)
  638. return
  639. }
  640. if job.JobType == string(models.JobTypeBenchmark) {
  641. log.Info("url=" + setting.BenchmarkServerHost + "?username=" + ctx.User.Name)
  642. ctx.Redirect(setting.BenchmarkServerHost + "?username=" + ctx.User.Name)
  643. } else if job.JobType == string(models.JobTypeSnn4imagenet) {
  644. ctx.Redirect(setting.Snn4imagenetServerHost)
  645. } else if job.JobType == string(models.JobTypeBrainScore) {
  646. ctx.Redirect(setting.BrainScoreServerHost)
  647. } else {
  648. log.Error("JobType error:%s", job.JobType, ctx.Data["msgID"])
  649. }
  650. }
  651. func downloadCode(repo *models.Repository, codePath string) error {
  652. if err := git.Clone(repo.RepoPath(), codePath, git.CloneRepoOptions{}); err != nil {
  653. log.Error("Failed to clone repository: %s (%v)", repo.FullName(), err)
  654. return err
  655. }
  656. configFile, err := os.OpenFile(codePath+"/.git/config", os.O_RDWR, 0666)
  657. if err != nil {
  658. log.Error("open file(%s) failed:%v", codePath+"/,git/config", err)
  659. return err
  660. }
  661. defer configFile.Close()
  662. pos := int64(0)
  663. reader := bufio.NewReader(configFile)
  664. for {
  665. line, err := reader.ReadString('\n')
  666. if err != nil {
  667. if err == io.EOF {
  668. log.Error("not find the remote-url")
  669. return nil
  670. } else {
  671. log.Error("read error: %v", err)
  672. return err
  673. }
  674. }
  675. if strings.Contains(line, "url") && strings.Contains(line, ".git") {
  676. originUrl := "\turl = " + repo.CloneLink().HTTPS + "\n"
  677. if len(line) > len(originUrl) {
  678. originUrl += strings.Repeat(" ", len(line)-len(originUrl))
  679. }
  680. bytes := []byte(originUrl)
  681. _, err := configFile.WriteAt(bytes, pos)
  682. if err != nil {
  683. log.Error("WriteAt failed:%v", err)
  684. return err
  685. }
  686. break
  687. }
  688. pos += int64(len(line))
  689. }
  690. return nil
  691. }
  692. func downloadRateCode(repo *models.Repository, taskName, rateOwnerName, rateRepoName, codePath, benchmarkCategory, gpuType string) error {
  693. err := os.MkdirAll(codePath, os.ModePerm)
  694. if err != nil {
  695. log.Error("mkdir codePath failed", err.Error())
  696. return err
  697. }
  698. repoExt, err := models.GetRepositoryByOwnerAndName(rateOwnerName, rateRepoName)
  699. if err != nil {
  700. log.Error("GetRepositoryByOwnerAndName(%s) failed", rateRepoName, err.Error())
  701. return err
  702. }
  703. if err := git.Clone(repoExt.RepoPath(), codePath, git.CloneRepoOptions{}); err != nil {
  704. log.Error("Failed to clone repository: %s (%v)", repoExt.FullName(), err)
  705. return err
  706. }
  707. fileName := codePath + cloudbrain.TaskInfoName
  708. f, err := os.OpenFile(fileName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, os.ModePerm)
  709. if err != nil {
  710. log.Error("OpenFile failed", err.Error())
  711. return err
  712. }
  713. defer f.Close()
  714. data, err := json.Marshal(models.TaskInfo{
  715. Username: repo.Owner.Name,
  716. TaskName: taskName,
  717. CodeName: repo.Name,
  718. BenchmarkCategory: strings.Split(benchmarkCategory, ","),
  719. CodeLink: strings.TrimSuffix(repo.CloneLink().HTTPS, ".git"),
  720. GpuType: gpuType,
  721. })
  722. if err != nil {
  723. log.Error("json.Marshal failed", err.Error())
  724. return err
  725. }
  726. _, err = f.Write(data)
  727. if err != nil {
  728. log.Error("WriteString failed", err.Error())
  729. return err
  730. }
  731. return nil
  732. }
  733. func uploadCodeToMinio(codePath, jobName, parentDir string) error {
  734. files, err := readDir(codePath)
  735. if err != nil {
  736. log.Error("readDir(%s) failed: %s", codePath, err.Error())
  737. return err
  738. }
  739. for _, file := range files {
  740. if file.IsDir() {
  741. if err = uploadCodeToMinio(codePath+file.Name()+"/", jobName, parentDir+file.Name()+"/"); err != nil {
  742. log.Error("uploadCodeToMinio(%s) failed: %s", file.Name(), err.Error())
  743. return err
  744. }
  745. } else {
  746. destObject := setting.CBCodePathPrefix + jobName + parentDir + file.Name()
  747. sourceFile := codePath + file.Name()
  748. err = storage.Attachments.UploadObject(destObject, sourceFile)
  749. if err != nil {
  750. log.Error("UploadObject(%s) failed: %s", file.Name(), err.Error())
  751. return err
  752. }
  753. }
  754. }
  755. return nil
  756. }
  757. func mkModelPath(modelPath string) error {
  758. err := os.MkdirAll(modelPath, os.ModePerm)
  759. if err != nil {
  760. log.Error("MkdirAll(%s) failed:%v", modelPath, err)
  761. return err
  762. }
  763. fileName := modelPath + "README"
  764. f, err := os.OpenFile(fileName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, os.ModePerm)
  765. if err != nil {
  766. log.Error("OpenFile failed", err.Error())
  767. return err
  768. }
  769. defer f.Close()
  770. _, err = f.WriteString("You can put the model file into this directory and download it by the web page.")
  771. if err != nil {
  772. log.Error("WriteString failed", err.Error())
  773. return err
  774. }
  775. return nil
  776. }
  777. func deleteJobStorage(jobName string, cloudbrainType int) error {
  778. //delete local
  779. localJobPath := setting.JobPath + jobName
  780. err := os.RemoveAll(localJobPath)
  781. if err != nil {
  782. log.Error("RemoveAll(%s) failed:%v", localJobPath, err)
  783. }
  784. //delete oss
  785. if cloudbrainType == models.TypeCloudBrainOne {
  786. dirPath := setting.CBCodePathPrefix + jobName + "/"
  787. err = storage.Attachments.DeleteDir(dirPath)
  788. if err != nil {
  789. log.Error("DeleteDir(%s) failed:%v", localJobPath, err)
  790. }
  791. } else if cloudbrainType == models.TypeCloudBrainTwo {
  792. //dirPath := setting.CodePathPrefix + jobName + "/"
  793. //err = storage.ObsRemoveObject(setting.Bucket, dirPath)
  794. //if err != nil {
  795. // log.Error("ObsRemoveObject(%s) failed:%v", localJobPath, err)
  796. //}
  797. log.Info("no need to delete")
  798. } else {
  799. log.Error("cloudbrainType(%d) error", cloudbrainType)
  800. }
  801. return nil
  802. }
  803. func SyncCloudbrainStatus() {
  804. cloudBrains, err := models.GetCloudBrainUnStoppedJob()
  805. if err != nil {
  806. log.Error("GetCloudBrainUnStoppedJob failed:", err.Error())
  807. return
  808. }
  809. for _, task := range cloudBrains {
  810. if task.Type == models.TypeCloudBrainOne {
  811. result, err := cloudbrain.GetJob(task.JobID)
  812. if err != nil {
  813. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  814. continue
  815. }
  816. if result != nil {
  817. jobRes, _ := models.ConvertToJobResultPayload(result.Payload)
  818. taskRoles := jobRes.TaskRoles
  819. taskRes, _ := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
  820. task.Status = taskRes.TaskStatuses[0].State
  821. if task.Status != string(models.JobWaiting) {
  822. task.Duration = time.Now().Unix() - taskRes.TaskStatuses[0].StartAt.Unix()
  823. err = models.UpdateJob(task)
  824. if err != nil {
  825. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  826. }
  827. var maxDuration int64
  828. if task.JobType == string(models.JobTypeBenchmark) {
  829. maxDuration = setting.BenchmarkMaxDuration
  830. } else {
  831. maxDuration = setting.MaxDuration
  832. }
  833. if task.Duration >= maxDuration {
  834. log.Info("begin to stop job(%s), because of the duration", task.JobName)
  835. err = cloudbrain.StopJob(task.JobID)
  836. if err != nil {
  837. log.Error("StopJob(%s) failed:%v", task.JobName, err)
  838. continue
  839. }
  840. task.Status = string(models.JobStopped)
  841. err = models.UpdateJob(task)
  842. if err != nil {
  843. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  844. }
  845. }
  846. }
  847. }
  848. } else if task.Type == models.TypeCloudBrainTwo {
  849. if task.JobType == string(models.JobTypeDebug) {
  850. result, err := modelarts.GetJob(task.JobID)
  851. if err != nil {
  852. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  853. continue
  854. }
  855. if result != nil {
  856. task.Status = result.Status
  857. err = models.UpdateJob(task)
  858. if err != nil {
  859. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  860. continue
  861. }
  862. }
  863. } else if task.JobType == string(models.JobTypeTrain) {
  864. result, err := modelarts.GetTrainJob(task.JobID, strconv.FormatInt(task.VersionID, 10))
  865. if err != nil {
  866. log.Error("GetTrainJob(%s) failed:%v", task.JobName, err)
  867. continue
  868. }
  869. if result != nil {
  870. task.Status = modelarts.TransTrainJobStatus(result.IntStatus)
  871. task.Duration = result.Duration
  872. task.TrainJobDuration = result.TrainJobDuration
  873. if result.Duration != 0 {
  874. task.TrainJobDuration = util.AddZero(result.Duration/3600000) + ":" + util.AddZero(result.Duration%3600000/60000) + ":" + util.AddZero(result.Duration%60000/1000)
  875. } else {
  876. task.TrainJobDuration = "00:00:00"
  877. }
  878. err = models.UpdateJob(task)
  879. if err != nil {
  880. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  881. continue
  882. }
  883. }
  884. } else {
  885. log.Error("task.JobType(%s) is error:%s", task.JobName, task.JobType)
  886. }
  887. } else {
  888. log.Error("task.Type(%s) is error:%d", task.JobName, task.Type)
  889. }
  890. }
  891. return
  892. }
  893. func CloudBrainBenchmarkIndex(ctx *context.Context) {
  894. MustEnableCloudbrain(ctx)
  895. repo := ctx.Repo.Repository
  896. page := ctx.QueryInt("page")
  897. if page <= 0 {
  898. page = 1
  899. }
  900. var jobTypes []string
  901. jobTypes = append(jobTypes, string(models.JobTypeBenchmark))
  902. ciTasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  903. ListOptions: models.ListOptions{
  904. Page: page,
  905. PageSize: setting.UI.IssuePagingNum,
  906. },
  907. RepoID: repo.ID,
  908. Type: models.TypeCloudBrainOne,
  909. JobTypes: jobTypes,
  910. })
  911. if err != nil {
  912. ctx.ServerError("Get debugjob faild:", err)
  913. return
  914. }
  915. for i, task := range ciTasks {
  916. ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  917. ciTasks[i].Cloudbrain.ComputeResource = task.ComputeResource
  918. var duration int64
  919. if task.Status == string(models.JobRunning) {
  920. duration = time.Now().Unix() - int64(task.Cloudbrain.CreatedUnix)
  921. } else {
  922. duration = int64(task.Cloudbrain.UpdatedUnix) - int64(task.Cloudbrain.CreatedUnix)
  923. }
  924. ciTasks[i].TrainJobDuration = util.AddZero(duration/3600000) + ":" + util.AddZero(duration%3600000/60000) + ":" + util.AddZero(duration%60000/1000)
  925. }
  926. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  927. ctx.Data["Page"] = pager
  928. ctx.Data["PageIsCloudBrain"] = true
  929. ctx.Data["Tasks"] = ciTasks
  930. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  931. ctx.Data["RepoIsEmpty"] = repo.IsEmpty
  932. ctx.HTML(200, tplCloudBrainBenchmarkIndex)
  933. }
  934. func GetChildTypes(ctx *context.Context) {
  935. benchmarkTypeID := ctx.QueryInt("benchmark_type_id")
  936. re := make(map[string]interface{})
  937. for {
  938. if benchmarkTypes == nil {
  939. if err := json.Unmarshal([]byte(setting.BenchmarkTypes), &benchmarkTypes); err != nil {
  940. log.Error("json.Unmarshal BenchmarkTypes(%s) failed:%v", setting.BenchmarkTypes, err, ctx.Data["MsgID"])
  941. re["errMsg"] = "system error"
  942. break
  943. }
  944. }
  945. var isExist bool
  946. for _, benchmarkType := range benchmarkTypes.BenchmarkType {
  947. if benchmarkTypeID == benchmarkType.Id {
  948. isExist = true
  949. re["child_types"] = benchmarkType.Second
  950. re["result_code"] = "0"
  951. break
  952. }
  953. }
  954. if !isExist {
  955. re["result_code"] = "1"
  956. log.Error("no such benchmark_type_id", ctx.Data["MsgID"])
  957. re["errMsg"] = "system error"
  958. break
  959. }
  960. break
  961. }
  962. ctx.JSON(200, re)
  963. }
  964. func CloudBrainBenchmarkNew(ctx *context.Context) {
  965. err := cloudBrainNewDataPrepare(ctx)
  966. if err != nil {
  967. ctx.ServerError("get new cloudbrain info failed", err)
  968. return
  969. }
  970. ctx.HTML(200, tplCloudBrainBenchmarkNew)
  971. }
  972. func getBenchmarkAttachment(benchmarkTypeID, benchmarkChildTypeID int) (*models.BenchmarkDataset, error) {
  973. var childInfo *models.BenchmarkDataset
  974. if benchmarkTypes == nil {
  975. if err := json.Unmarshal([]byte(setting.BenchmarkTypes), &benchmarkTypes); err != nil {
  976. log.Error("json.Unmarshal BenchmarkTypes(%s) failed:%v", setting.BenchmarkTypes, err)
  977. return childInfo, err
  978. }
  979. }
  980. var isExist bool
  981. for _, benchmarkType := range benchmarkTypes.BenchmarkType {
  982. if benchmarkType.Id == benchmarkTypeID {
  983. for _, childType := range benchmarkType.Second {
  984. if childType.Id == benchmarkChildTypeID {
  985. childInfo = childType
  986. isExist = true
  987. break
  988. }
  989. }
  990. break
  991. }
  992. }
  993. if !isExist {
  994. log.Error("no such benchmark_type_id&benchmark_child_type_id")
  995. return childInfo, errors.New("no such benchmark_type_id&benchmark_child_type_id")
  996. }
  997. return childInfo, nil
  998. }
  999. func getBenchmarkGpuQueue(gpuQueue string) (string, error) {
  1000. queue := ""
  1001. if benchmarkGpuInfos == nil {
  1002. if err := json.Unmarshal([]byte(setting.BenchmarkGpuTypes), &benchmarkGpuInfos); err != nil {
  1003. log.Error("json.Unmarshal BenchmarkGpuTypes(%s) failed:%v", setting.BenchmarkGpuTypes, err)
  1004. return queue, err
  1005. }
  1006. }
  1007. var isExist bool
  1008. for _, gpuInfo := range benchmarkGpuInfos.GpuInfo {
  1009. if gpuQueue == gpuInfo.Queue {
  1010. isExist = true
  1011. queue = gpuQueue
  1012. break
  1013. }
  1014. }
  1015. if !isExist {
  1016. log.Error("no such gpuQueue, %s", gpuQueue)
  1017. return queue, errors.New("no such gpuQueue")
  1018. }
  1019. return queue, nil
  1020. }
  1021. func getBenchmarkResourceSpec(resourceSpecID int) (int, error) {
  1022. var id int
  1023. if benchmarkResourceSpecs == nil {
  1024. if err := json.Unmarshal([]byte(setting.BenchmarkResourceSpecs), &benchmarkResourceSpecs); err != nil {
  1025. log.Error("json.Unmarshal BenchmarkResourceSpecs(%s) failed:%v", setting.BenchmarkResourceSpecs, err)
  1026. return id, err
  1027. }
  1028. }
  1029. var isExist bool
  1030. for _, resourceSpec := range benchmarkResourceSpecs.ResourceSpec {
  1031. if resourceSpecID == resourceSpec.Id {
  1032. isExist = true
  1033. id = resourceSpecID
  1034. break
  1035. }
  1036. }
  1037. if !isExist {
  1038. log.Error("no such resourceSpecID, %d", resourceSpecID)
  1039. return id, errors.New("no such resourceSpec")
  1040. }
  1041. return id, nil
  1042. }
  1043. func CloudBrainBenchmarkCreate(ctx *context.Context, form auth.CreateCloudBrainForm) {
  1044. ctx.Data["PageIsCloudBrain"] = true
  1045. openiJobName := form.JobName
  1046. jobName := util.ConvertToApiJobName(openiJobName)
  1047. image := form.Image
  1048. gpuQueue := form.GpuType
  1049. command := cloudbrain.CommandBenchmark
  1050. codePath := setting.JobPath + jobName + cloudbrain.CodeMountPath
  1051. resourceSpecId := cloudbrain.BenchMarkResourceID
  1052. benchmarkTypeID := form.BenchmarkTypeID
  1053. benchmarkChildTypeID := form.BenchmarkChildTypeID
  1054. if !jobNamePattern.MatchString(jobName) {
  1055. cloudBrainNewDataPrepare(ctx)
  1056. ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_jobname_err"), tplCloudBrainBenchmarkNew, &form)
  1057. return
  1058. }
  1059. childInfo, err := getBenchmarkAttachment(benchmarkTypeID, benchmarkChildTypeID)
  1060. if err != nil {
  1061. log.Error("getBenchmarkAttachment failed:%v", err, ctx.Data["MsgID"])
  1062. cloudBrainNewDataPrepare(ctx)
  1063. ctx.RenderWithErr("benchmark type error", tplCloudBrainBenchmarkNew, &form)
  1064. return
  1065. }
  1066. _, err = getBenchmarkGpuQueue(gpuQueue)
  1067. if err != nil {
  1068. log.Error("getBenchmarkGpuQueue failed:%v", err, ctx.Data["MsgID"])
  1069. cloudBrainNewDataPrepare(ctx)
  1070. ctx.RenderWithErr("gpu queue error", tplCloudBrainBenchmarkNew, &form)
  1071. return
  1072. }
  1073. _, err = getBenchmarkResourceSpec(resourceSpecId)
  1074. if err != nil {
  1075. log.Error("getBenchmarkResourceSpec failed:%v", err, ctx.Data["MsgID"])
  1076. cloudBrainNewDataPrepare(ctx)
  1077. ctx.RenderWithErr("resource spec error", tplCloudBrainBenchmarkNew, &form)
  1078. return
  1079. }
  1080. count, err := models.GetCloudbrainCountByUserID(ctx.User.ID, string(models.JobTypeBenchmark))
  1081. if err != nil {
  1082. log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])
  1083. cloudBrainNewDataPrepare(ctx)
  1084. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1085. return
  1086. } else {
  1087. if count >= 1 {
  1088. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  1089. cloudBrainNewDataPrepare(ctx)
  1090. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplCloudBrainBenchmarkNew, &form)
  1091. return
  1092. }
  1093. }
  1094. _, err = models.GetCloudbrainByName(jobName)
  1095. if err == nil {
  1096. log.Error("the job name did already exist", ctx.Data["MsgID"])
  1097. cloudBrainNewDataPrepare(ctx)
  1098. ctx.RenderWithErr("the job name did already exist", tplCloudBrainBenchmarkNew, &form)
  1099. return
  1100. } else {
  1101. if !models.IsErrJobNotExist(err) {
  1102. log.Error("GetCloudbrainByName failed, %v", err, ctx.Data["MsgID"])
  1103. cloudBrainNewDataPrepare(ctx)
  1104. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1105. return
  1106. }
  1107. }
  1108. repo := ctx.Repo.Repository
  1109. os.RemoveAll(codePath)
  1110. if err := downloadCode(repo, codePath); err != nil {
  1111. log.Error("downloadCode failed, %v", err, ctx.Data["MsgID"])
  1112. cloudBrainNewDataPrepare(ctx)
  1113. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1114. return
  1115. }
  1116. if _, err := os.Stat(codePath + "/train.py"); err != nil {
  1117. if os.IsNotExist(err) {
  1118. // file does not exist
  1119. log.Error("train.py does not exist, %v", err, ctx.Data["MsgID"])
  1120. cloudBrainNewDataPrepare(ctx)
  1121. ctx.RenderWithErr("train.py does not exist", tplCloudBrainBenchmarkNew, &form)
  1122. } else {
  1123. log.Error("Stat failed, %v", err, ctx.Data["MsgID"])
  1124. cloudBrainNewDataPrepare(ctx)
  1125. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1126. }
  1127. return
  1128. } else if _, err := os.Stat(codePath + "/test.py"); err != nil {
  1129. if os.IsNotExist(err) {
  1130. // file does not exist
  1131. log.Error("test.py does not exist, %v", err, ctx.Data["MsgID"])
  1132. cloudBrainNewDataPrepare(ctx)
  1133. ctx.RenderWithErr("test.py does not exist", tplCloudBrainBenchmarkNew, &form)
  1134. } else {
  1135. log.Error("Stat failed, %v", err, ctx.Data["MsgID"])
  1136. cloudBrainNewDataPrepare(ctx)
  1137. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1138. }
  1139. return
  1140. }
  1141. if err := uploadCodeToMinio(codePath+"/", jobName, cloudbrain.CodeMountPath+"/"); err != nil {
  1142. log.Error("uploadCodeToMinio failed, %v", err, ctx.Data["MsgID"])
  1143. cloudBrainNewDataPrepare(ctx)
  1144. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1145. return
  1146. }
  1147. benchmarkPath := setting.JobPath + jobName + cloudbrain.BenchMarkMountPath
  1148. var gpuType string
  1149. for _, gpuInfo := range gpuInfos.GpuInfo {
  1150. if gpuInfo.Queue == gpuQueue {
  1151. gpuType = gpuInfo.Value
  1152. }
  1153. }
  1154. if err := downloadRateCode(repo, jobName, childInfo.Owner, childInfo.RepoName, benchmarkPath, form.BenchmarkCategory, gpuType); err != nil {
  1155. log.Error("downloadRateCode failed, %v", err, ctx.Data["MsgID"])
  1156. //cloudBrainNewDataPrepare(ctx)
  1157. //ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1158. //return
  1159. }
  1160. if err := uploadCodeToMinio(benchmarkPath+"/", jobName, cloudbrain.BenchMarkMountPath+"/"); err != nil {
  1161. log.Error("uploadCodeToMinio failed, %v", err, ctx.Data["MsgID"])
  1162. //cloudBrainNewDataPrepare(ctx)
  1163. //ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1164. //return
  1165. }
  1166. err = cloudbrain.GenerateTask(ctx, openiJobName, jobName, image, command, childInfo.Attachment, storage.GetMinioPath(jobName, cloudbrain.CodeMountPath+"/"),
  1167. storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/"),
  1168. storage.GetMinioPath(jobName, cloudbrain.BenchMarkMountPath+"/"), storage.GetMinioPath(jobName, cloudbrain.Snn4imagenetMountPath+"/"),
  1169. storage.GetMinioPath(jobName, cloudbrain.BrainScoreMountPath+"/"), string(models.JobTypeBenchmark), gpuQueue, form.Description,
  1170. benchmarkTypeID, benchmarkChildTypeID, resourceSpecId)
  1171. if err != nil {
  1172. cloudBrainNewDataPrepare(ctx)
  1173. ctx.RenderWithErr(err.Error(), tplCloudBrainBenchmarkNew, &form)
  1174. return
  1175. }
  1176. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/cloudbrain/benchmark")
  1177. }
  1178. func BenchmarkDel(ctx *context.Context) {
  1179. if err := deleteCloudbrainJob(ctx); err != nil {
  1180. log.Error("deleteCloudbrainJob failed: %v", err, ctx.Data["msgID"])
  1181. ctx.ServerError(err.Error(), err)
  1182. return
  1183. }
  1184. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/cloudbrain/benchmark")
  1185. }