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
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
4 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
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
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
5 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
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
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
3 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
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
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
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
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. 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. }
  293. }
  294. cloudBrainShow(ctx, tplCloudBrainBenchmarkShow)
  295. }
  296. func CloudBrainShow(ctx *context.Context) {
  297. cloudBrainShow(ctx, tplCloudBrainShow)
  298. }
  299. func cloudBrainShow(ctx *context.Context, tpName base.TplName) {
  300. ctx.Data["PageIsCloudBrain"] = true
  301. var jobName = ctx.Params(":jobname")
  302. task, err := models.GetCloudbrainByName(jobName)
  303. if err != nil {
  304. ctx.Data["error"] = err.Error()
  305. }
  306. result, err := cloudbrain.GetJob(task.JobID)
  307. if err != nil {
  308. ctx.Data["error"] = err.Error()
  309. }
  310. if result != nil {
  311. jobRes, _ := models.ConvertToJobResultPayload(result.Payload)
  312. jobRes.Resource.Memory = strings.ReplaceAll(jobRes.Resource.Memory, "Mi", "MB")
  313. spec := "GPU数:" + strconv.Itoa(jobRes.Resource.NvidiaComGpu) + ",CPU数:" + strconv.Itoa(jobRes.Resource.CPU) + ",内存(MB):" + jobRes.Resource.Memory
  314. ctx.Data["resource_spec"] = spec
  315. taskRoles := jobRes.TaskRoles
  316. if jobRes.JobStatus.State != string(models.JobFailed) {
  317. taskRes, _ := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
  318. ctx.Data["taskRes"] = taskRes
  319. task.Status = taskRes.TaskStatuses[0].State
  320. task.ContainerID = taskRes.TaskStatuses[0].ContainerID
  321. task.ContainerIp = taskRes.TaskStatuses[0].ContainerIP
  322. err = models.UpdateJob(task)
  323. if err != nil {
  324. ctx.Data["error"] = err.Error()
  325. }
  326. } else {
  327. task.Status = jobRes.JobStatus.State
  328. taskRes := models.TaskPod{TaskStatuses: []models.TaskStatuses{
  329. {
  330. State: jobRes.JobStatus.State,
  331. },
  332. }}
  333. ctx.Data["taskRes"] = taskRes
  334. jobRes.JobStatus.StartTime = time.Unix(int64(task.CreatedUnix), 0).Format("2006-01-02 15:04:05")
  335. jobRes.JobStatus.EndTime = time.Unix(int64(task.UpdatedUnix), 0).Format("2006-01-02 15:04:05")
  336. }
  337. ctx.Data["result"] = jobRes
  338. } else {
  339. log.Info("error:" + err.Error())
  340. }
  341. user, err := models.GetUserByID(task.UserID)
  342. if err == nil {
  343. task.User = user
  344. }
  345. var duration int64
  346. if task.Status == string(models.JobRunning) {
  347. duration = time.Now().Unix() - int64(task.CreatedUnix)
  348. } else {
  349. duration = int64(task.UpdatedUnix) - int64(task.CreatedUnix)
  350. }
  351. if benchmarkTypes != nil {
  352. for _, benchmarkType := range benchmarkTypes.BenchmarkType {
  353. if task.BenchmarkTypeID == benchmarkType.Id {
  354. ctx.Data["BenchmarkTypeName"] = benchmarkType.First
  355. for _, benchmarkChildType := range benchmarkType.Second {
  356. if task.BenchmarkChildTypeID == benchmarkChildType.Id {
  357. ctx.Data["BenchmarkChildTypeName"] = benchmarkChildType.Value
  358. break
  359. }
  360. }
  361. break
  362. }
  363. }
  364. }
  365. ctx.Data["duration"] = util.AddZero(duration/3600000) + ":" + util.AddZero(duration%3600000/60000) + ":" + util.AddZero(duration%60000/1000)
  366. ctx.Data["task"] = task
  367. // ctx.Data["jobID"] = task.JobID
  368. ctx.Data["jobName"] = task.JobName
  369. version_list_task := make([]*models.Cloudbrain, 0)
  370. version_list_task = append(version_list_task, task)
  371. ctx.Data["version_list_task"] = version_list_task
  372. ctx.HTML(200, tpName)
  373. }
  374. func CloudBrainDebug(ctx *context.Context) {
  375. debugUrl := setting.DebugServerHost + "jpylab_" + ctx.Cloudbrain.JobID + "_" + ctx.Cloudbrain.SubTaskName
  376. ctx.Redirect(debugUrl)
  377. }
  378. func CloudBrainCommitImage(ctx *context.Context, form auth.CommitImageCloudBrainForm) {
  379. err := cloudbrain.CommitImage(ctx.Cloudbrain.JobID, models.CommitImageParams{
  380. Ip: ctx.Cloudbrain.ContainerIp,
  381. TaskContainerId: ctx.Cloudbrain.ContainerID,
  382. ImageDescription: form.Description,
  383. ImageTag: form.Tag,
  384. })
  385. if err != nil {
  386. log.Error("CommitImage(%s) failed:%v", ctx.Cloudbrain.JobName, err.Error(), ctx.Data["msgID"])
  387. ctx.JSON(200, map[string]string{
  388. "result_code": "-1",
  389. "error_msg": "CommitImage failed",
  390. })
  391. return
  392. }
  393. ctx.JSON(200, map[string]string{
  394. "result_code": "0",
  395. "error_msg": "",
  396. })
  397. }
  398. func CloudBrainStop(ctx *context.Context) {
  399. var jobID = ctx.Params(":jobid")
  400. var resultCode = "0"
  401. var errorMsg = ""
  402. var status = ""
  403. task := ctx.Cloudbrain
  404. for {
  405. if task.Status == string(models.JobStopped) || task.Status == string(models.JobFailed) || task.Status == string(models.JobSucceeded) {
  406. log.Error("the job(%s) has been stopped", task.JobName, ctx.Data["msgID"])
  407. resultCode = "-1"
  408. errorMsg = "system error"
  409. break
  410. }
  411. err := cloudbrain.StopJob(jobID)
  412. if err != nil {
  413. log.Error("StopJob(%s) failed:%v", task.JobName, err, ctx.Data["msgID"])
  414. resultCode = "-1"
  415. errorMsg = "system error"
  416. break
  417. }
  418. task.Status = string(models.JobStopped)
  419. err = models.UpdateJob(task)
  420. if err != nil {
  421. log.Error("UpdateJob(%s) failed:%v", task.JobName, err, ctx.Data["msgID"])
  422. resultCode = "-1"
  423. errorMsg = "system error"
  424. break
  425. }
  426. status = task.Status
  427. break
  428. }
  429. ctx.JSON(200, map[string]string{
  430. "result_code": resultCode,
  431. "error_msg": errorMsg,
  432. "status": status,
  433. "job_id": jobID,
  434. })
  435. }
  436. func StopJobsByUserID(userID int64) {
  437. cloudBrains, err := models.GetCloudbrainsNeededStopByUserID(userID)
  438. if err != nil {
  439. log.Warn("Failed to get cloudBrain info", err)
  440. return
  441. }
  442. StopJobs(cloudBrains)
  443. }
  444. func StopJobsByRepoID(repoID int64) {
  445. cloudBrains, err := models.GetCloudbrainsNeededStopByRepoID(repoID)
  446. if err != nil {
  447. log.Warn("Failed to get cloudBrain info", err)
  448. return
  449. }
  450. StopJobs(cloudBrains)
  451. }
  452. /**
  453. */
  454. func StopJobs(cloudBrains []*models.Cloudbrain) {
  455. for _, taskInfo := range cloudBrains {
  456. if taskInfo.Type == models.TypeCloudBrainOne {
  457. err := retry(3, time.Second*30, func() error {
  458. return cloudbrain.StopJob(taskInfo.JobID)
  459. })
  460. logErrorAndUpdateJobStatus(err, taskInfo)
  461. } else {
  462. if taskInfo.JobType == string(models.JobTypeTrain) {
  463. err := retry(3, time.Second*30, func() error {
  464. _, err := modelarts.StopTrainJob(taskInfo.JobID, strconv.FormatInt(taskInfo.VersionID, 10))
  465. return err
  466. })
  467. logErrorAndUpdateJobStatus(err, taskInfo)
  468. } else {
  469. param := models.NotebookAction{
  470. Action: models.ActionStop,
  471. }
  472. err := retry(3, time.Second*30, func() error {
  473. _, err := modelarts.ManageNotebook(taskInfo.JobID, param)
  474. return err
  475. })
  476. logErrorAndUpdateJobStatus(err, taskInfo)
  477. }
  478. }
  479. }
  480. }
  481. func retry(attempts int, sleep time.Duration, f func() error) (err error) {
  482. for i := 0; i < attempts; i++ {
  483. if i > 0 {
  484. log.Warn("retrying after error:", err)
  485. time.Sleep(sleep)
  486. }
  487. err = f()
  488. if err == nil {
  489. return nil
  490. }
  491. }
  492. return fmt.Errorf("after %d attempts, last error: %s", attempts, err)
  493. }
  494. func logErrorAndUpdateJobStatus(err error, taskInfo *models.Cloudbrain) {
  495. if err != nil {
  496. log.Warn("Failed to stop cloudBrain job:"+taskInfo.JobID, err)
  497. } else {
  498. taskInfo.Status = string(models.JobStopped)
  499. err = models.UpdateJob(taskInfo)
  500. if err != nil {
  501. log.Warn("UpdateJob failed", err)
  502. }
  503. }
  504. }
  505. func CloudBrainDel(ctx *context.Context) {
  506. var listType = ctx.Query("debugListType")
  507. if err := deleteCloudbrainJob(ctx); err != nil {
  508. log.Error("deleteCloudbrainJob failed: %v", err, ctx.Data["msgID"])
  509. ctx.ServerError(err.Error(), err)
  510. return
  511. }
  512. var isAdminPage = ctx.Query("isadminpage")
  513. if ctx.IsUserSiteAdmin() && isAdminPage == "true" {
  514. ctx.Redirect(setting.AppSubURL + "/admin" + "/cloudbrains")
  515. } else {
  516. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=" + listType)
  517. }
  518. }
  519. func deleteCloudbrainJob(ctx *context.Context) error {
  520. task := ctx.Cloudbrain
  521. if task.Status != string(models.JobStopped) && task.Status != string(models.JobFailed) && task.Status != string(models.JobSucceeded) {
  522. log.Error("the job(%s) has not been stopped", task.JobName, ctx.Data["msgID"])
  523. return errors.New("the job has not been stopped")
  524. }
  525. err := models.DeleteJob(task)
  526. if err != nil {
  527. log.Error("DeleteJob failed: %v", err, ctx.Data["msgID"])
  528. return err
  529. }
  530. deleteJobStorage(task.JobName, models.TypeCloudBrainOne)
  531. return nil
  532. }
  533. func CloudBrainShowModels(ctx *context.Context) {
  534. ctx.Data["PageIsCloudBrain"] = true
  535. jobID := ctx.Params(":jobid")
  536. parentDir := ctx.Query("parentDir")
  537. dirArray := strings.Split(parentDir, "/")
  538. task, err := models.GetCloudbrainByJobID(jobID)
  539. if err != nil {
  540. log.Error("no such job!", ctx.Data["msgID"])
  541. ctx.ServerError("no such job:", err)
  542. return
  543. }
  544. //get dirs
  545. dirs, err := GetModelDirs(task.JobName, parentDir)
  546. if err != nil {
  547. log.Error("GetModelDirs failed:%v", err.Error(), ctx.Data["msgID"])
  548. ctx.ServerError("GetModelDirs failed:", err)
  549. return
  550. }
  551. var fileInfos []storage.FileInfo
  552. err = json.Unmarshal([]byte(dirs), &fileInfos)
  553. if err != nil {
  554. log.Error("json.Unmarshal failed:%v", err.Error(), ctx.Data["msgID"])
  555. ctx.ServerError("json.Unmarshal failed:", err)
  556. return
  557. }
  558. for i, fileInfo := range fileInfos {
  559. temp, _ := time.Parse("2006-01-02 15:04:05", fileInfo.ModTime)
  560. fileInfos[i].ModTime = temp.Local().Format("2006-01-02 15:04:05")
  561. }
  562. sort.Slice(fileInfos, func(i, j int) bool {
  563. return fileInfos[i].ModTime > fileInfos[j].ModTime
  564. })
  565. ctx.Data["Path"] = dirArray
  566. ctx.Data["Dirs"] = fileInfos
  567. ctx.Data["task"] = task
  568. ctx.Data["JobID"] = jobID
  569. ctx.HTML(200, tplCloudBrainShowModels)
  570. }
  571. func GetPublicImages(ctx *context.Context) {
  572. getImages(ctx, cloudbrain.Public)
  573. }
  574. func GetCustomImages(ctx *context.Context) {
  575. getImages(ctx, cloudbrain.Custom)
  576. }
  577. func getImages(ctx *context.Context, imageType string) {
  578. log.Info("Get images begin")
  579. page := ctx.QueryInt("page")
  580. size := ctx.QueryInt("size")
  581. name := ctx.Query("name")
  582. getImagesResult, err := cloudbrain.GetImagesPageable(page, size, imageType, name)
  583. if err != nil {
  584. log.Error("Can not get images:%v", err)
  585. ctx.JSON(http.StatusOK, models.GetImagesPayload{
  586. Count: 0,
  587. TotalPages: 0,
  588. ImageInfo: []*models.ImageInfo{},
  589. })
  590. } else {
  591. ctx.JSON(http.StatusOK, getImagesResult.Payload)
  592. }
  593. log.Info("Get images end")
  594. }
  595. func GetModelDirs(jobName string, parentDir string) (string, error) {
  596. var req string
  597. modelActualPath := storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/")
  598. if parentDir == "" {
  599. req = "baseDir=" + modelActualPath
  600. } else {
  601. req = "baseDir=" + modelActualPath + "&parentDir=" + parentDir
  602. }
  603. return getDirs(req)
  604. }
  605. func CloudBrainDownloadModel(ctx *context.Context) {
  606. parentDir := ctx.Query("parentDir")
  607. fileName := ctx.Query("fileName")
  608. jobName := ctx.Query("jobName")
  609. filePath := "jobs/" + jobName + "/model/" + parentDir
  610. url, err := storage.Attachments.PresignedGetURL(filePath, fileName)
  611. if err != nil {
  612. log.Error("PresignedGetURL failed: %v", err.Error(), ctx.Data["msgID"])
  613. ctx.ServerError("PresignedGetURL", err)
  614. return
  615. }
  616. http.Redirect(ctx.Resp, ctx.Req.Request, url, http.StatusMovedPermanently)
  617. }
  618. func GetRate(ctx *context.Context) {
  619. isObjectDetcionAll := ctx.QueryBool("isObjectDetcionAll")
  620. if isObjectDetcionAll {
  621. ctx.Redirect(setting.BenchmarkServerHost + "?username=admin")
  622. return
  623. }
  624. var jobID = ctx.Params(":jobid")
  625. job, err := models.GetCloudbrainByJobID(jobID)
  626. if err != nil {
  627. ctx.ServerError("GetCloudbrainByJobID failed", err)
  628. return
  629. }
  630. if job.JobType == string(models.JobTypeBenchmark) {
  631. log.Info("url=" + setting.BenchmarkServerHost + "?username=" + ctx.User.Name)
  632. ctx.Redirect(setting.BenchmarkServerHost + "?username=" + ctx.User.Name)
  633. } else if job.JobType == string(models.JobTypeSnn4imagenet) {
  634. ctx.Redirect(setting.Snn4imagenetServerHost)
  635. } else if job.JobType == string(models.JobTypeBrainScore) {
  636. ctx.Redirect(setting.BrainScoreServerHost)
  637. } else {
  638. log.Error("JobType error:%s", job.JobType, ctx.Data["msgID"])
  639. }
  640. }
  641. func downloadCode(repo *models.Repository, codePath string) error {
  642. if err := git.Clone(repo.RepoPath(), codePath, git.CloneRepoOptions{}); err != nil {
  643. log.Error("Failed to clone repository: %s (%v)", repo.FullName(), err)
  644. return err
  645. }
  646. configFile, err := os.OpenFile(codePath+"/.git/config", os.O_RDWR, 0666)
  647. if err != nil {
  648. log.Error("open file(%s) failed:%v", codePath+"/,git/config", err)
  649. return err
  650. }
  651. defer configFile.Close()
  652. pos := int64(0)
  653. reader := bufio.NewReader(configFile)
  654. for {
  655. line, err := reader.ReadString('\n')
  656. if err != nil {
  657. if err == io.EOF {
  658. log.Error("not find the remote-url")
  659. return nil
  660. } else {
  661. log.Error("read error: %v", err)
  662. return err
  663. }
  664. }
  665. if strings.Contains(line, "url") && strings.Contains(line, ".git") {
  666. originUrl := "\turl = " + repo.CloneLink().HTTPS + "\n"
  667. if len(line) > len(originUrl) {
  668. originUrl += strings.Repeat(" ", len(line)-len(originUrl))
  669. }
  670. bytes := []byte(originUrl)
  671. _, err := configFile.WriteAt(bytes, pos)
  672. if err != nil {
  673. log.Error("WriteAt failed:%v", err)
  674. return err
  675. }
  676. break
  677. }
  678. pos += int64(len(line))
  679. }
  680. return nil
  681. }
  682. func downloadRateCode(repo *models.Repository, taskName, rateOwnerName, rateRepoName, codePath, benchmarkCategory, gpuType string) error {
  683. err := os.MkdirAll(codePath, os.ModePerm)
  684. if err != nil {
  685. log.Error("mkdir codePath failed", err.Error())
  686. return err
  687. }
  688. repoExt, err := models.GetRepositoryByOwnerAndName(rateOwnerName, rateRepoName)
  689. if err != nil {
  690. log.Error("GetRepositoryByOwnerAndName(%s) failed", rateRepoName, err.Error())
  691. return err
  692. }
  693. if err := git.Clone(repoExt.RepoPath(), codePath, git.CloneRepoOptions{}); err != nil {
  694. log.Error("Failed to clone repository: %s (%v)", repoExt.FullName(), err)
  695. return err
  696. }
  697. fileName := codePath + cloudbrain.TaskInfoName
  698. f, err := os.OpenFile(fileName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, os.ModePerm)
  699. if err != nil {
  700. log.Error("OpenFile failed", err.Error())
  701. return err
  702. }
  703. defer f.Close()
  704. data, err := json.Marshal(models.TaskInfo{
  705. Username: repo.Owner.Name,
  706. TaskName: taskName,
  707. CodeName: repo.Name,
  708. BenchmarkCategory: strings.Split(benchmarkCategory, ","),
  709. CodeLink: strings.TrimSuffix(repo.CloneLink().HTTPS, ".git"),
  710. GpuType: gpuType,
  711. })
  712. if err != nil {
  713. log.Error("json.Marshal failed", err.Error())
  714. return err
  715. }
  716. _, err = f.Write(data)
  717. if err != nil {
  718. log.Error("WriteString failed", err.Error())
  719. return err
  720. }
  721. return nil
  722. }
  723. func uploadCodeToMinio(codePath, jobName, parentDir string) error {
  724. files, err := readDir(codePath)
  725. if err != nil {
  726. log.Error("readDir(%s) failed: %s", codePath, err.Error())
  727. return err
  728. }
  729. for _, file := range files {
  730. if file.IsDir() {
  731. if err = uploadCodeToMinio(codePath+file.Name()+"/", jobName, parentDir+file.Name()+"/"); err != nil {
  732. log.Error("uploadCodeToMinio(%s) failed: %s", file.Name(), err.Error())
  733. return err
  734. }
  735. } else {
  736. destObject := setting.CBCodePathPrefix + jobName + parentDir + file.Name()
  737. sourceFile := codePath + file.Name()
  738. err = storage.Attachments.UploadObject(destObject, sourceFile)
  739. if err != nil {
  740. log.Error("UploadObject(%s) failed: %s", file.Name(), err.Error())
  741. return err
  742. }
  743. }
  744. }
  745. return nil
  746. }
  747. func mkModelPath(modelPath string) error {
  748. err := os.MkdirAll(modelPath, os.ModePerm)
  749. if err != nil {
  750. log.Error("MkdirAll(%s) failed:%v", modelPath, err)
  751. return err
  752. }
  753. fileName := modelPath + "README"
  754. f, err := os.OpenFile(fileName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, os.ModePerm)
  755. if err != nil {
  756. log.Error("OpenFile failed", err.Error())
  757. return err
  758. }
  759. defer f.Close()
  760. _, err = f.WriteString("You can put the model file into this directory and download it by the web page.")
  761. if err != nil {
  762. log.Error("WriteString failed", err.Error())
  763. return err
  764. }
  765. return nil
  766. }
  767. func deleteJobStorage(jobName string, cloudbrainType int) error {
  768. //delete local
  769. localJobPath := setting.JobPath + jobName
  770. err := os.RemoveAll(localJobPath)
  771. if err != nil {
  772. log.Error("RemoveAll(%s) failed:%v", localJobPath, err)
  773. }
  774. //delete oss
  775. if cloudbrainType == models.TypeCloudBrainOne {
  776. dirPath := setting.CBCodePathPrefix + jobName + "/"
  777. err = storage.Attachments.DeleteDir(dirPath)
  778. if err != nil {
  779. log.Error("DeleteDir(%s) failed:%v", localJobPath, err)
  780. }
  781. } else if cloudbrainType == models.TypeCloudBrainTwo {
  782. //dirPath := setting.CodePathPrefix + jobName + "/"
  783. //err = storage.ObsRemoveObject(setting.Bucket, dirPath)
  784. //if err != nil {
  785. // log.Error("ObsRemoveObject(%s) failed:%v", localJobPath, err)
  786. //}
  787. log.Info("no need to delete")
  788. } else {
  789. log.Error("cloudbrainType(%d) error", cloudbrainType)
  790. }
  791. return nil
  792. }
  793. func SyncCloudbrainStatus() {
  794. cloudBrains, err := models.GetCloudBrainUnStoppedJob()
  795. if err != nil {
  796. log.Error("GetCloudBrainUnStoppedJob failed:", err.Error())
  797. return
  798. }
  799. for _, task := range cloudBrains {
  800. if task.Type == models.TypeCloudBrainOne {
  801. result, err := cloudbrain.GetJob(task.JobID)
  802. if err != nil {
  803. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  804. continue
  805. }
  806. if result != nil {
  807. jobRes, _ := models.ConvertToJobResultPayload(result.Payload)
  808. taskRoles := jobRes.TaskRoles
  809. taskRes, _ := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
  810. task.Status = taskRes.TaskStatuses[0].State
  811. if task.Status != string(models.JobWaiting) {
  812. task.Duration = time.Now().Unix() - taskRes.TaskStatuses[0].StartAt.Unix()
  813. err = models.UpdateJob(task)
  814. if err != nil {
  815. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  816. }
  817. var maxDuration int64
  818. if task.JobType == string(models.JobTypeBenchmark) {
  819. maxDuration = setting.BenchmarkMaxDuration
  820. } else {
  821. maxDuration = setting.MaxDuration
  822. }
  823. if task.Duration >= maxDuration {
  824. log.Info("begin to stop job(%s), because of the duration", task.JobName)
  825. err = cloudbrain.StopJob(task.JobID)
  826. if err != nil {
  827. log.Error("StopJob(%s) failed:%v", task.JobName, err)
  828. continue
  829. }
  830. task.Status = string(models.JobStopped)
  831. err = models.UpdateJob(task)
  832. if err != nil {
  833. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  834. }
  835. }
  836. }
  837. }
  838. } else if task.Type == models.TypeCloudBrainTwo {
  839. if task.JobType == string(models.JobTypeDebug) {
  840. //result, err := modelarts.GetJob(task.JobID)
  841. result, err := modelarts.GetNotebook2(task.JobID)
  842. if err != nil {
  843. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  844. continue
  845. }
  846. if result != nil {
  847. task.Status = result.Status
  848. err = models.UpdateJob(task)
  849. if err != nil {
  850. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  851. continue
  852. }
  853. }
  854. } else if task.JobType == string(models.JobTypeTrain) {
  855. result, err := modelarts.GetTrainJob(task.JobID, strconv.FormatInt(task.VersionID, 10))
  856. if err != nil {
  857. log.Error("GetTrainJob(%s) failed:%v", task.JobName, err)
  858. continue
  859. }
  860. if result != nil {
  861. task.Status = modelarts.TransTrainJobStatus(result.IntStatus)
  862. task.Duration = result.Duration
  863. task.TrainJobDuration = result.TrainJobDuration
  864. if result.Duration != 0 {
  865. task.TrainJobDuration = util.AddZero(result.Duration/3600000) + ":" + util.AddZero(result.Duration%3600000/60000) + ":" + util.AddZero(result.Duration%60000/1000)
  866. } else {
  867. task.TrainJobDuration = "00:00:00"
  868. }
  869. err = models.UpdateJob(task)
  870. if err != nil {
  871. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  872. continue
  873. }
  874. }
  875. } else {
  876. log.Error("task.JobType(%s) is error:%s", task.JobName, task.JobType)
  877. }
  878. } else {
  879. log.Error("task.Type(%s) is error:%d", task.JobName, task.Type)
  880. }
  881. }
  882. return
  883. }
  884. func CloudBrainBenchmarkIndex(ctx *context.Context) {
  885. MustEnableCloudbrain(ctx)
  886. repo := ctx.Repo.Repository
  887. page := ctx.QueryInt("page")
  888. if page <= 0 {
  889. page = 1
  890. }
  891. var jobTypes []string
  892. jobTypes = append(jobTypes, string(models.JobTypeBenchmark))
  893. ciTasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  894. ListOptions: models.ListOptions{
  895. Page: page,
  896. PageSize: setting.UI.IssuePagingNum,
  897. },
  898. RepoID: repo.ID,
  899. Type: models.TypeCloudBrainOne,
  900. JobTypes: jobTypes,
  901. })
  902. if err != nil {
  903. ctx.ServerError("Get debugjob faild:", err)
  904. return
  905. }
  906. for i, task := range ciTasks {
  907. ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  908. ciTasks[i].Cloudbrain.ComputeResource = task.ComputeResource
  909. var duration int64
  910. if task.Status == string(models.JobRunning) {
  911. duration = time.Now().Unix() - int64(task.Cloudbrain.CreatedUnix)
  912. } else {
  913. duration = int64(task.Cloudbrain.UpdatedUnix) - int64(task.Cloudbrain.CreatedUnix)
  914. }
  915. ciTasks[i].TrainJobDuration = util.AddZero(duration/3600000) + ":" + util.AddZero(duration%3600000/60000) + ":" + util.AddZero(duration%60000/1000)
  916. }
  917. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  918. ctx.Data["Page"] = pager
  919. ctx.Data["PageIsCloudBrain"] = true
  920. ctx.Data["Tasks"] = ciTasks
  921. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  922. ctx.Data["RepoIsEmpty"] = repo.IsEmpty
  923. ctx.HTML(200, tplCloudBrainBenchmarkIndex)
  924. }
  925. func GetChildTypes(ctx *context.Context) {
  926. benchmarkTypeID := ctx.QueryInt("benchmark_type_id")
  927. re := make(map[string]interface{})
  928. for {
  929. if benchmarkTypes == nil {
  930. if err := json.Unmarshal([]byte(setting.BenchmarkTypes), &benchmarkTypes); err != nil {
  931. log.Error("json.Unmarshal BenchmarkTypes(%s) failed:%v", setting.BenchmarkTypes, err, ctx.Data["MsgID"])
  932. re["errMsg"] = "system error"
  933. break
  934. }
  935. }
  936. var isExist bool
  937. for _, benchmarkType := range benchmarkTypes.BenchmarkType {
  938. if benchmarkTypeID == benchmarkType.Id {
  939. isExist = true
  940. re["child_types"] = benchmarkType.Second
  941. re["result_code"] = "0"
  942. break
  943. }
  944. }
  945. if !isExist {
  946. re["result_code"] = "1"
  947. log.Error("no such benchmark_type_id", ctx.Data["MsgID"])
  948. re["errMsg"] = "system error"
  949. break
  950. }
  951. break
  952. }
  953. ctx.JSON(200, re)
  954. }
  955. func CloudBrainBenchmarkNew(ctx *context.Context) {
  956. ctx.Data["description"] = ""
  957. ctx.Data["benchmarkTypeID"] = -1
  958. ctx.Data["benchmark_child_types_id_hidden"] = -1
  959. err := cloudBrainNewDataPrepare(ctx)
  960. if err != nil {
  961. ctx.ServerError("get new cloudbrain info failed", err)
  962. return
  963. }
  964. ctx.HTML(200, tplCloudBrainBenchmarkNew)
  965. }
  966. func getBenchmarkAttachment(benchmarkTypeID, benchmarkChildTypeID int) (*models.BenchmarkDataset, error) {
  967. var childInfo *models.BenchmarkDataset
  968. if benchmarkTypes == nil {
  969. if err := json.Unmarshal([]byte(setting.BenchmarkTypes), &benchmarkTypes); err != nil {
  970. log.Error("json.Unmarshal BenchmarkTypes(%s) failed:%v", setting.BenchmarkTypes, err)
  971. return childInfo, err
  972. }
  973. }
  974. var isExist bool
  975. for _, benchmarkType := range benchmarkTypes.BenchmarkType {
  976. if benchmarkType.Id == benchmarkTypeID {
  977. for _, childType := range benchmarkType.Second {
  978. if childType.Id == benchmarkChildTypeID {
  979. childInfo = childType
  980. isExist = true
  981. break
  982. }
  983. }
  984. break
  985. }
  986. }
  987. if !isExist {
  988. log.Error("no such benchmark_type_id&benchmark_child_type_id")
  989. return childInfo, errors.New("no such benchmark_type_id&benchmark_child_type_id")
  990. }
  991. return childInfo, nil
  992. }
  993. func getBenchmarkGpuQueue(gpuQueue string) (string, error) {
  994. queue := ""
  995. if benchmarkGpuInfos == nil {
  996. if err := json.Unmarshal([]byte(setting.BenchmarkGpuTypes), &benchmarkGpuInfos); err != nil {
  997. log.Error("json.Unmarshal BenchmarkGpuTypes(%s) failed:%v", setting.BenchmarkGpuTypes, err)
  998. return queue, err
  999. }
  1000. }
  1001. var isExist bool
  1002. for _, gpuInfo := range benchmarkGpuInfos.GpuInfo {
  1003. if gpuQueue == gpuInfo.Queue {
  1004. isExist = true
  1005. queue = gpuQueue
  1006. break
  1007. }
  1008. }
  1009. if !isExist {
  1010. log.Error("no such gpuQueue, %s", gpuQueue)
  1011. return queue, errors.New("no such gpuQueue")
  1012. }
  1013. return queue, nil
  1014. }
  1015. func getBenchmarkResourceSpec(resourceSpecID int) (int, error) {
  1016. var id int
  1017. if benchmarkResourceSpecs == nil {
  1018. if err := json.Unmarshal([]byte(setting.BenchmarkResourceSpecs), &benchmarkResourceSpecs); err != nil {
  1019. log.Error("json.Unmarshal BenchmarkResourceSpecs(%s) failed:%v", setting.BenchmarkResourceSpecs, err)
  1020. return id, err
  1021. }
  1022. }
  1023. var isExist bool
  1024. for _, resourceSpec := range benchmarkResourceSpecs.ResourceSpec {
  1025. if resourceSpecID == resourceSpec.Id {
  1026. isExist = true
  1027. id = resourceSpecID
  1028. break
  1029. }
  1030. }
  1031. if !isExist {
  1032. log.Error("no such resourceSpecID, %d", resourceSpecID)
  1033. return id, errors.New("no such resourceSpec")
  1034. }
  1035. return id, nil
  1036. }
  1037. func CloudBrainBenchmarkCreate(ctx *context.Context, form auth.CreateCloudBrainForm) {
  1038. ctx.Data["PageIsCloudBrain"] = true
  1039. jobName := form.JobName
  1040. image := form.Image
  1041. gpuQueue := form.GpuType
  1042. command := cloudbrain.CommandBenchmark
  1043. codePath := setting.JobPath + jobName + cloudbrain.CodeMountPath
  1044. resourceSpecId := cloudbrain.BenchMarkResourceID
  1045. benchmarkTypeID := form.BenchmarkTypeID
  1046. benchmarkChildTypeID := form.BenchmarkChildTypeID
  1047. ctx.Data["description"] = form.Description
  1048. ctx.Data["benchmarkTypeID"] = benchmarkTypeID
  1049. ctx.Data["benchmark_child_types_id_hidden"] = benchmarkChildTypeID
  1050. if !jobNamePattern.MatchString(jobName) {
  1051. cloudBrainNewDataPrepare(ctx)
  1052. ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_jobname_err"), tplCloudBrainBenchmarkNew, &form)
  1053. return
  1054. }
  1055. childInfo, err := getBenchmarkAttachment(benchmarkTypeID, benchmarkChildTypeID)
  1056. if err != nil {
  1057. log.Error("getBenchmarkAttachment failed:%v", err, ctx.Data["MsgID"])
  1058. cloudBrainNewDataPrepare(ctx)
  1059. ctx.RenderWithErr("benchmark type error", tplCloudBrainBenchmarkNew, &form)
  1060. return
  1061. }
  1062. _, err = getBenchmarkGpuQueue(gpuQueue)
  1063. if err != nil {
  1064. log.Error("getBenchmarkGpuQueue failed:%v", err, ctx.Data["MsgID"])
  1065. cloudBrainNewDataPrepare(ctx)
  1066. ctx.RenderWithErr("gpu queue error", tplCloudBrainBenchmarkNew, &form)
  1067. return
  1068. }
  1069. _, err = getBenchmarkResourceSpec(resourceSpecId)
  1070. if err != nil {
  1071. log.Error("getBenchmarkResourceSpec failed:%v", err, ctx.Data["MsgID"])
  1072. cloudBrainNewDataPrepare(ctx)
  1073. ctx.RenderWithErr("resource spec error", tplCloudBrainBenchmarkNew, &form)
  1074. return
  1075. }
  1076. count, err := models.GetCloudbrainCountByUserID(ctx.User.ID, string(models.JobTypeBenchmark))
  1077. if err != nil {
  1078. log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])
  1079. cloudBrainNewDataPrepare(ctx)
  1080. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1081. return
  1082. } else {
  1083. if count >= 1 {
  1084. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  1085. cloudBrainNewDataPrepare(ctx)
  1086. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplCloudBrainBenchmarkNew, &form)
  1087. return
  1088. }
  1089. }
  1090. _, err = models.GetCloudbrainByName(jobName)
  1091. if err == nil {
  1092. log.Error("the job name did already exist", ctx.Data["MsgID"])
  1093. cloudBrainNewDataPrepare(ctx)
  1094. ctx.RenderWithErr("the job name did already exist", tplCloudBrainBenchmarkNew, &form)
  1095. return
  1096. } else {
  1097. if !models.IsErrJobNotExist(err) {
  1098. log.Error("GetCloudbrainByName failed, %v", err, ctx.Data["MsgID"])
  1099. cloudBrainNewDataPrepare(ctx)
  1100. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1101. return
  1102. }
  1103. }
  1104. repo := ctx.Repo.Repository
  1105. os.RemoveAll(codePath)
  1106. if err := downloadCode(repo, codePath); err != nil {
  1107. log.Error("downloadCode failed, %v", err, ctx.Data["MsgID"])
  1108. cloudBrainNewDataPrepare(ctx)
  1109. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1110. return
  1111. }
  1112. if _, err := os.Stat(codePath + "/train.py"); err != nil {
  1113. if os.IsNotExist(err) {
  1114. // file does not exist
  1115. log.Error("train.py does not exist, %v", err, ctx.Data["MsgID"])
  1116. cloudBrainNewDataPrepare(ctx)
  1117. ctx.RenderWithErr("train.py does not exist", tplCloudBrainBenchmarkNew, &form)
  1118. } else {
  1119. log.Error("Stat failed, %v", err, ctx.Data["MsgID"])
  1120. cloudBrainNewDataPrepare(ctx)
  1121. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1122. }
  1123. return
  1124. } else if _, err := os.Stat(codePath + "/test.py"); err != nil {
  1125. if os.IsNotExist(err) {
  1126. // file does not exist
  1127. log.Error("test.py does not exist, %v", err, ctx.Data["MsgID"])
  1128. cloudBrainNewDataPrepare(ctx)
  1129. ctx.RenderWithErr("test.py does not exist", tplCloudBrainBenchmarkNew, &form)
  1130. } else {
  1131. log.Error("Stat failed, %v", err, ctx.Data["MsgID"])
  1132. cloudBrainNewDataPrepare(ctx)
  1133. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1134. }
  1135. return
  1136. }
  1137. if err := uploadCodeToMinio(codePath+"/", jobName, cloudbrain.CodeMountPath+"/"); err != nil {
  1138. log.Error("uploadCodeToMinio failed, %v", err, ctx.Data["MsgID"])
  1139. cloudBrainNewDataPrepare(ctx)
  1140. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1141. return
  1142. }
  1143. benchmarkPath := setting.JobPath + jobName + cloudbrain.BenchMarkMountPath
  1144. var gpuType string
  1145. for _, gpuInfo := range gpuInfos.GpuInfo {
  1146. if gpuInfo.Queue == gpuQueue {
  1147. gpuType = gpuInfo.Value
  1148. }
  1149. }
  1150. if err := downloadRateCode(repo, jobName, childInfo.Owner, childInfo.RepoName, benchmarkPath, form.BenchmarkCategory, gpuType); err != nil {
  1151. log.Error("downloadRateCode failed, %v", err, ctx.Data["MsgID"])
  1152. //cloudBrainNewDataPrepare(ctx)
  1153. //ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1154. //return
  1155. }
  1156. if err := uploadCodeToMinio(benchmarkPath+"/", jobName, cloudbrain.BenchMarkMountPath+"/"); err != nil {
  1157. log.Error("uploadCodeToMinio failed, %v", err, ctx.Data["MsgID"])
  1158. //cloudBrainNewDataPrepare(ctx)
  1159. //ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1160. //return
  1161. }
  1162. err = cloudbrain.GenerateTask(ctx, jobName, image, command, childInfo.Attachment, storage.GetMinioPath(jobName, cloudbrain.CodeMountPath+"/"),
  1163. storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/"),
  1164. storage.GetMinioPath(jobName, cloudbrain.BenchMarkMountPath+"/"), storage.GetMinioPath(jobName, cloudbrain.Snn4imagenetMountPath+"/"),
  1165. storage.GetMinioPath(jobName, cloudbrain.BrainScoreMountPath+"/"), string(models.JobTypeBenchmark), gpuQueue, form.Description,
  1166. benchmarkTypeID, benchmarkChildTypeID, resourceSpecId)
  1167. if err != nil {
  1168. cloudBrainNewDataPrepare(ctx)
  1169. ctx.RenderWithErr(err.Error(), tplCloudBrainBenchmarkNew, &form)
  1170. return
  1171. }
  1172. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/cloudbrain/benchmark")
  1173. }
  1174. func BenchmarkDel(ctx *context.Context) {
  1175. if err := deleteCloudbrainJob(ctx); err != nil {
  1176. log.Error("deleteCloudbrainJob failed: %v", err, ctx.Data["msgID"])
  1177. ctx.ServerError(err.Error(), err)
  1178. return
  1179. }
  1180. var isAdminPage = ctx.Query("isadminpage")
  1181. if ctx.IsUserSiteAdmin() && isAdminPage == "true" {
  1182. ctx.Redirect(setting.AppSubURL + "/admin" + "/cloudbrains")
  1183. } else {
  1184. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/cloudbrain/benchmark")
  1185. }
  1186. }