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 41 kB

4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
3 years ago
3 years ago
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
4 years ago
3 years ago
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
4 years ago
4 years ago
3 years ago
4 years ago
3 years ago
3 years ago
4 years ago
4 years ago
4 years ago
3 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
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
3 years ago
3 years ago
3 years ago
4 years ago
3 years ago
5 years ago
3 years ago
3 years ago
4 years ago
5 years ago
5 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
4 years ago
4 years ago
3 years ago
3 years ago
4 years ago
3 years ago
3 years ago
3 years ago
4 years ago
3 years ago
3 years ago
4 years ago
3 years ago
3 years ago
3 years ago
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
4 years ago
3 years ago
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
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
3 years ago
4 years ago
4 years ago
4 years ago
3 years ago
3 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347
  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 jobID = ctx.Params(":jobid")
  302. task, err := models.GetCloudbrainByJobID(jobID)
  303. if err != nil {
  304. ctx.Data["error"] = err.Error()
  305. }
  306. result, err := cloudbrain.GetJob(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"] = 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. if err := deleteCloudbrainJob(ctx); err != nil {
  507. log.Error("deleteCloudbrainJob failed: %v", err, ctx.Data["msgID"])
  508. ctx.ServerError(err.Error(), err)
  509. return
  510. }
  511. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=all")
  512. }
  513. func deleteCloudbrainJob(ctx *context.Context) error {
  514. task := ctx.Cloudbrain
  515. if task.Status != string(models.JobStopped) && task.Status != string(models.JobFailed) && task.Status != string(models.JobSucceeded) {
  516. log.Error("the job(%s) has not been stopped", task.JobName, ctx.Data["msgID"])
  517. return errors.New("the job has not been stopped")
  518. }
  519. err := models.DeleteJob(task)
  520. if err != nil {
  521. log.Error("DeleteJob failed: %v", err, ctx.Data["msgID"])
  522. return err
  523. }
  524. deleteJobStorage(task.JobName, models.TypeCloudBrainOne)
  525. return nil
  526. }
  527. func CloudBrainShowModels(ctx *context.Context) {
  528. ctx.Data["PageIsCloudBrain"] = true
  529. jobID := ctx.Params(":jobid")
  530. parentDir := ctx.Query("parentDir")
  531. dirArray := strings.Split(parentDir, "/")
  532. task, err := models.GetCloudbrainByJobID(jobID)
  533. if err != nil {
  534. log.Error("no such job!", ctx.Data["msgID"])
  535. ctx.ServerError("no such job:", err)
  536. return
  537. }
  538. //get dirs
  539. dirs, err := GetModelDirs(task.JobName, parentDir)
  540. if err != nil {
  541. log.Error("GetModelDirs failed:%v", err.Error(), ctx.Data["msgID"])
  542. ctx.ServerError("GetModelDirs failed:", err)
  543. return
  544. }
  545. var fileInfos []storage.FileInfo
  546. err = json.Unmarshal([]byte(dirs), &fileInfos)
  547. if err != nil {
  548. log.Error("json.Unmarshal failed:%v", err.Error(), ctx.Data["msgID"])
  549. ctx.ServerError("json.Unmarshal failed:", err)
  550. return
  551. }
  552. for i, fileInfo := range fileInfos {
  553. temp, _ := time.Parse("2006-01-02 15:04:05", fileInfo.ModTime)
  554. fileInfos[i].ModTime = temp.Local().Format("2006-01-02 15:04:05")
  555. }
  556. sort.Slice(fileInfos, func(i, j int) bool {
  557. return fileInfos[i].ModTime > fileInfos[j].ModTime
  558. })
  559. ctx.Data["Path"] = dirArray
  560. ctx.Data["Dirs"] = fileInfos
  561. ctx.Data["task"] = task
  562. ctx.Data["JobID"] = jobID
  563. ctx.HTML(200, tplCloudBrainShowModels)
  564. }
  565. func GetPublicImages(ctx *context.Context) {
  566. getImages(ctx, cloudbrain.Public)
  567. }
  568. func GetCustomImages(ctx *context.Context) {
  569. getImages(ctx, cloudbrain.Custom)
  570. }
  571. func getImages(ctx *context.Context, imageType string) {
  572. log.Info("Get images begin")
  573. page := ctx.QueryInt("page")
  574. size := ctx.QueryInt("size")
  575. name := ctx.Query("name")
  576. getImagesResult, err := cloudbrain.GetImagesPageable(page, size, imageType, name)
  577. if err != nil {
  578. log.Error("Can not get images:%v", err)
  579. ctx.JSON(http.StatusOK, models.GetImagesPayload{
  580. Count: 0,
  581. TotalPages: 0,
  582. ImageInfo: []*models.ImageInfo{},
  583. })
  584. } else {
  585. ctx.JSON(http.StatusOK, getImagesResult.Payload)
  586. }
  587. log.Info("Get images end")
  588. }
  589. func GetModelDirs(jobName string, parentDir string) (string, error) {
  590. var req string
  591. modelActualPath := storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/")
  592. if parentDir == "" {
  593. req = "baseDir=" + modelActualPath
  594. } else {
  595. req = "baseDir=" + modelActualPath + "&parentDir=" + parentDir
  596. }
  597. return getDirs(req)
  598. }
  599. func CloudBrainDownloadModel(ctx *context.Context) {
  600. parentDir := ctx.Query("parentDir")
  601. fileName := ctx.Query("fileName")
  602. jobName := ctx.Query("jobName")
  603. filePath := "jobs/" + jobName + "/model/" + parentDir
  604. url, err := storage.Attachments.PresignedGetURL(filePath, fileName)
  605. if err != nil {
  606. log.Error("PresignedGetURL failed: %v", err.Error(), ctx.Data["msgID"])
  607. ctx.ServerError("PresignedGetURL", err)
  608. return
  609. }
  610. http.Redirect(ctx.Resp, ctx.Req.Request, url, http.StatusMovedPermanently)
  611. }
  612. func GetRate(ctx *context.Context) {
  613. isObjectDetcionAll := ctx.QueryBool("isObjectDetcionAll")
  614. if isObjectDetcionAll {
  615. ctx.Redirect(setting.BenchmarkServerHost + "?username=admin")
  616. return
  617. }
  618. var jobID = ctx.Params(":jobid")
  619. job, err := models.GetCloudbrainByJobID(jobID)
  620. if err != nil {
  621. ctx.ServerError("GetCloudbrainByJobID failed", err)
  622. return
  623. }
  624. if job.JobType == string(models.JobTypeBenchmark) {
  625. log.Info("url=" + setting.BenchmarkServerHost + "?username=" + ctx.User.Name)
  626. ctx.Redirect(setting.BenchmarkServerHost + "?username=" + ctx.User.Name)
  627. } else if job.JobType == string(models.JobTypeSnn4imagenet) {
  628. ctx.Redirect(setting.Snn4imagenetServerHost)
  629. } else if job.JobType == string(models.JobTypeBrainScore) {
  630. ctx.Redirect(setting.BrainScoreServerHost)
  631. } else {
  632. log.Error("JobType error:%s", job.JobType, ctx.Data["msgID"])
  633. }
  634. }
  635. func downloadCode(repo *models.Repository, codePath string) error {
  636. if err := git.Clone(repo.RepoPath(), codePath, git.CloneRepoOptions{}); err != nil {
  637. log.Error("Failed to clone repository: %s (%v)", repo.FullName(), err)
  638. return err
  639. }
  640. configFile, err := os.OpenFile(codePath+"/.git/config", os.O_RDWR, 0666)
  641. if err != nil {
  642. log.Error("open file(%s) failed:%v", codePath+"/,git/config", err)
  643. return err
  644. }
  645. defer configFile.Close()
  646. pos := int64(0)
  647. reader := bufio.NewReader(configFile)
  648. for {
  649. line, err := reader.ReadString('\n')
  650. if err != nil {
  651. if err == io.EOF {
  652. log.Error("not find the remote-url")
  653. return nil
  654. } else {
  655. log.Error("read error: %v", err)
  656. return err
  657. }
  658. }
  659. if strings.Contains(line, "url") && strings.Contains(line, ".git") {
  660. originUrl := "\turl = " + repo.CloneLink().HTTPS + "\n"
  661. if len(line) > len(originUrl) {
  662. originUrl += strings.Repeat(" ", len(line)-len(originUrl))
  663. }
  664. bytes := []byte(originUrl)
  665. _, err := configFile.WriteAt(bytes, pos)
  666. if err != nil {
  667. log.Error("WriteAt failed:%v", err)
  668. return err
  669. }
  670. break
  671. }
  672. pos += int64(len(line))
  673. }
  674. return nil
  675. }
  676. func downloadRateCode(repo *models.Repository, taskName, rateOwnerName, rateRepoName, codePath, benchmarkCategory, gpuType string) error {
  677. err := os.MkdirAll(codePath, os.ModePerm)
  678. if err != nil {
  679. log.Error("mkdir codePath failed", err.Error())
  680. return err
  681. }
  682. repoExt, err := models.GetRepositoryByOwnerAndName(rateOwnerName, rateRepoName)
  683. if err != nil {
  684. log.Error("GetRepositoryByOwnerAndName(%s) failed", rateRepoName, err.Error())
  685. return err
  686. }
  687. if err := git.Clone(repoExt.RepoPath(), codePath, git.CloneRepoOptions{}); err != nil {
  688. log.Error("Failed to clone repository: %s (%v)", repoExt.FullName(), err)
  689. return err
  690. }
  691. fileName := codePath + cloudbrain.TaskInfoName
  692. f, err := os.OpenFile(fileName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, os.ModePerm)
  693. if err != nil {
  694. log.Error("OpenFile failed", err.Error())
  695. return err
  696. }
  697. defer f.Close()
  698. data, err := json.Marshal(models.TaskInfo{
  699. Username: repo.Owner.Name,
  700. TaskName: taskName,
  701. CodeName: repo.Name,
  702. BenchmarkCategory: strings.Split(benchmarkCategory, ","),
  703. CodeLink: strings.TrimSuffix(repo.CloneLink().HTTPS, ".git"),
  704. GpuType: gpuType,
  705. })
  706. if err != nil {
  707. log.Error("json.Marshal failed", err.Error())
  708. return err
  709. }
  710. _, err = f.Write(data)
  711. if err != nil {
  712. log.Error("WriteString failed", err.Error())
  713. return err
  714. }
  715. return nil
  716. }
  717. func uploadCodeToMinio(codePath, jobName, parentDir string) error {
  718. files, err := readDir(codePath)
  719. if err != nil {
  720. log.Error("readDir(%s) failed: %s", codePath, err.Error())
  721. return err
  722. }
  723. for _, file := range files {
  724. if file.IsDir() {
  725. if err = uploadCodeToMinio(codePath+file.Name()+"/", jobName, parentDir+file.Name()+"/"); err != nil {
  726. log.Error("uploadCodeToMinio(%s) failed: %s", file.Name(), err.Error())
  727. return err
  728. }
  729. } else {
  730. destObject := setting.CBCodePathPrefix + jobName + parentDir + file.Name()
  731. sourceFile := codePath + file.Name()
  732. err = storage.Attachments.UploadObject(destObject, sourceFile)
  733. if err != nil {
  734. log.Error("UploadObject(%s) failed: %s", file.Name(), err.Error())
  735. return err
  736. }
  737. }
  738. }
  739. return nil
  740. }
  741. func mkModelPath(modelPath string) error {
  742. err := os.MkdirAll(modelPath, os.ModePerm)
  743. if err != nil {
  744. log.Error("MkdirAll(%s) failed:%v", modelPath, err)
  745. return err
  746. }
  747. fileName := modelPath + "README"
  748. f, err := os.OpenFile(fileName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, os.ModePerm)
  749. if err != nil {
  750. log.Error("OpenFile failed", err.Error())
  751. return err
  752. }
  753. defer f.Close()
  754. _, err = f.WriteString("You can put the model file into this directory and download it by the web page.")
  755. if err != nil {
  756. log.Error("WriteString failed", err.Error())
  757. return err
  758. }
  759. return nil
  760. }
  761. func deleteJobStorage(jobName string, cloudbrainType int) error {
  762. //delete local
  763. localJobPath := setting.JobPath + jobName
  764. err := os.RemoveAll(localJobPath)
  765. if err != nil {
  766. log.Error("RemoveAll(%s) failed:%v", localJobPath, err)
  767. }
  768. //delete oss
  769. if cloudbrainType == models.TypeCloudBrainOne {
  770. dirPath := setting.CBCodePathPrefix + jobName + "/"
  771. err = storage.Attachments.DeleteDir(dirPath)
  772. if err != nil {
  773. log.Error("DeleteDir(%s) failed:%v", localJobPath, err)
  774. }
  775. } else if cloudbrainType == models.TypeCloudBrainTwo {
  776. //dirPath := setting.CodePathPrefix + jobName + "/"
  777. //err = storage.ObsRemoveObject(setting.Bucket, dirPath)
  778. //if err != nil {
  779. // log.Error("ObsRemoveObject(%s) failed:%v", localJobPath, err)
  780. //}
  781. log.Info("no need to delete")
  782. } else {
  783. log.Error("cloudbrainType(%d) error", cloudbrainType)
  784. }
  785. return nil
  786. }
  787. func SyncCloudbrainStatus() {
  788. cloudBrains, err := models.GetCloudBrainUnStoppedJob()
  789. if err != nil {
  790. log.Error("GetCloudBrainUnStoppedJob failed:", err.Error())
  791. return
  792. }
  793. for _, task := range cloudBrains {
  794. if task.Type == models.TypeCloudBrainOne {
  795. result, err := cloudbrain.GetJob(task.JobID)
  796. if err != nil {
  797. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  798. continue
  799. }
  800. if result != nil {
  801. jobRes, _ := models.ConvertToJobResultPayload(result.Payload)
  802. taskRoles := jobRes.TaskRoles
  803. taskRes, _ := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
  804. task.Status = taskRes.TaskStatuses[0].State
  805. if task.Status != string(models.JobWaiting) {
  806. task.Duration = time.Now().Unix() - taskRes.TaskStatuses[0].StartAt.Unix()
  807. err = models.UpdateJob(task)
  808. if err != nil {
  809. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  810. }
  811. var maxDuration int64
  812. if task.JobType == string(models.JobTypeBenchmark) {
  813. maxDuration = setting.BenchmarkMaxDuration
  814. } else {
  815. maxDuration = setting.MaxDuration
  816. }
  817. if task.Duration >= maxDuration {
  818. log.Info("begin to stop job(%s), because of the duration", task.JobName)
  819. err = cloudbrain.StopJob(task.JobID)
  820. if err != nil {
  821. log.Error("StopJob(%s) failed:%v", task.JobName, err)
  822. continue
  823. }
  824. task.Status = string(models.JobStopped)
  825. err = models.UpdateJob(task)
  826. if err != nil {
  827. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  828. }
  829. }
  830. }
  831. }
  832. } else if task.Type == models.TypeCloudBrainTwo {
  833. if task.JobType == string(models.JobTypeDebug) {
  834. //result, err := modelarts.GetJob(task.JobID)
  835. result, err := modelarts.GetNotebook2(task.JobID)
  836. if err != nil {
  837. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  838. continue
  839. }
  840. if result != nil {
  841. task.Status = result.Status
  842. err = models.UpdateJob(task)
  843. if err != nil {
  844. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  845. continue
  846. }
  847. }
  848. } else if task.JobType == string(models.JobTypeTrain) {
  849. result, err := modelarts.GetTrainJob(task.JobID, strconv.FormatInt(task.VersionID, 10))
  850. if err != nil {
  851. log.Error("GetTrainJob(%s) failed:%v", task.JobName, err)
  852. continue
  853. }
  854. if result != nil {
  855. task.Status = modelarts.TransTrainJobStatus(result.IntStatus)
  856. task.Duration = result.Duration
  857. task.TrainJobDuration = result.TrainJobDuration
  858. if result.Duration != 0 {
  859. task.TrainJobDuration = util.AddZero(result.Duration/3600000) + ":" + util.AddZero(result.Duration%3600000/60000) + ":" + util.AddZero(result.Duration%60000/1000)
  860. } else {
  861. task.TrainJobDuration = "00:00:00"
  862. }
  863. err = models.UpdateJob(task)
  864. if err != nil {
  865. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  866. continue
  867. }
  868. }
  869. } else {
  870. log.Error("task.JobType(%s) is error:%s", task.JobName, task.JobType)
  871. }
  872. } else {
  873. log.Error("task.Type(%s) is error:%d", task.JobName, task.Type)
  874. }
  875. }
  876. return
  877. }
  878. func CloudBrainBenchmarkIndex(ctx *context.Context) {
  879. MustEnableCloudbrain(ctx)
  880. repo := ctx.Repo.Repository
  881. page := ctx.QueryInt("page")
  882. if page <= 0 {
  883. page = 1
  884. }
  885. var jobTypes []string
  886. jobTypes = append(jobTypes, string(models.JobTypeBenchmark))
  887. ciTasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  888. ListOptions: models.ListOptions{
  889. Page: page,
  890. PageSize: setting.UI.IssuePagingNum,
  891. },
  892. RepoID: repo.ID,
  893. Type: models.TypeCloudBrainOne,
  894. JobTypes: jobTypes,
  895. })
  896. if err != nil {
  897. ctx.ServerError("Get debugjob faild:", err)
  898. return
  899. }
  900. for i, task := range ciTasks {
  901. ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  902. ciTasks[i].Cloudbrain.ComputeResource = task.ComputeResource
  903. var duration int64
  904. if task.Status == string(models.JobRunning) {
  905. duration = time.Now().Unix() - int64(task.Cloudbrain.CreatedUnix)
  906. } else {
  907. duration = int64(task.Cloudbrain.UpdatedUnix) - int64(task.Cloudbrain.CreatedUnix)
  908. }
  909. ciTasks[i].TrainJobDuration = util.AddZero(duration/3600000) + ":" + util.AddZero(duration%3600000/60000) + ":" + util.AddZero(duration%60000/1000)
  910. }
  911. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  912. ctx.Data["Page"] = pager
  913. ctx.Data["PageIsCloudBrain"] = true
  914. ctx.Data["Tasks"] = ciTasks
  915. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  916. ctx.Data["RepoIsEmpty"] = repo.IsEmpty
  917. ctx.HTML(200, tplCloudBrainBenchmarkIndex)
  918. }
  919. func GetChildTypes(ctx *context.Context) {
  920. benchmarkTypeID := ctx.QueryInt("benchmark_type_id")
  921. re := make(map[string]interface{})
  922. for {
  923. if benchmarkTypes == nil {
  924. if err := json.Unmarshal([]byte(setting.BenchmarkTypes), &benchmarkTypes); err != nil {
  925. log.Error("json.Unmarshal BenchmarkTypes(%s) failed:%v", setting.BenchmarkTypes, err, ctx.Data["MsgID"])
  926. re["errMsg"] = "system error"
  927. break
  928. }
  929. }
  930. var isExist bool
  931. for _, benchmarkType := range benchmarkTypes.BenchmarkType {
  932. if benchmarkTypeID == benchmarkType.Id {
  933. isExist = true
  934. re["child_types"] = benchmarkType.Second
  935. re["result_code"] = "0"
  936. break
  937. }
  938. }
  939. if !isExist {
  940. re["result_code"] = "1"
  941. log.Error("no such benchmark_type_id", ctx.Data["MsgID"])
  942. re["errMsg"] = "system error"
  943. break
  944. }
  945. break
  946. }
  947. ctx.JSON(200, re)
  948. }
  949. func CloudBrainBenchmarkNew(ctx *context.Context) {
  950. err := cloudBrainNewDataPrepare(ctx)
  951. if err != nil {
  952. ctx.ServerError("get new cloudbrain info failed", err)
  953. return
  954. }
  955. ctx.HTML(200, tplCloudBrainBenchmarkNew)
  956. }
  957. func getBenchmarkAttachment(benchmarkTypeID, benchmarkChildTypeID int) (*models.BenchmarkDataset, error) {
  958. var childInfo *models.BenchmarkDataset
  959. if benchmarkTypes == nil {
  960. if err := json.Unmarshal([]byte(setting.BenchmarkTypes), &benchmarkTypes); err != nil {
  961. log.Error("json.Unmarshal BenchmarkTypes(%s) failed:%v", setting.BenchmarkTypes, err)
  962. return childInfo, err
  963. }
  964. }
  965. var isExist bool
  966. for _, benchmarkType := range benchmarkTypes.BenchmarkType {
  967. if benchmarkType.Id == benchmarkTypeID {
  968. for _, childType := range benchmarkType.Second {
  969. if childType.Id == benchmarkChildTypeID {
  970. childInfo = childType
  971. isExist = true
  972. break
  973. }
  974. }
  975. break
  976. }
  977. }
  978. if !isExist {
  979. log.Error("no such benchmark_type_id&benchmark_child_type_id")
  980. return childInfo, errors.New("no such benchmark_type_id&benchmark_child_type_id")
  981. }
  982. return childInfo, nil
  983. }
  984. func getBenchmarkGpuQueue(gpuQueue string) (string, error) {
  985. queue := ""
  986. if benchmarkGpuInfos == nil {
  987. if err := json.Unmarshal([]byte(setting.BenchmarkGpuTypes), &benchmarkGpuInfos); err != nil {
  988. log.Error("json.Unmarshal BenchmarkGpuTypes(%s) failed:%v", setting.BenchmarkGpuTypes, err)
  989. return queue, err
  990. }
  991. }
  992. var isExist bool
  993. for _, gpuInfo := range benchmarkGpuInfos.GpuInfo {
  994. if gpuQueue == gpuInfo.Queue {
  995. isExist = true
  996. queue = gpuQueue
  997. break
  998. }
  999. }
  1000. if !isExist {
  1001. log.Error("no such gpuQueue, %s", gpuQueue)
  1002. return queue, errors.New("no such gpuQueue")
  1003. }
  1004. return queue, nil
  1005. }
  1006. func getBenchmarkResourceSpec(resourceSpecID int) (int, error) {
  1007. var id int
  1008. if benchmarkResourceSpecs == nil {
  1009. if err := json.Unmarshal([]byte(setting.BenchmarkResourceSpecs), &benchmarkResourceSpecs); err != nil {
  1010. log.Error("json.Unmarshal BenchmarkResourceSpecs(%s) failed:%v", setting.BenchmarkResourceSpecs, err)
  1011. return id, err
  1012. }
  1013. }
  1014. var isExist bool
  1015. for _, resourceSpec := range benchmarkResourceSpecs.ResourceSpec {
  1016. if resourceSpecID == resourceSpec.Id {
  1017. isExist = true
  1018. id = resourceSpecID
  1019. break
  1020. }
  1021. }
  1022. if !isExist {
  1023. log.Error("no such resourceSpecID, %d", resourceSpecID)
  1024. return id, errors.New("no such resourceSpec")
  1025. }
  1026. return id, nil
  1027. }
  1028. func CloudBrainBenchmarkCreate(ctx *context.Context, form auth.CreateCloudBrainForm) {
  1029. ctx.Data["PageIsCloudBrain"] = true
  1030. jobName := form.JobName
  1031. image := form.Image
  1032. gpuQueue := form.GpuType
  1033. command := cloudbrain.CommandBenchmark
  1034. codePath := setting.JobPath + jobName + cloudbrain.CodeMountPath
  1035. resourceSpecId := cloudbrain.BenchMarkResourceID
  1036. benchmarkTypeID := form.BenchmarkTypeID
  1037. benchmarkChildTypeID := form.BenchmarkChildTypeID
  1038. if !jobNamePattern.MatchString(jobName) {
  1039. cloudBrainNewDataPrepare(ctx)
  1040. ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_jobname_err"), tplCloudBrainBenchmarkNew, &form)
  1041. return
  1042. }
  1043. childInfo, err := getBenchmarkAttachment(benchmarkTypeID, benchmarkChildTypeID)
  1044. if err != nil {
  1045. log.Error("getBenchmarkAttachment failed:%v", err, ctx.Data["MsgID"])
  1046. cloudBrainNewDataPrepare(ctx)
  1047. ctx.RenderWithErr("benchmark type error", tplCloudBrainBenchmarkNew, &form)
  1048. return
  1049. }
  1050. _, err = getBenchmarkGpuQueue(gpuQueue)
  1051. if err != nil {
  1052. log.Error("getBenchmarkGpuQueue failed:%v", err, ctx.Data["MsgID"])
  1053. cloudBrainNewDataPrepare(ctx)
  1054. ctx.RenderWithErr("gpu queue error", tplCloudBrainBenchmarkNew, &form)
  1055. return
  1056. }
  1057. _, err = getBenchmarkResourceSpec(resourceSpecId)
  1058. if err != nil {
  1059. log.Error("getBenchmarkResourceSpec failed:%v", err, ctx.Data["MsgID"])
  1060. cloudBrainNewDataPrepare(ctx)
  1061. ctx.RenderWithErr("resource spec error", tplCloudBrainBenchmarkNew, &form)
  1062. return
  1063. }
  1064. count, err := models.GetCloudbrainCountByUserID(ctx.User.ID, string(models.JobTypeBenchmark))
  1065. if err != nil {
  1066. log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])
  1067. cloudBrainNewDataPrepare(ctx)
  1068. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1069. return
  1070. } else {
  1071. if count >= 1 {
  1072. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  1073. cloudBrainNewDataPrepare(ctx)
  1074. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplCloudBrainBenchmarkNew, &form)
  1075. return
  1076. }
  1077. }
  1078. _, err = models.GetCloudbrainByName(jobName)
  1079. if err == nil {
  1080. log.Error("the job name did already exist", ctx.Data["MsgID"])
  1081. cloudBrainNewDataPrepare(ctx)
  1082. ctx.RenderWithErr("the job name did already exist", tplCloudBrainBenchmarkNew, &form)
  1083. return
  1084. } else {
  1085. if !models.IsErrJobNotExist(err) {
  1086. log.Error("GetCloudbrainByName failed, %v", err, ctx.Data["MsgID"])
  1087. cloudBrainNewDataPrepare(ctx)
  1088. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1089. return
  1090. }
  1091. }
  1092. repo := ctx.Repo.Repository
  1093. os.RemoveAll(codePath)
  1094. if err := downloadCode(repo, codePath); err != nil {
  1095. log.Error("downloadCode failed, %v", err, ctx.Data["MsgID"])
  1096. cloudBrainNewDataPrepare(ctx)
  1097. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1098. return
  1099. }
  1100. if _, err := os.Stat(codePath + "/train.py"); err != nil {
  1101. if os.IsNotExist(err) {
  1102. // file does not exist
  1103. log.Error("train.py does not exist, %v", err, ctx.Data["MsgID"])
  1104. cloudBrainNewDataPrepare(ctx)
  1105. ctx.RenderWithErr("train.py does not exist", tplCloudBrainBenchmarkNew, &form)
  1106. } else {
  1107. log.Error("Stat failed, %v", err, ctx.Data["MsgID"])
  1108. cloudBrainNewDataPrepare(ctx)
  1109. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1110. }
  1111. return
  1112. } else if _, err := os.Stat(codePath + "/test.py"); err != nil {
  1113. if os.IsNotExist(err) {
  1114. // file does not exist
  1115. log.Error("test.py does not exist, %v", err, ctx.Data["MsgID"])
  1116. cloudBrainNewDataPrepare(ctx)
  1117. ctx.RenderWithErr("test.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. }
  1125. if err := uploadCodeToMinio(codePath+"/", jobName, cloudbrain.CodeMountPath+"/"); err != nil {
  1126. log.Error("uploadCodeToMinio failed, %v", err, ctx.Data["MsgID"])
  1127. cloudBrainNewDataPrepare(ctx)
  1128. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1129. return
  1130. }
  1131. benchmarkPath := setting.JobPath + jobName + cloudbrain.BenchMarkMountPath
  1132. var gpuType string
  1133. for _, gpuInfo := range gpuInfos.GpuInfo {
  1134. if gpuInfo.Queue == gpuQueue {
  1135. gpuType = gpuInfo.Value
  1136. }
  1137. }
  1138. if err := downloadRateCode(repo, jobName, childInfo.Owner, childInfo.RepoName, benchmarkPath, form.BenchmarkCategory, gpuType); err != nil {
  1139. log.Error("downloadRateCode failed, %v", err, ctx.Data["MsgID"])
  1140. //cloudBrainNewDataPrepare(ctx)
  1141. //ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1142. //return
  1143. }
  1144. if err := uploadCodeToMinio(benchmarkPath+"/", jobName, cloudbrain.BenchMarkMountPath+"/"); err != nil {
  1145. log.Error("uploadCodeToMinio failed, %v", err, ctx.Data["MsgID"])
  1146. //cloudBrainNewDataPrepare(ctx)
  1147. //ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1148. //return
  1149. }
  1150. err = cloudbrain.GenerateTask(ctx, jobName, image, command, childInfo.Attachment, storage.GetMinioPath(jobName, cloudbrain.CodeMountPath+"/"),
  1151. storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/"),
  1152. storage.GetMinioPath(jobName, cloudbrain.BenchMarkMountPath+"/"), storage.GetMinioPath(jobName, cloudbrain.Snn4imagenetMountPath+"/"),
  1153. storage.GetMinioPath(jobName, cloudbrain.BrainScoreMountPath+"/"), string(models.JobTypeBenchmark), gpuQueue, form.Description,
  1154. benchmarkTypeID, benchmarkChildTypeID, resourceSpecId)
  1155. if err != nil {
  1156. cloudBrainNewDataPrepare(ctx)
  1157. ctx.RenderWithErr(err.Error(), tplCloudBrainBenchmarkNew, &form)
  1158. return
  1159. }
  1160. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/cloudbrain/benchmark")
  1161. }
  1162. func BenchmarkDel(ctx *context.Context) {
  1163. if err := deleteCloudbrainJob(ctx); err != nil {
  1164. log.Error("deleteCloudbrainJob failed: %v", err, ctx.Data["msgID"])
  1165. ctx.ServerError(err.Error(), err)
  1166. return
  1167. }
  1168. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/cloudbrain/benchmark")
  1169. }