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 54 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
3 years ago
3 years ago
4 years ago
3 years ago
3 years ago
4 years ago
3 years ago
3 years ago
4 years ago
3 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
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
4 years ago
4 years ago
3 years ago
4 years ago
3 years ago
5 years ago
4 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
4 years ago
3 years ago
3 years ago
3 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
5 years ago
4 years ago
4 years ago
5 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
4 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
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
3 years ago
5 years ago
3 years ago
5 years ago
3 years ago
3 years ago
5 years ago
3 years ago
5 years ago
3 years ago
5 years ago
3 years ago
5 years ago
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
4 years ago
4 years ago
3 years ago
3 years ago
4 years ago
5 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
5 years ago
3 years ago
3 years ago
3 years ago
5 years ago
5 years ago
5 years ago
3 years ago
3 years ago
5 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
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
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
5 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
3 years ago
3 years ago
4 years ago
4 years ago
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
3 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
4 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
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
4 years ago
3 years ago
4 years ago
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
3 years ago
4 years ago
3 years ago
3 years ago
3 years ago
4 years ago
3 years ago
4 years ago
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741
  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/modules/timeutil"
  16. "github.com/unknwon/i18n"
  17. "code.gitea.io/gitea/models"
  18. "code.gitea.io/gitea/modules/auth"
  19. "code.gitea.io/gitea/modules/base"
  20. "code.gitea.io/gitea/modules/cloudbrain"
  21. "code.gitea.io/gitea/modules/context"
  22. "code.gitea.io/gitea/modules/git"
  23. "code.gitea.io/gitea/modules/log"
  24. "code.gitea.io/gitea/modules/modelarts"
  25. "code.gitea.io/gitea/modules/setting"
  26. "code.gitea.io/gitea/modules/storage"
  27. "code.gitea.io/gitea/modules/util"
  28. )
  29. const (
  30. tplCloudBrainNew base.TplName = "repo/cloudbrain/new"
  31. tplCloudBrainShow base.TplName = "repo/cloudbrain/show"
  32. tplCloudBrainShowModels base.TplName = "repo/cloudbrain/models/index"
  33. tplCloudBrainBenchmarkIndex base.TplName = "repo/cloudbrain/benchmark/index"
  34. tplCloudBrainBenchmarkNew base.TplName = "repo/cloudbrain/benchmark/new"
  35. tplCloudBrainBenchmarkShow base.TplName = "repo/cloudbrain/benchmark/show"
  36. tplCloudBrainTrainJobNew base.TplName = "repo/cloudbrain/trainjob/new"
  37. tplCloudBrainTrainJobShow base.TplName = "repo/cloudbrain/trainjob/show"
  38. )
  39. var (
  40. gpuInfos *models.GpuInfos
  41. categories *models.Categories
  42. benchmarkTypes *models.BenchmarkTypes
  43. benchmarkGpuInfos *models.GpuInfos
  44. benchmarkResourceSpecs *models.ResourceSpecs
  45. trainGpuInfos *models.GpuInfos
  46. )
  47. const BENCHMARK_TYPE_CODE = "repo.cloudbrain.benchmark.types"
  48. var benchmarkTypesMap = make(map[string]*models.BenchmarkTypes, 0)
  49. var jobNamePattern = regexp.MustCompile(`^[a-z0-9][a-z0-9-_]{1,34}[a-z0-9-]$`)
  50. // MustEnableDataset check if repository enable internal cb
  51. func MustEnableCloudbrain(ctx *context.Context) {
  52. if !ctx.Repo.CanRead(models.UnitTypeCloudBrain) {
  53. ctx.NotFound("MustEnableCloudbrain", nil)
  54. return
  55. }
  56. }
  57. func cutString(str string, lens int) string {
  58. if len(str) < lens {
  59. return str
  60. }
  61. return str[:lens]
  62. }
  63. func jobNamePrefixValid(s string) string {
  64. lowStr := strings.ToLower(s)
  65. re := regexp.MustCompile(`[^a-z0-9_\\-]+`)
  66. removeSpecial := re.ReplaceAllString(lowStr, "")
  67. re = regexp.MustCompile(`^[_\\-]+`)
  68. return re.ReplaceAllString(removeSpecial, "")
  69. }
  70. func cloudBrainNewDataPrepare(ctx *context.Context) error {
  71. ctx.Data["PageIsCloudBrain"] = true
  72. t := time.Now()
  73. var displayJobName = jobNamePrefixValid(cutString(ctx.User.Name, 5)) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  74. ctx.Data["display_job_name"] = displayJobName
  75. result, err := cloudbrain.GetImages()
  76. if err != nil {
  77. ctx.Data["error"] = err.Error()
  78. log.Error("cloudbrain.GetImages failed:", err.Error(), ctx.Data["MsgID"])
  79. }
  80. for i, payload := range result.Payload.ImageInfo {
  81. if strings.HasPrefix(result.Payload.ImageInfo[i].Place, "192.168") {
  82. result.Payload.ImageInfo[i].PlaceView = payload.Place[strings.Index(payload.Place, "/"):len(payload.Place)]
  83. } else {
  84. result.Payload.ImageInfo[i].PlaceView = payload.Place
  85. }
  86. }
  87. ctx.Data["images"] = result.Payload.ImageInfo
  88. resultPublic, err := cloudbrain.GetPublicImages()
  89. if err != nil {
  90. ctx.Data["error"] = err.Error()
  91. log.Error("cloudbrain.GetPublicImages failed:", err.Error(), ctx.Data["MsgID"])
  92. }
  93. for i, payload := range resultPublic.Payload.ImageInfo {
  94. if strings.HasPrefix(resultPublic.Payload.ImageInfo[i].Place, "192.168") {
  95. resultPublic.Payload.ImageInfo[i].PlaceView = payload.Place[strings.Index(payload.Place, "/"):len(payload.Place)]
  96. } else {
  97. resultPublic.Payload.ImageInfo[i].PlaceView = payload.Place
  98. }
  99. }
  100. ctx.Data["public_images"] = resultPublic.Payload.ImageInfo
  101. attachs, err := models.GetAllUserAttachments(ctx.User.ID)
  102. if err != nil {
  103. log.Error("GetAllUserAttachments failed: %v", err, ctx.Data["MsgID"])
  104. return err
  105. }
  106. ctx.Data["attachments"] = attachs
  107. ctx.Data["command"] = cloudbrain.Command
  108. ctx.Data["code_path"] = cloudbrain.CodeMountPath
  109. ctx.Data["dataset_path"] = cloudbrain.DataSetMountPath
  110. ctx.Data["model_path"] = cloudbrain.ModelMountPath
  111. ctx.Data["benchmark_path"] = cloudbrain.BenchMarkMountPath
  112. ctx.Data["is_benchmark_enabled"] = setting.IsBenchmarkEnabled
  113. if categories == nil {
  114. json.Unmarshal([]byte(setting.BenchmarkCategory), &categories)
  115. }
  116. ctx.Data["benchmark_categories"] = categories.Category
  117. ctx.Data["benchmark_types"] = GetBenchmarkTypes(ctx).BenchmarkType
  118. if gpuInfos == nil {
  119. json.Unmarshal([]byte(setting.GpuTypes), &gpuInfos)
  120. }
  121. ctx.Data["gpu_types"] = gpuInfos.GpuInfo
  122. if trainGpuInfos == nil {
  123. json.Unmarshal([]byte(setting.TrainGpuTypes), &trainGpuInfos)
  124. }
  125. ctx.Data["train_gpu_types"] = trainGpuInfos.GpuInfo
  126. if benchmarkGpuInfos == nil {
  127. json.Unmarshal([]byte(setting.BenchmarkGpuTypes), &benchmarkGpuInfos)
  128. }
  129. ctx.Data["benchmark_gpu_types"] = benchmarkGpuInfos.GpuInfo
  130. if benchmarkResourceSpecs == nil {
  131. json.Unmarshal([]byte(setting.BenchmarkResourceSpecs), &benchmarkResourceSpecs)
  132. }
  133. ctx.Data["benchmark_resource_specs"] = benchmarkResourceSpecs.ResourceSpec
  134. if cloudbrain.ResourceSpecs == nil {
  135. json.Unmarshal([]byte(setting.ResourceSpecs), &cloudbrain.ResourceSpecs)
  136. }
  137. ctx.Data["resource_specs"] = cloudbrain.ResourceSpecs.ResourceSpec
  138. if cloudbrain.TrainResourceSpecs == nil {
  139. json.Unmarshal([]byte(setting.TrainResourceSpecs), &cloudbrain.TrainResourceSpecs)
  140. }
  141. ctx.Data["train_resource_specs"] = cloudbrain.TrainResourceSpecs.ResourceSpec
  142. ctx.Data["params"] = ""
  143. ctx.Data["branchName"] = ctx.Repo.BranchName
  144. ctx.Data["snn4imagenet_path"] = cloudbrain.Snn4imagenetMountPath
  145. ctx.Data["is_snn4imagenet_enabled"] = setting.IsSnn4imagenetEnabled
  146. ctx.Data["brainscore_path"] = cloudbrain.BrainScoreMountPath
  147. ctx.Data["is_brainscore_enabled"] = setting.IsBrainScoreEnabled
  148. ctx.Data["cloudbraintype"] = models.TypeCloudBrainOne
  149. return nil
  150. }
  151. func CloudBrainNew(ctx *context.Context) {
  152. err := cloudBrainNewDataPrepare(ctx)
  153. if err != nil {
  154. ctx.ServerError("get new cloudbrain info failed", err)
  155. return
  156. }
  157. ctx.HTML(200, tplCloudBrainNew)
  158. }
  159. func CloudBrainCreate(ctx *context.Context, form auth.CreateCloudBrainForm) {
  160. ctx.Data["PageIsCloudBrain"] = true
  161. displayJobName := form.DisplayJobName
  162. jobName := util.ConvertDisplayJobNameToJobName(displayJobName)
  163. image := form.Image
  164. uuid := form.Attachment
  165. jobType := form.JobType
  166. gpuQueue := form.GpuType
  167. codePath := setting.JobPath + jobName + cloudbrain.CodeMountPath
  168. resourceSpecId := form.ResourceSpecId
  169. branchName := form.BranchName
  170. repo := ctx.Repo.Repository
  171. tpl := tplCloudBrainNew
  172. command := cloudbrain.Command
  173. if jobType == string(models.JobTypeTrain) {
  174. tpl = tplCloudBrainTrainJobNew
  175. commandTrain, err := getTrainJobCommand(form)
  176. if err != nil {
  177. log.Error("getTrainJobCommand failed: %v", err)
  178. ctx.RenderWithErr(err.Error(), tpl, &form)
  179. return
  180. }
  181. command = commandTrain
  182. }
  183. tasks, err := models.GetCloudbrainsByDisplayJobName(repo.ID, jobType, displayJobName)
  184. if err == nil {
  185. if len(tasks) != 0 {
  186. log.Error("the job name did already exist", ctx.Data["MsgID"])
  187. cloudBrainNewDataPrepare(ctx)
  188. ctx.RenderWithErr("the job name did already exist", tpl, &form)
  189. return
  190. }
  191. } else {
  192. if !models.IsErrJobNotExist(err) {
  193. log.Error("system error, %v", err, ctx.Data["MsgID"])
  194. cloudBrainNewDataPrepare(ctx)
  195. ctx.RenderWithErr("system error", tpl, &form)
  196. return
  197. }
  198. }
  199. if !jobNamePattern.MatchString(displayJobName) {
  200. ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_jobname_err"), tpl, &form)
  201. return
  202. }
  203. if jobType != string(models.JobTypeBenchmark) && jobType != string(models.JobTypeDebug) && jobType != string(models.JobTypeSnn4imagenet) && jobType != string(models.JobTypeBrainScore) && jobType != string(models.JobTypeTrain) {
  204. log.Error("jobtype error:", jobType, ctx.Data["MsgID"])
  205. cloudBrainNewDataPrepare(ctx)
  206. ctx.RenderWithErr("jobtype error", tpl, &form)
  207. return
  208. }
  209. count, err := models.GetCloudbrainCountByUserID(ctx.User.ID, jobType)
  210. if err != nil {
  211. log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])
  212. cloudBrainNewDataPrepare(ctx)
  213. ctx.RenderWithErr("system error", tpl, &form)
  214. return
  215. } else {
  216. if count >= 1 {
  217. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  218. cloudBrainNewDataPrepare(ctx)
  219. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tpl, &form)
  220. return
  221. }
  222. }
  223. if branchName == "" {
  224. branchName = cloudbrain.DefaultBranchName
  225. }
  226. downloadCode(repo, codePath, branchName)
  227. uploadCodeToMinio(codePath+"/", jobName, cloudbrain.CodeMountPath+"/")
  228. modelPath := setting.JobPath + jobName + cloudbrain.ModelMountPath + "/"
  229. mkModelPath(modelPath)
  230. uploadCodeToMinio(modelPath, jobName, cloudbrain.ModelMountPath+"/")
  231. benchmarkPath := setting.JobPath + jobName + cloudbrain.BenchMarkMountPath
  232. if setting.IsBenchmarkEnabled && jobType == string(models.JobTypeBenchmark) {
  233. var gpuType string
  234. for _, gpuInfo := range gpuInfos.GpuInfo {
  235. if gpuInfo.Queue == gpuQueue {
  236. gpuType = gpuInfo.Value
  237. }
  238. }
  239. downloadRateCode(repo, jobName, setting.BenchmarkOwner, setting.BenchmarkName, benchmarkPath, form.BenchmarkCategory, gpuType)
  240. uploadCodeToMinio(benchmarkPath+"/", jobName, cloudbrain.BenchMarkMountPath+"/")
  241. }
  242. snn4imagenetPath := setting.JobPath + jobName + cloudbrain.Snn4imagenetMountPath
  243. if setting.IsSnn4imagenetEnabled && jobType == string(models.JobTypeSnn4imagenet) {
  244. downloadRateCode(repo, jobName, setting.Snn4imagenetOwner, setting.Snn4imagenetName, snn4imagenetPath, "", "")
  245. uploadCodeToMinio(snn4imagenetPath+"/", jobName, cloudbrain.Snn4imagenetMountPath+"/")
  246. }
  247. brainScorePath := setting.JobPath + jobName + cloudbrain.BrainScoreMountPath
  248. if setting.IsBrainScoreEnabled && jobType == string(models.JobTypeBrainScore) {
  249. downloadRateCode(repo, jobName, setting.BrainScoreOwner, setting.BrainScoreName, brainScorePath, "", "")
  250. uploadCodeToMinio(brainScorePath+"/", jobName, cloudbrain.BrainScoreMountPath+"/")
  251. }
  252. err = cloudbrain.GenerateTask(ctx, displayJobName, jobName, image, command, uuid, storage.GetMinioPath(jobName, cloudbrain.CodeMountPath+"/"),
  253. storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/"),
  254. storage.GetMinioPath(jobName, cloudbrain.BenchMarkMountPath+"/"), storage.GetMinioPath(jobName, cloudbrain.Snn4imagenetMountPath+"/"),
  255. storage.GetMinioPath(jobName, cloudbrain.BrainScoreMountPath+"/"), jobType, gpuQueue, form.Description, branchName, form.BootFile, form.Params,
  256. 0, 0, resourceSpecId)
  257. if err != nil {
  258. cloudBrainNewDataPrepare(ctx)
  259. ctx.RenderWithErr(err.Error(), tpl, &form)
  260. return
  261. }
  262. if jobType == string(models.JobTypeTrain) {
  263. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job?listType=all")
  264. } else {
  265. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=all")
  266. }
  267. }
  268. func CloudBrainRestart(ctx *context.Context) {
  269. var ID = ctx.Params(":id")
  270. var resultCode = "0"
  271. var errorMsg = ""
  272. var status = string(models.JobWaiting)
  273. task := ctx.Cloudbrain
  274. for {
  275. if task.Status != string(models.JobStopped) && task.Status != string(models.JobSucceeded) && task.Status != string(models.JobFailed) {
  276. log.Error("the job(%s) is not stopped", task.JobName, ctx.Data["MsgID"])
  277. resultCode = "-1"
  278. errorMsg = "the job is not stopped"
  279. break
  280. }
  281. if task.Image == "" || task.GpuQueue == "" || task.Type != models.TypeCloudBrainOne {
  282. log.Error("the job(%s) version is too old", task.JobName, ctx.Data["MsgID"])
  283. resultCode = "-1"
  284. errorMsg = "the job's version is too old and can not be restarted"
  285. break
  286. }
  287. if !ctx.IsSigned || (ctx.User.ID != task.UserID && !ctx.IsUserSiteAdmin()) {
  288. log.Error("the user has no right ro restart the job", task.JobName, ctx.Data["MsgID"])
  289. resultCode = "-1"
  290. errorMsg = "you have no right to restart the job"
  291. break
  292. }
  293. count, err := models.GetCloudbrainCountByUserID(ctx.User.ID, string(models.JobTypeDebug))
  294. if err != nil {
  295. log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])
  296. resultCode = "-1"
  297. errorMsg = "system error"
  298. break
  299. } else {
  300. if count >= 1 {
  301. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  302. resultCode = "-1"
  303. errorMsg = "you have already a running or waiting task, can not create more"
  304. break
  305. }
  306. }
  307. err = cloudbrain.RestartTask(ctx, task, &ID)
  308. if err != nil {
  309. log.Error("RestartTask failed:%v", err.Error(), ctx.Data["MsgID"])
  310. resultCode = "-1"
  311. errorMsg = "system error"
  312. break
  313. }
  314. break
  315. }
  316. ctx.JSON(200, map[string]string{
  317. "result_code": resultCode,
  318. "error_msg": errorMsg,
  319. "status": status,
  320. "id": ID,
  321. })
  322. }
  323. func CloudBrainBenchMarkShow(ctx *context.Context) {
  324. cloudBrainShow(ctx, tplCloudBrainBenchmarkShow, models.JobTypeBenchmark)
  325. }
  326. func CloudBrainShow(ctx *context.Context) {
  327. cloudBrainShow(ctx, tplCloudBrainShow, models.JobTypeDebug)
  328. }
  329. func CloudBrainTrainJobShow(ctx *context.Context) {
  330. cloudBrainShow(ctx, tplCloudBrainTrainJobShow, models.JobTypeTrain)
  331. }
  332. func cloudBrainShow(ctx *context.Context, tpName base.TplName, jobType models.JobType) {
  333. ctx.Data["PageIsCloudBrain"] = true
  334. debugListType := ctx.Query("debugListType")
  335. var task *models.Cloudbrain
  336. var err error
  337. if jobType == models.JobTypeTrain {
  338. task, err = models.GetCloudbrainByJobID(ctx.Params(":jobid"))
  339. } else {
  340. task, err = models.GetCloudbrainByIDWithDeleted(ctx.Params(":id"))
  341. }
  342. if err != nil {
  343. log.Info("error:" + err.Error())
  344. ctx.Data["error"] = err.Error()
  345. return
  346. }
  347. result, err := cloudbrain.GetJob(task.JobID)
  348. if err != nil {
  349. log.Info("error:" + err.Error())
  350. ctx.Data["error"] = err.Error()
  351. return
  352. }
  353. if result != nil {
  354. jobRes, _ := models.ConvertToJobResultPayload(result.Payload)
  355. jobRes.Resource.Memory = strings.ReplaceAll(jobRes.Resource.Memory, "Mi", "MB")
  356. spec := "GPU数:" + strconv.Itoa(jobRes.Resource.NvidiaComGpu) + ",CPU数:" + strconv.Itoa(jobRes.Resource.CPU) + ",内存(MB):" + jobRes.Resource.Memory
  357. ctx.Data["resource_spec"] = spec
  358. if task.JobType == string(models.JobTypeTrain) {
  359. if trainGpuInfos == nil {
  360. json.Unmarshal([]byte(setting.TrainGpuTypes), &trainGpuInfos)
  361. }
  362. for _, resourceType := range trainGpuInfos.GpuInfo {
  363. if resourceType.Queue == jobRes.Config.GpuType {
  364. ctx.Data["resource_type"] = resourceType.Value
  365. }
  366. }
  367. } else {
  368. if gpuInfos == nil {
  369. json.Unmarshal([]byte(setting.GpuTypes), &gpuInfos)
  370. }
  371. for _, resourceType := range gpuInfos.GpuInfo {
  372. if resourceType.Queue == jobRes.Config.GpuType {
  373. ctx.Data["resource_type"] = resourceType.Value
  374. }
  375. }
  376. }
  377. taskRoles := jobRes.TaskRoles
  378. if jobRes.JobStatus.State != string(models.JobFailed) {
  379. taskRes, _ := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
  380. ctx.Data["taskRes"] = taskRes
  381. task.Status = taskRes.TaskStatuses[0].State
  382. task.ContainerID = taskRes.TaskStatuses[0].ContainerID
  383. task.ContainerIp = taskRes.TaskStatuses[0].ContainerIP
  384. models.ParseAndSetDurationFromCloudBrainOne(jobRes, task)
  385. if task.DeletedAt.IsZero() { //normal record
  386. err = models.UpdateJob(task)
  387. if err != nil {
  388. ctx.Data["error"] = err.Error()
  389. return
  390. }
  391. } else { //deleted record
  392. }
  393. } else {
  394. task.Status = jobRes.JobStatus.State
  395. taskRes := models.TaskPod{TaskStatuses: []models.TaskStatuses{
  396. {
  397. State: jobRes.JobStatus.State,
  398. },
  399. }}
  400. ctx.Data["taskRes"] = taskRes
  401. jobRes.JobStatus.StartTime = time.Unix(int64(task.CreatedUnix), 0).Format("2006-01-02 15:04:05")
  402. jobRes.JobStatus.EndTime = time.Unix(int64(task.UpdatedUnix), 0).Format("2006-01-02 15:04:05")
  403. }
  404. ctx.Data["result"] = jobRes
  405. } else {
  406. log.Info("error:" + err.Error())
  407. return
  408. }
  409. user, err := models.GetUserByID(task.UserID)
  410. if err == nil {
  411. task.User = user
  412. }
  413. if task.BenchmarkTypeID > 0 {
  414. for _, benchmarkType := range GetBenchmarkTypes(ctx).BenchmarkType {
  415. if task.BenchmarkTypeID == benchmarkType.Id {
  416. ctx.Data["BenchmarkTypeName"] = benchmarkType.First
  417. for _, benchmarkChildType := range benchmarkType.Second {
  418. if task.BenchmarkChildTypeID == benchmarkChildType.Id {
  419. ctx.Data["BenchmarkChildTypeName"] = benchmarkChildType.Value
  420. break
  421. }
  422. }
  423. break
  424. }
  425. }
  426. }
  427. if task.TrainJobDuration == "" {
  428. if task.Duration == 0 {
  429. var duration int64
  430. if task.Status == string(models.JobRunning) {
  431. duration = time.Now().Unix() - int64(task.CreatedUnix)
  432. } else {
  433. duration = int64(task.UpdatedUnix) - int64(task.CreatedUnix)
  434. }
  435. task.Duration = duration
  436. }
  437. task.TrainJobDuration = models.ConvertDurationToStr(task.Duration)
  438. }
  439. ctx.Data["duration"] = task.TrainJobDuration
  440. if len(task.Parameters) > 0 {
  441. var parameters models.Parameters
  442. err := json.Unmarshal([]byte(task.Parameters), &parameters)
  443. if err != nil {
  444. log.Error("Failed to Unmarshal Parameters: %s (%v)", task.Parameters, err)
  445. task.Parameters = ""
  446. } else {
  447. if len(parameters.Parameter) > 0 {
  448. paramTemp := ""
  449. for _, Parameter := range parameters.Parameter {
  450. param := Parameter.Label + " = " + Parameter.Value + "; "
  451. paramTemp = paramTemp + param
  452. }
  453. task.Parameters = paramTemp[:len(paramTemp)-2]
  454. } else {
  455. task.Parameters = ""
  456. }
  457. }
  458. }
  459. attachment, err := models.GetAttachmentByUUID(task.Uuid)
  460. if err == nil {
  461. ctx.Data["datasetname"] = attachment.Name
  462. } else {
  463. ctx.Data["datasetname"] = ""
  464. }
  465. ctx.Data["task"] = task
  466. ctx.Data["jobName"] = task.JobName
  467. ctx.Data["displayJobName"] = task.DisplayJobName
  468. version_list_task := make([]*models.Cloudbrain, 0)
  469. version_list_task = append(version_list_task, task)
  470. ctx.Data["version_list_task"] = version_list_task
  471. ctx.Data["debugListType"] = debugListType
  472. ctx.Data["code_path"] = cloudbrain.CodeMountPath
  473. ctx.Data["dataset_path"] = cloudbrain.DataSetMountPath
  474. ctx.Data["model_path"] = cloudbrain.ModelMountPath
  475. ctx.Data["canDownload"] = cloudbrain.CanDeleteJob(ctx, task)
  476. ctx.HTML(200, tpName)
  477. }
  478. func CloudBrainDebug(ctx *context.Context) {
  479. task := ctx.Cloudbrain
  480. debugUrl := setting.DebugServerHost + "jpylab_" + task.JobID + "_" + task.SubTaskName
  481. ctx.Redirect(debugUrl)
  482. }
  483. func CloudBrainCommitImage(ctx *context.Context, form auth.CommitImageCloudBrainForm) {
  484. err := cloudbrain.CommitImage(ctx.Cloudbrain.JobID, models.CommitImageParams{
  485. Ip: ctx.Cloudbrain.ContainerIp,
  486. TaskContainerId: ctx.Cloudbrain.ContainerID,
  487. ImageDescription: form.Description,
  488. ImageTag: form.Tag,
  489. })
  490. if err != nil {
  491. log.Error("CommitImage(%s) failed:%v", ctx.Cloudbrain.JobName, err.Error(), ctx.Data["msgID"])
  492. ctx.JSON(200, map[string]string{
  493. "result_code": "-1",
  494. "error_msg": "CommitImage failed",
  495. })
  496. return
  497. }
  498. ctx.JSON(200, map[string]string{
  499. "result_code": "0",
  500. "error_msg": "",
  501. })
  502. }
  503. func CloudBrainStop(ctx *context.Context) {
  504. var ID = ctx.Params(":id")
  505. var resultCode = "0"
  506. var errorMsg = ""
  507. var status = ""
  508. task := ctx.Cloudbrain
  509. for {
  510. if task.Status == string(models.JobStopped) || task.Status == string(models.JobFailed) || task.Status == string(models.JobSucceeded) {
  511. log.Error("the job(%s) has been stopped", task.JobName, ctx.Data["msgID"])
  512. resultCode = "-1"
  513. errorMsg = "system error"
  514. break
  515. }
  516. err := cloudbrain.StopJob(task.JobID)
  517. if err != nil {
  518. log.Error("StopJob(%s) failed:%v", task.JobName, err, ctx.Data["msgID"])
  519. resultCode = "-1"
  520. errorMsg = "system error"
  521. break
  522. }
  523. task.Status = string(models.JobStopped)
  524. if task.EndTime == 0 {
  525. task.EndTime = timeutil.TimeStampNow()
  526. }
  527. task.ComputeAndSetDuration()
  528. err = models.UpdateJob(task)
  529. if err != nil {
  530. log.Error("UpdateJob(%s) failed:%v", task.JobName, err, ctx.Data["msgID"])
  531. resultCode = "-1"
  532. errorMsg = "system error"
  533. break
  534. }
  535. status = task.Status
  536. break
  537. }
  538. ctx.JSON(200, map[string]interface{}{
  539. "result_code": resultCode,
  540. "error_msg": errorMsg,
  541. "status": status,
  542. "id": ID,
  543. "StatusOK": 0,
  544. })
  545. }
  546. func StopJobsByUserID(userID int64) {
  547. cloudBrains, err := models.GetCloudbrainsNeededStopByUserID(userID)
  548. if err != nil {
  549. log.Warn("Failed to get cloudBrain info", err)
  550. return
  551. }
  552. StopJobs(cloudBrains)
  553. }
  554. func StopJobsByRepoID(repoID int64) {
  555. cloudBrains, err := models.GetCloudbrainsNeededStopByRepoID(repoID)
  556. if err != nil {
  557. log.Warn("Failed to get cloudBrain info", err)
  558. return
  559. }
  560. StopJobs(cloudBrains)
  561. }
  562. /**
  563. */
  564. func StopJobs(cloudBrains []*models.Cloudbrain) {
  565. for _, taskInfo := range cloudBrains {
  566. if taskInfo.Type == models.TypeCloudBrainOne {
  567. err := retry(3, time.Second*30, func() error {
  568. return cloudbrain.StopJob(taskInfo.JobID)
  569. })
  570. logErrorAndUpdateJobStatus(err, taskInfo)
  571. } else {
  572. if taskInfo.JobType == string(models.JobTypeTrain) {
  573. err := retry(3, time.Second*30, func() error {
  574. _, err := modelarts.StopTrainJob(taskInfo.JobID, strconv.FormatInt(taskInfo.VersionID, 10))
  575. return err
  576. })
  577. logErrorAndUpdateJobStatus(err, taskInfo)
  578. } else {
  579. param := models.NotebookAction{
  580. Action: models.ActionStop,
  581. }
  582. err := retry(3, time.Second*30, func() error {
  583. _, err := modelarts.ManageNotebook(taskInfo.JobID, param)
  584. return err
  585. })
  586. logErrorAndUpdateJobStatus(err, taskInfo)
  587. }
  588. }
  589. }
  590. }
  591. func retry(attempts int, sleep time.Duration, f func() error) (err error) {
  592. for i := 0; i < attempts; i++ {
  593. if i > 0 {
  594. log.Warn("retrying after error:", err)
  595. time.Sleep(sleep)
  596. }
  597. err = f()
  598. if err == nil {
  599. return nil
  600. }
  601. }
  602. return fmt.Errorf("after %d attempts, last error: %s", attempts, err)
  603. }
  604. func logErrorAndUpdateJobStatus(err error, taskInfo *models.Cloudbrain) {
  605. if err != nil {
  606. log.Warn("Failed to stop cloudBrain job:"+taskInfo.JobID, err)
  607. } else {
  608. taskInfo.Status = string(models.JobStopped)
  609. if taskInfo.EndTime == 0 {
  610. taskInfo.EndTime = timeutil.TimeStampNow()
  611. }
  612. taskInfo.ComputeAndSetDuration()
  613. err = models.UpdateJob(taskInfo)
  614. if err != nil {
  615. log.Warn("UpdateJob failed", err)
  616. }
  617. }
  618. }
  619. func CloudBrainDel(ctx *context.Context) {
  620. var listType = ctx.Query("debugListType")
  621. if err := deleteCloudbrainJob(ctx); err != nil {
  622. log.Error("deleteCloudbrainJob failed: %v", err, ctx.Data["msgID"])
  623. ctx.ServerError(err.Error(), err)
  624. return
  625. }
  626. var isAdminPage = ctx.Query("isadminpage")
  627. if ctx.IsUserSiteAdmin() && isAdminPage == "true" {
  628. ctx.Redirect(setting.AppSubURL + "/admin" + "/cloudbrains")
  629. } else {
  630. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=" + listType)
  631. }
  632. }
  633. func deleteCloudbrainJob(ctx *context.Context) error {
  634. task := ctx.Cloudbrain
  635. if task.Status != string(models.JobStopped) && task.Status != string(models.JobFailed) && task.Status != string(models.JobSucceeded) {
  636. log.Error("the job(%s) has not been stopped", task.JobName, ctx.Data["msgID"])
  637. return errors.New("the job has not been stopped")
  638. }
  639. err := models.DeleteJob(task)
  640. if err != nil {
  641. log.Error("DeleteJob failed: %v", err, ctx.Data["msgID"])
  642. return err
  643. }
  644. deleteJobStorage(task.JobName, models.TypeCloudBrainOne)
  645. return nil
  646. }
  647. func CloudBrainShowModels(ctx *context.Context) {
  648. ctx.Data["PageIsCloudBrain"] = true
  649. ID := ctx.Params(":id")
  650. parentDir := ctx.Query("parentDir")
  651. dirArray := strings.Split(parentDir, "/")
  652. task, err := models.GetCloudbrainByID(ID)
  653. if err != nil {
  654. log.Error("no such job!", ctx.Data["msgID"])
  655. ctx.ServerError("no such job:", err)
  656. return
  657. }
  658. //get dirs
  659. dirs, err := GetModelDirs(task.JobName, parentDir)
  660. if err != nil {
  661. log.Error("GetModelDirs failed:%v", err.Error(), ctx.Data["msgID"])
  662. ctx.ServerError("GetModelDirs failed:", err)
  663. return
  664. }
  665. var fileInfos []storage.FileInfo
  666. err = json.Unmarshal([]byte(dirs), &fileInfos)
  667. if err != nil {
  668. log.Error("json.Unmarshal failed:%v", err.Error(), ctx.Data["msgID"])
  669. ctx.ServerError("json.Unmarshal failed:", err)
  670. return
  671. }
  672. for i, fileInfo := range fileInfos {
  673. temp, _ := time.Parse("2006-01-02 15:04:05", fileInfo.ModTime)
  674. fileInfos[i].ModTime = temp.Local().Format("2006-01-02 15:04:05")
  675. }
  676. sort.Slice(fileInfos, func(i, j int) bool {
  677. return fileInfos[i].ModTime > fileInfos[j].ModTime
  678. })
  679. ctx.Data["Path"] = dirArray
  680. ctx.Data["Dirs"] = fileInfos
  681. ctx.Data["task"] = task
  682. ctx.Data["ID"] = ID
  683. ctx.HTML(200, tplCloudBrainShowModels)
  684. }
  685. func GetPublicImages(ctx *context.Context) {
  686. getImages(ctx, cloudbrain.Public)
  687. }
  688. func GetCustomImages(ctx *context.Context) {
  689. getImages(ctx, cloudbrain.Custom)
  690. }
  691. func getImages(ctx *context.Context, imageType string) {
  692. log.Info("Get images begin")
  693. page := ctx.QueryInt("page")
  694. size := ctx.QueryInt("size")
  695. name := ctx.Query("name")
  696. getImagesResult, err := cloudbrain.GetImagesPageable(page, size, imageType, name)
  697. if err != nil {
  698. log.Error("Can not get images:%v", err)
  699. ctx.JSON(http.StatusOK, models.GetImagesPayload{
  700. Count: 0,
  701. TotalPages: 0,
  702. ImageInfo: []*models.ImageInfo{},
  703. })
  704. } else {
  705. ctx.JSON(http.StatusOK, getImagesResult.Payload)
  706. }
  707. log.Info("Get images end")
  708. }
  709. func GetModelDirs(jobName string, parentDir string) (string, error) {
  710. var req string
  711. modelActualPath := storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/")
  712. if parentDir == "" {
  713. req = "baseDir=" + modelActualPath
  714. } else {
  715. req = "baseDir=" + modelActualPath + "&parentDir=" + parentDir
  716. }
  717. return getDirs(req)
  718. }
  719. func CloudBrainDownloadModel(ctx *context.Context) {
  720. parentDir := ctx.Query("parentDir")
  721. fileName := ctx.Query("fileName")
  722. jobName := ctx.Query("jobName")
  723. filePath := "jobs/" + jobName + "/model/" + parentDir
  724. url, err := storage.Attachments.PresignedGetURL(filePath, fileName)
  725. if err != nil {
  726. log.Error("PresignedGetURL failed: %v", err.Error(), ctx.Data["msgID"])
  727. ctx.ServerError("PresignedGetURL", err)
  728. return
  729. }
  730. http.Redirect(ctx.Resp, ctx.Req.Request, url, http.StatusMovedPermanently)
  731. }
  732. func GetRate(ctx *context.Context) {
  733. isObjectDetcionAll := ctx.QueryBool("isObjectDetcionAll")
  734. if isObjectDetcionAll {
  735. ctx.Redirect(setting.BenchmarkServerHost + "?username=admin")
  736. return
  737. }
  738. var ID = ctx.Params(":id")
  739. job, err := models.GetCloudbrainByID(ID)
  740. if err != nil {
  741. ctx.ServerError("GetCloudbrainByJobID failed", err)
  742. return
  743. }
  744. if job.JobType == string(models.JobTypeBenchmark) {
  745. log.Info("url=" + setting.BenchmarkServerHost + "?username=" + ctx.User.Name)
  746. ctx.Redirect(setting.BenchmarkServerHost + "?username=" + ctx.User.Name)
  747. } else if job.JobType == string(models.JobTypeSnn4imagenet) {
  748. ctx.Redirect(setting.Snn4imagenetServerHost)
  749. } else if job.JobType == string(models.JobTypeBrainScore) {
  750. ctx.Redirect(setting.BrainScoreServerHost)
  751. } else {
  752. log.Error("JobType error:%s", job.JobType, ctx.Data["msgID"])
  753. }
  754. }
  755. func downloadCode(repo *models.Repository, codePath, branchName string) error {
  756. if err := git.Clone(repo.RepoPath(), codePath, git.CloneRepoOptions{Branch: branchName}); err != nil {
  757. log.Error("Failed to clone repository: %s (%v)", repo.FullName(), err)
  758. return err
  759. }
  760. configFile, err := os.OpenFile(codePath+"/.git/config", os.O_RDWR, 0666)
  761. if err != nil {
  762. log.Error("open file(%s) failed:%v", codePath+"/,git/config", err)
  763. return err
  764. }
  765. defer configFile.Close()
  766. pos := int64(0)
  767. reader := bufio.NewReader(configFile)
  768. for {
  769. line, err := reader.ReadString('\n')
  770. if err != nil {
  771. if err == io.EOF {
  772. log.Error("not find the remote-url")
  773. return nil
  774. } else {
  775. log.Error("read error: %v", err)
  776. return err
  777. }
  778. }
  779. if strings.Contains(line, "url") && strings.Contains(line, ".git") {
  780. originUrl := "\turl = " + repo.CloneLink().HTTPS + "\n"
  781. if len(line) > len(originUrl) {
  782. originUrl += strings.Repeat(" ", len(line)-len(originUrl))
  783. }
  784. bytes := []byte(originUrl)
  785. _, err := configFile.WriteAt(bytes, pos)
  786. if err != nil {
  787. log.Error("WriteAt failed:%v", err)
  788. return err
  789. }
  790. break
  791. }
  792. pos += int64(len(line))
  793. }
  794. return nil
  795. }
  796. func downloadRateCode(repo *models.Repository, taskName, rateOwnerName, rateRepoName, codePath, benchmarkCategory, gpuType string) error {
  797. err := os.MkdirAll(codePath, os.ModePerm)
  798. if err != nil {
  799. log.Error("mkdir codePath failed", err.Error())
  800. return err
  801. }
  802. repoExt, err := models.GetRepositoryByOwnerAndName(rateOwnerName, rateRepoName)
  803. if err != nil {
  804. log.Error("GetRepositoryByOwnerAndName(%s) failed", rateRepoName, err.Error())
  805. return err
  806. }
  807. if err := git.Clone(repoExt.RepoPath(), codePath, git.CloneRepoOptions{}); err != nil {
  808. log.Error("Failed to clone repository: %s (%v)", repoExt.FullName(), err)
  809. return err
  810. }
  811. fileName := codePath + cloudbrain.TaskInfoName
  812. f, err := os.OpenFile(fileName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, os.ModePerm)
  813. if err != nil {
  814. log.Error("OpenFile failed", err.Error())
  815. return err
  816. }
  817. defer f.Close()
  818. data, err := json.Marshal(models.TaskInfo{
  819. Username: repo.Owner.Name,
  820. TaskName: taskName,
  821. CodeName: repo.Name,
  822. BenchmarkCategory: strings.Split(benchmarkCategory, ","),
  823. CodeLink: strings.TrimSuffix(repo.CloneLink().HTTPS, ".git"),
  824. GpuType: gpuType,
  825. })
  826. if err != nil {
  827. log.Error("json.Marshal failed", err.Error())
  828. return err
  829. }
  830. _, err = f.Write(data)
  831. if err != nil {
  832. log.Error("WriteString failed", err.Error())
  833. return err
  834. }
  835. return nil
  836. }
  837. func uploadCodeToMinio(codePath, jobName, parentDir string) error {
  838. files, err := readDir(codePath)
  839. if err != nil {
  840. log.Error("readDir(%s) failed: %s", codePath, err.Error())
  841. return err
  842. }
  843. for _, file := range files {
  844. if file.IsDir() {
  845. if err = uploadCodeToMinio(codePath+file.Name()+"/", jobName, parentDir+file.Name()+"/"); err != nil {
  846. log.Error("uploadCodeToMinio(%s) failed: %s", file.Name(), err.Error())
  847. return err
  848. }
  849. } else {
  850. destObject := setting.CBCodePathPrefix + jobName + parentDir + file.Name()
  851. sourceFile := codePath + file.Name()
  852. err = storage.Attachments.UploadObject(destObject, sourceFile)
  853. if err != nil {
  854. log.Error("UploadObject(%s) failed: %s", file.Name(), err.Error())
  855. return err
  856. }
  857. }
  858. }
  859. return nil
  860. }
  861. func mkModelPath(modelPath string) error {
  862. err := os.MkdirAll(modelPath, os.ModePerm)
  863. if err != nil {
  864. log.Error("MkdirAll(%s) failed:%v", modelPath, err)
  865. return err
  866. }
  867. fileName := modelPath + "README"
  868. f, err := os.OpenFile(fileName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, os.ModePerm)
  869. if err != nil {
  870. log.Error("OpenFile failed", err.Error())
  871. return err
  872. }
  873. defer f.Close()
  874. _, err = f.WriteString("You can put the model file into this directory and download it by the web page.")
  875. if err != nil {
  876. log.Error("WriteString failed", err.Error())
  877. return err
  878. }
  879. return nil
  880. }
  881. func deleteJobStorage(jobName string, cloudbrainType int) error {
  882. //delete local
  883. localJobPath := setting.JobPath + jobName
  884. err := os.RemoveAll(localJobPath)
  885. if err != nil {
  886. log.Error("RemoveAll(%s) failed:%v", localJobPath, err)
  887. }
  888. //delete oss
  889. if cloudbrainType == models.TypeCloudBrainOne {
  890. dirPath := setting.CBCodePathPrefix + jobName + "/"
  891. err = storage.Attachments.DeleteDir(dirPath)
  892. if err != nil {
  893. log.Error("DeleteDir(%s) failed:%v", localJobPath, err)
  894. }
  895. } else if cloudbrainType == models.TypeCloudBrainTwo {
  896. //dirPath := setting.CodePathPrefix + jobName + "/"
  897. //err = storage.ObsRemoveObject(setting.Bucket, dirPath)
  898. //if err != nil {
  899. // log.Error("ObsRemoveObject(%s) failed:%v", localJobPath, err)
  900. //}
  901. log.Info("no need to delete")
  902. } else {
  903. log.Error("cloudbrainType(%d) error", cloudbrainType)
  904. }
  905. return nil
  906. }
  907. func SyncCloudbrainStatus() {
  908. cloudBrains, err := models.GetCloudBrainUnStoppedJob()
  909. if err != nil {
  910. log.Error("GetCloudBrainUnStoppedJob failed:", err.Error())
  911. return
  912. }
  913. for _, task := range cloudBrains {
  914. if task.Type == models.TypeCloudBrainOne {
  915. result, err := cloudbrain.GetJob(task.JobID)
  916. if err != nil {
  917. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  918. continue
  919. }
  920. if result != nil {
  921. jobRes, _ := models.ConvertToJobResultPayload(result.Payload)
  922. taskRoles := jobRes.TaskRoles
  923. taskRes, _ := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
  924. task.Status = taskRes.TaskStatuses[0].State
  925. if task.Status != string(models.JobWaiting) {
  926. models.ParseAndSetDurationFromCloudBrainOne(jobRes, task)
  927. err = models.UpdateJob(task)
  928. if err != nil {
  929. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  930. }
  931. var maxDuration int64
  932. if task.JobType == string(models.JobTypeBenchmark) {
  933. maxDuration = setting.BenchmarkMaxDuration
  934. } else {
  935. maxDuration = setting.MaxDuration
  936. }
  937. if task.Duration >= maxDuration {
  938. log.Info("begin to stop job(%s), because of the duration", task.JobName)
  939. err = cloudbrain.StopJob(task.JobID)
  940. if err != nil {
  941. log.Error("StopJob(%s) failed:%v", task.JobName, err)
  942. continue
  943. }
  944. task.Status = string(models.JobStopped)
  945. if task.EndTime == 0 {
  946. task.EndTime = timeutil.TimeStampNow()
  947. }
  948. task.ComputeAndSetDuration()
  949. err = models.UpdateJob(task)
  950. if err != nil {
  951. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  952. }
  953. }
  954. }
  955. }
  956. } else if task.Type == models.TypeCloudBrainTwo {
  957. if task.JobType == string(models.JobTypeDebug) {
  958. //result, err := modelarts.GetJob(task.JobID)
  959. result, err := modelarts.GetNotebook2(task.JobID)
  960. if err != nil {
  961. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  962. continue
  963. }
  964. if result != nil {
  965. task.Status = result.Status
  966. if task.StartTime == 0 && result.Lease.UpdateTime > 0 {
  967. task.StartTime = timeutil.TimeStamp(result.Lease.UpdateTime / 1000)
  968. }
  969. if task.EndTime == 0 && models.IsModelArtsDebugJobTerminal(task.Status) {
  970. task.EndTime = timeutil.TimeStampNow()
  971. }
  972. task.ComputeAndSetDuration()
  973. err = models.UpdateJob(task)
  974. if err != nil {
  975. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  976. continue
  977. }
  978. }
  979. } else if task.JobType == string(models.JobTypeTrain) || task.JobType == string(models.JobTypeInference) {
  980. result, err := modelarts.GetTrainJob(task.JobID, strconv.FormatInt(task.VersionID, 10))
  981. if err != nil {
  982. log.Error("GetTrainJob(%s) failed:%v", task.JobName, err)
  983. continue
  984. }
  985. if result != nil {
  986. task.Status = modelarts.TransTrainJobStatus(result.IntStatus)
  987. task.Duration = result.Duration / 1000
  988. task.TrainJobDuration = result.TrainJobDuration
  989. if task.StartTime == 0 && result.StartTime > 0 {
  990. task.StartTime = timeutil.TimeStamp(result.StartTime / 1000)
  991. }
  992. task.TrainJobDuration = models.ConvertDurationToStr(task.Duration)
  993. if task.EndTime == 0 && models.IsTrainJobTerminal(task.Status) && task.StartTime > 0 {
  994. task.EndTime = task.StartTime.Add(task.Duration)
  995. }
  996. err = models.UpdateJob(task)
  997. if err != nil {
  998. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  999. continue
  1000. }
  1001. }
  1002. } else {
  1003. log.Error("task.JobType(%s) is error:%s", task.JobName, task.JobType)
  1004. }
  1005. } else {
  1006. log.Error("task.Type(%s) is error:%d", task.JobName, task.Type)
  1007. }
  1008. }
  1009. return
  1010. }
  1011. func HandleTaskWithNoDuration(ctx *context.Context) {
  1012. log.Info("HandleTaskWithNoDuration start")
  1013. count := 0
  1014. start := time.Now().Unix()
  1015. for {
  1016. cloudBrains, err := models.GetStoppedJobWithNoDurationJob()
  1017. if err != nil {
  1018. log.Error("HandleTaskWithNoTrainJobDuration failed:", err.Error())
  1019. break
  1020. }
  1021. if len(cloudBrains) == 0 {
  1022. log.Info("HandleTaskWithNoTrainJobDuration:no task need handle")
  1023. break
  1024. }
  1025. handleNoDurationTask(cloudBrains)
  1026. count += len(cloudBrains)
  1027. if len(cloudBrains) < 100 {
  1028. log.Info("HandleTaskWithNoTrainJobDuration:task less than 100")
  1029. break
  1030. }
  1031. if time.Now().Unix()-start > 600 {
  1032. log.Info("HandleTaskWithNoDuration : time out")
  1033. ctx.JSON(200, fmt.Sprintf("task stop for time out,count=%d", count))
  1034. return
  1035. }
  1036. }
  1037. log.Info("HandleTaskWithNoTrainJobDuration:count=%d", count)
  1038. ctx.JSON(200, fmt.Sprintf("success,count=%d", count))
  1039. }
  1040. func handleNoDurationTask(cloudBrains []*models.Cloudbrain) {
  1041. for _, task := range cloudBrains {
  1042. time.Sleep(time.Millisecond * 100)
  1043. log.Info("Handle job ,%+v", task)
  1044. if task.Type == models.TypeCloudBrainOne {
  1045. result, err := cloudbrain.GetJob(task.JobID)
  1046. if err != nil {
  1047. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  1048. updateDefaultDuration(task)
  1049. continue
  1050. }
  1051. if result != nil {
  1052. if result.Msg != "success" {
  1053. updateDefaultDuration(task)
  1054. continue
  1055. }
  1056. jobRes, err := models.ConvertToJobResultPayload(result.Payload)
  1057. if err != nil || len(jobRes.TaskRoles) == 0 {
  1058. updateDefaultDuration(task)
  1059. continue
  1060. }
  1061. taskRoles := jobRes.TaskRoles
  1062. taskRes, err := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
  1063. if err != nil || len(taskRes.TaskStatuses) == 0 {
  1064. updateDefaultDuration(task)
  1065. continue
  1066. }
  1067. task.Status = taskRes.TaskStatuses[0].State
  1068. log.Info("task startTime = %v endTime= %v ,jobId=%d", jobRes.JobStatus.StartTime, jobRes.JobStatus.EndTime, task.ID)
  1069. if jobRes.JobStatus.CreatedTime > 0 {
  1070. task.StartTime = timeutil.TimeStamp(jobRes.JobStatus.CreatedTime / 1000)
  1071. if jobRes.JobStatus.CompletedTime > 0 {
  1072. task.EndTime = timeutil.TimeStamp(jobRes.JobStatus.CompletedTime / 1000)
  1073. } else {
  1074. task.EndTime = task.UpdatedUnix
  1075. }
  1076. } else {
  1077. task.StartTime = 0
  1078. task.EndTime = 0
  1079. }
  1080. if task.EndTime < task.StartTime {
  1081. log.Info("endTime[%v] is less than starTime[%v],jobId=%d", task.EndTime, task.StartTime, task.ID)
  1082. st := task.StartTime
  1083. task.StartTime = task.EndTime
  1084. task.EndTime = st
  1085. }
  1086. task.Duration = task.EndTime.AsTime().Unix() - task.StartTime.AsTime().Unix()
  1087. task.TrainJobDuration = models.ConvertDurationToStr(task.Duration)
  1088. err = models.UpdateJob(task)
  1089. if err != nil {
  1090. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  1091. }
  1092. }
  1093. } else if task.Type == models.TypeCloudBrainTwo {
  1094. if task.JobType == string(models.JobTypeDebug) {
  1095. //result, err := modelarts.GetJob(task.JobID)
  1096. result, err := modelarts.GetNotebook2(task.JobID)
  1097. if err != nil {
  1098. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  1099. updateDefaultDuration(task)
  1100. continue
  1101. }
  1102. if result != nil {
  1103. task.Status = result.Status
  1104. startTime := result.Lease.CreateTime
  1105. duration := result.Lease.Duration / 1000
  1106. if startTime > 0 {
  1107. task.StartTime = timeutil.TimeStamp(startTime / 1000)
  1108. task.EndTime = task.StartTime.Add(duration)
  1109. }
  1110. task.ComputeAndSetDuration()
  1111. err = models.UpdateJob(task)
  1112. if err != nil {
  1113. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  1114. continue
  1115. }
  1116. }
  1117. } else if task.JobType == string(models.JobTypeTrain) || task.JobType == string(models.JobTypeInference) {
  1118. result, err := modelarts.GetTrainJob(task.JobID, strconv.FormatInt(task.VersionID, 10))
  1119. if err != nil {
  1120. log.Error("GetTrainJob(%s) failed:%v", task.JobName, err)
  1121. updateDefaultDuration(task)
  1122. continue
  1123. }
  1124. if result != nil {
  1125. startTime := result.StartTime / 1000
  1126. if startTime > 0 {
  1127. task.StartTime = timeutil.TimeStamp(startTime)
  1128. task.EndTime = task.StartTime.Add(result.Duration / 1000)
  1129. }
  1130. task.ComputeAndSetDuration()
  1131. err = models.UpdateJob(task)
  1132. if err != nil {
  1133. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  1134. continue
  1135. }
  1136. }
  1137. } else {
  1138. log.Error("task.JobType(%s) is error:%s", task.JobName, task.JobType)
  1139. }
  1140. } else {
  1141. log.Error("task.Type(%s) is error:%d", task.JobName, task.Type)
  1142. }
  1143. }
  1144. }
  1145. func updateDefaultDuration(task *models.Cloudbrain) {
  1146. log.Info("updateDefaultDuration: taskId=%d", task.ID)
  1147. task.StartTime = task.CreatedUnix
  1148. task.EndTime = task.UpdatedUnix
  1149. task.ComputeAndSetDuration()
  1150. err := models.UpdateJob(task)
  1151. if err != nil {
  1152. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  1153. }
  1154. }
  1155. func CloudBrainBenchmarkIndex(ctx *context.Context) {
  1156. MustEnableCloudbrain(ctx)
  1157. repo := ctx.Repo.Repository
  1158. page := ctx.QueryInt("page")
  1159. if page <= 0 {
  1160. page = 1
  1161. }
  1162. var jobTypes []string
  1163. jobTypes = append(jobTypes, string(models.JobTypeBenchmark))
  1164. ciTasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  1165. ListOptions: models.ListOptions{
  1166. Page: page,
  1167. PageSize: setting.UI.IssuePagingNum,
  1168. },
  1169. RepoID: repo.ID,
  1170. Type: models.TypeCloudBrainOne,
  1171. JobTypes: jobTypes,
  1172. })
  1173. if err != nil {
  1174. ctx.ServerError("Get debugjob faild:", err)
  1175. return
  1176. }
  1177. for i, task := range ciTasks {
  1178. ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  1179. ciTasks[i].Cloudbrain.ComputeResource = task.ComputeResource
  1180. if ciTasks[i].TrainJobDuration == "" {
  1181. if ciTasks[i].Duration == 0 {
  1182. var duration int64
  1183. if task.Status == string(models.JobRunning) {
  1184. duration = time.Now().Unix() - int64(task.Cloudbrain.CreatedUnix)
  1185. } else {
  1186. duration = int64(task.Cloudbrain.UpdatedUnix) - int64(task.Cloudbrain.CreatedUnix)
  1187. }
  1188. ciTasks[i].Duration = duration
  1189. }
  1190. ciTasks[i].TrainJobDuration = models.ConvertDurationToStr(ciTasks[i].Duration)
  1191. }
  1192. ciTasks[i].BenchmarkTypeName = ""
  1193. if task.BenchmarkTypeID > 0 {
  1194. for _, benchmarkType := range GetBenchmarkTypes(ctx).BenchmarkType {
  1195. if task.BenchmarkTypeID == benchmarkType.Id {
  1196. ciTasks[i].BenchmarkTypeRankLink = benchmarkType.RankLink
  1197. ciTasks[i].BenchmarkTypeName = benchmarkType.First
  1198. break
  1199. }
  1200. }
  1201. }
  1202. }
  1203. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  1204. ctx.Data["Page"] = pager
  1205. ctx.Data["PageIsCloudBrain"] = true
  1206. ctx.Data["Tasks"] = ciTasks
  1207. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  1208. ctx.Data["RepoIsEmpty"] = repo.IsEmpty
  1209. ctx.HTML(200, tplCloudBrainBenchmarkIndex)
  1210. }
  1211. func GetChildTypes(ctx *context.Context) {
  1212. benchmarkTypeID := ctx.QueryInt("benchmark_type_id")
  1213. re := make(map[string]interface{})
  1214. for {
  1215. var isExist bool
  1216. for _, benchmarkType := range GetBenchmarkTypes(ctx).BenchmarkType {
  1217. if benchmarkTypeID == benchmarkType.Id {
  1218. isExist = true
  1219. re["child_types"] = benchmarkType.Second
  1220. re["result_code"] = "0"
  1221. break
  1222. }
  1223. }
  1224. if !isExist {
  1225. re["result_code"] = "1"
  1226. log.Error("no such benchmark_type_id", ctx.Data["MsgID"])
  1227. re["errMsg"] = "system error"
  1228. break
  1229. }
  1230. break
  1231. }
  1232. ctx.JSON(200, re)
  1233. }
  1234. func CloudBrainBenchmarkNew(ctx *context.Context) {
  1235. ctx.Data["description"] = ""
  1236. ctx.Data["benchmarkTypeID"] = -1
  1237. ctx.Data["benchmark_child_types_id_hidden"] = -1
  1238. err := cloudBrainNewDataPrepare(ctx)
  1239. if err != nil {
  1240. ctx.ServerError("get new cloudbrain info failed", err)
  1241. return
  1242. }
  1243. ctx.HTML(200, tplCloudBrainBenchmarkNew)
  1244. }
  1245. func getBenchmarkAttachment(benchmarkTypeID, benchmarkChildTypeID int, ctx *context.Context) (*models.BenchmarkDataset, error) {
  1246. var childInfo *models.BenchmarkDataset
  1247. var isExist bool
  1248. for _, benchmarkType := range GetBenchmarkTypes(ctx).BenchmarkType {
  1249. if benchmarkType.Id == benchmarkTypeID {
  1250. for _, childType := range benchmarkType.Second {
  1251. if childType.Id == benchmarkChildTypeID {
  1252. childInfo = childType
  1253. isExist = true
  1254. break
  1255. }
  1256. }
  1257. break
  1258. }
  1259. }
  1260. if !isExist {
  1261. log.Error("no such benchmark_type_id&benchmark_child_type_id")
  1262. return childInfo, errors.New("no such benchmark_type_id&benchmark_child_type_id")
  1263. }
  1264. return childInfo, nil
  1265. }
  1266. func getBenchmarkGpuQueue(gpuQueue string) (string, error) {
  1267. queue := ""
  1268. if benchmarkGpuInfos == nil {
  1269. if err := json.Unmarshal([]byte(setting.BenchmarkGpuTypes), &benchmarkGpuInfos); err != nil {
  1270. log.Error("json.Unmarshal BenchmarkGpuTypes(%s) failed:%v", setting.BenchmarkGpuTypes, err)
  1271. return queue, err
  1272. }
  1273. }
  1274. var isExist bool
  1275. for _, gpuInfo := range benchmarkGpuInfos.GpuInfo {
  1276. if gpuQueue == gpuInfo.Queue {
  1277. isExist = true
  1278. queue = gpuQueue
  1279. break
  1280. }
  1281. }
  1282. if !isExist {
  1283. log.Error("no such gpuQueue, %s", gpuQueue)
  1284. return queue, errors.New("no such gpuQueue")
  1285. }
  1286. return queue, nil
  1287. }
  1288. func getBenchmarkResourceSpec(resourceSpecID int) (int, error) {
  1289. var id int
  1290. if benchmarkResourceSpecs == nil {
  1291. if err := json.Unmarshal([]byte(setting.BenchmarkResourceSpecs), &benchmarkResourceSpecs); err != nil {
  1292. log.Error("json.Unmarshal BenchmarkResourceSpecs(%s) failed:%v", setting.BenchmarkResourceSpecs, err)
  1293. return id, err
  1294. }
  1295. }
  1296. var isExist bool
  1297. for _, resourceSpec := range benchmarkResourceSpecs.ResourceSpec {
  1298. if resourceSpecID == resourceSpec.Id {
  1299. isExist = true
  1300. id = resourceSpecID
  1301. break
  1302. }
  1303. }
  1304. if !isExist {
  1305. log.Error("no such resourceSpecID, %d", resourceSpecID)
  1306. return id, errors.New("no such resourceSpec")
  1307. }
  1308. return id, nil
  1309. }
  1310. func CloudBrainBenchmarkCreate(ctx *context.Context, form auth.CreateCloudBrainForm) {
  1311. ctx.Data["PageIsCloudBrain"] = true
  1312. displayJobName := form.DisplayJobName
  1313. jobName := util.ConvertDisplayJobNameToJobName(displayJobName)
  1314. image := form.Image
  1315. gpuQueue := form.GpuType
  1316. command := cloudbrain.CommandBenchmark
  1317. codePath := setting.JobPath + jobName + cloudbrain.CodeMountPath
  1318. resourceSpecId := cloudbrain.BenchMarkResourceID
  1319. benchmarkTypeID := form.BenchmarkTypeID
  1320. benchmarkChildTypeID := form.BenchmarkChildTypeID
  1321. ctx.Data["description"] = form.Description
  1322. ctx.Data["benchmarkTypeID"] = benchmarkTypeID
  1323. ctx.Data["benchmark_child_types_id_hidden"] = benchmarkChildTypeID
  1324. repo := ctx.Repo.Repository
  1325. tasks, err := models.GetCloudbrainsByDisplayJobName(repo.ID, string(models.JobTypeBenchmark), displayJobName)
  1326. if err == nil {
  1327. if len(tasks) != 0 {
  1328. log.Error("the job name did already exist", ctx.Data["MsgID"])
  1329. cloudBrainNewDataPrepare(ctx)
  1330. ctx.RenderWithErr("the job name did already exist", tplCloudBrainBenchmarkNew, &form)
  1331. return
  1332. }
  1333. } else {
  1334. if !models.IsErrJobNotExist(err) {
  1335. log.Error("system error, %v", err, ctx.Data["MsgID"])
  1336. cloudBrainNewDataPrepare(ctx)
  1337. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1338. return
  1339. }
  1340. }
  1341. if !jobNamePattern.MatchString(jobName) {
  1342. cloudBrainNewDataPrepare(ctx)
  1343. ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_jobname_err"), tplCloudBrainBenchmarkNew, &form)
  1344. return
  1345. }
  1346. childInfo, err := getBenchmarkAttachment(benchmarkTypeID, benchmarkChildTypeID, ctx)
  1347. if err != nil {
  1348. log.Error("getBenchmarkAttachment failed:%v", err, ctx.Data["MsgID"])
  1349. cloudBrainNewDataPrepare(ctx)
  1350. ctx.RenderWithErr("benchmark type error", tplCloudBrainBenchmarkNew, &form)
  1351. return
  1352. }
  1353. _, err = getBenchmarkGpuQueue(gpuQueue)
  1354. if err != nil {
  1355. log.Error("getBenchmarkGpuQueue failed:%v", err, ctx.Data["MsgID"])
  1356. cloudBrainNewDataPrepare(ctx)
  1357. ctx.RenderWithErr("gpu queue error", tplCloudBrainBenchmarkNew, &form)
  1358. return
  1359. }
  1360. _, err = getBenchmarkResourceSpec(resourceSpecId)
  1361. if err != nil {
  1362. log.Error("getBenchmarkResourceSpec failed:%v", err, ctx.Data["MsgID"])
  1363. cloudBrainNewDataPrepare(ctx)
  1364. ctx.RenderWithErr("resource spec error", tplCloudBrainBenchmarkNew, &form)
  1365. return
  1366. }
  1367. count, err := models.GetCloudbrainCountByUserID(ctx.User.ID, string(models.JobTypeBenchmark))
  1368. if err != nil {
  1369. log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])
  1370. cloudBrainNewDataPrepare(ctx)
  1371. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1372. return
  1373. } else {
  1374. if count >= 1 {
  1375. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  1376. cloudBrainNewDataPrepare(ctx)
  1377. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplCloudBrainBenchmarkNew, &form)
  1378. return
  1379. }
  1380. }
  1381. os.RemoveAll(codePath)
  1382. if err := downloadCode(repo, codePath, cloudbrain.DefaultBranchName); err != nil {
  1383. log.Error("downloadCode failed, %v", err, ctx.Data["MsgID"])
  1384. cloudBrainNewDataPrepare(ctx)
  1385. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1386. return
  1387. }
  1388. if _, err := os.Stat(codePath + "/train.py"); err != nil {
  1389. if os.IsNotExist(err) {
  1390. // file does not exist
  1391. log.Error("train.py does not exist, %v", err, ctx.Data["MsgID"])
  1392. cloudBrainNewDataPrepare(ctx)
  1393. ctx.RenderWithErr("train.py does not exist", tplCloudBrainBenchmarkNew, &form)
  1394. } else {
  1395. log.Error("Stat failed, %v", err, ctx.Data["MsgID"])
  1396. cloudBrainNewDataPrepare(ctx)
  1397. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1398. }
  1399. return
  1400. } else if _, err := os.Stat(codePath + "/test.py"); err != nil {
  1401. if os.IsNotExist(err) {
  1402. // file does not exist
  1403. log.Error("test.py does not exist, %v", err, ctx.Data["MsgID"])
  1404. cloudBrainNewDataPrepare(ctx)
  1405. ctx.RenderWithErr("test.py does not exist", tplCloudBrainBenchmarkNew, &form)
  1406. } else {
  1407. log.Error("Stat failed, %v", err, ctx.Data["MsgID"])
  1408. cloudBrainNewDataPrepare(ctx)
  1409. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1410. }
  1411. return
  1412. }
  1413. if err := uploadCodeToMinio(codePath+"/", jobName, cloudbrain.CodeMountPath+"/"); err != nil {
  1414. log.Error("uploadCodeToMinio failed, %v", err, ctx.Data["MsgID"])
  1415. cloudBrainNewDataPrepare(ctx)
  1416. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1417. return
  1418. }
  1419. benchmarkPath := setting.JobPath + jobName + cloudbrain.BenchMarkMountPath
  1420. var gpuType string
  1421. for _, gpuInfo := range gpuInfos.GpuInfo {
  1422. if gpuInfo.Queue == gpuQueue {
  1423. gpuType = gpuInfo.Value
  1424. }
  1425. }
  1426. if err := downloadRateCode(repo, jobName, childInfo.Owner, childInfo.RepoName, benchmarkPath, form.BenchmarkCategory, gpuType); err != nil {
  1427. log.Error("downloadRateCode failed, %v", err, ctx.Data["MsgID"])
  1428. //cloudBrainNewDataPrepare(ctx)
  1429. //ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1430. //return
  1431. }
  1432. if err := uploadCodeToMinio(benchmarkPath+"/", jobName, cloudbrain.BenchMarkMountPath+"/"); err != nil {
  1433. log.Error("uploadCodeToMinio failed, %v", err, ctx.Data["MsgID"])
  1434. //cloudBrainNewDataPrepare(ctx)
  1435. //ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1436. //return
  1437. }
  1438. err = cloudbrain.GenerateTask(ctx, displayJobName, jobName, image, command, childInfo.Attachment, storage.GetMinioPath(jobName, cloudbrain.CodeMountPath+"/"),
  1439. storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/"),
  1440. storage.GetMinioPath(jobName, cloudbrain.BenchMarkMountPath+"/"), storage.GetMinioPath(jobName, cloudbrain.Snn4imagenetMountPath+"/"),
  1441. storage.GetMinioPath(jobName, cloudbrain.BrainScoreMountPath+"/"), string(models.JobTypeBenchmark), gpuQueue, form.Description, cloudbrain.DefaultBranchName, "", "",
  1442. benchmarkTypeID, benchmarkChildTypeID, resourceSpecId)
  1443. if err != nil {
  1444. cloudBrainNewDataPrepare(ctx)
  1445. ctx.RenderWithErr(err.Error(), tplCloudBrainBenchmarkNew, &form)
  1446. return
  1447. }
  1448. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/cloudbrain/benchmark")
  1449. }
  1450. func BenchmarkDel(ctx *context.Context) {
  1451. if err := deleteCloudbrainJob(ctx); err != nil {
  1452. log.Error("deleteCloudbrainJob failed: %v", err, ctx.Data["msgID"])
  1453. ctx.ServerError(err.Error(), err)
  1454. return
  1455. }
  1456. var isAdminPage = ctx.Query("isadminpage")
  1457. if ctx.IsUserSiteAdmin() && isAdminPage == "true" {
  1458. ctx.Redirect(setting.AppSubURL + "/admin" + "/cloudbrains")
  1459. } else {
  1460. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/cloudbrain/benchmark")
  1461. }
  1462. }
  1463. func CloudBrainTrainJobNew(ctx *context.Context) {
  1464. err := cloudBrainNewDataPrepare(ctx)
  1465. if err != nil {
  1466. ctx.ServerError("get new train-job info failed", err)
  1467. return
  1468. }
  1469. ctx.HTML(http.StatusOK, tplCloudBrainTrainJobNew)
  1470. }
  1471. func getTrainJobCommand(form auth.CreateCloudBrainForm) (string, error) {
  1472. var command string
  1473. bootFile := form.BootFile
  1474. params := form.Params
  1475. if !strings.HasSuffix(bootFile, ".py") {
  1476. log.Error("bootFile(%s) format error", bootFile)
  1477. return command, errors.New("bootFile format error")
  1478. }
  1479. var parameters models.Parameters
  1480. var param string
  1481. if len(params) != 0 {
  1482. err := json.Unmarshal([]byte(params), &parameters)
  1483. if err != nil {
  1484. log.Error("Failed to Unmarshal params: %s (%v)", params, err)
  1485. return command, err
  1486. }
  1487. for _, parameter := range parameters.Parameter {
  1488. param += " --" + parameter.Label + "=" + parameter.Value
  1489. }
  1490. }
  1491. command += "python /code/" + bootFile + param + " > " + cloudbrain.ModelMountPath + "/" + form.DisplayJobName + "-" + cloudbrain.LogFile
  1492. return command, nil
  1493. }
  1494. func CloudBrainTrainJobDel(ctx *context.Context) {
  1495. var listType = ctx.Query("listType")
  1496. if err := deleteCloudbrainJob(ctx); err != nil {
  1497. log.Error("deleteCloudbrainJob failed: %v", err, ctx.Data["msgID"])
  1498. ctx.ServerError(err.Error(), err)
  1499. return
  1500. }
  1501. var isAdminPage = ctx.Query("isadminpage")
  1502. if ctx.IsUserSiteAdmin() && isAdminPage == "true" {
  1503. ctx.Redirect(setting.AppSubURL + "/admin" + "/cloudbrains")
  1504. } else {
  1505. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job?listType=" + listType)
  1506. }
  1507. }
  1508. func GetBenchmarkTypes(ctx *context.Context) *models.BenchmarkTypes {
  1509. var lang = ctx.Locale.Language()
  1510. if benchmarkTypesMap[lang] == nil {
  1511. var val = i18n.Tr(lang, BENCHMARK_TYPE_CODE)
  1512. //use config
  1513. val = setting.BenchmarkTypes
  1514. var tempType *models.BenchmarkTypes
  1515. if err := json.Unmarshal([]byte(val), &tempType); err != nil {
  1516. log.Error("json.Unmarshal BenchmarkTypes(%s) failed:%v", val, err, ctx.Data["MsgID"])
  1517. return &models.BenchmarkTypes{}
  1518. }
  1519. benchmarkTypesMap[lang] = tempType
  1520. }
  1521. return benchmarkTypesMap[lang]
  1522. }