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 55 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
4 years ago
5 years ago
4 years ago
4 years ago
5 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
3 years ago
3 years ago
4 years ago
3 years ago
3 years ago
5 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
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 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
4 years ago
4 years ago
5 years ago
3 years ago
4 years ago
4 years ago
5 years ago
3 years ago
4 years ago
5 years ago
4 years ago
3 years ago
3 years ago
4 years ago
4 years ago
4 years ago
5 years ago
4 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
4 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
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
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
3 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
4 years ago
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
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
5 years ago
4 years ago
4 years ago
5 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
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
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780
  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. var hasSameResource bool
  294. if gpuInfos == nil {
  295. json.Unmarshal([]byte(setting.GpuTypes), &gpuInfos)
  296. }
  297. for _, resourceType := range gpuInfos.GpuInfo {
  298. if resourceType.Queue == task.GpuQueue {
  299. hasSameResource = true
  300. continue
  301. }
  302. }
  303. if !hasSameResource {
  304. log.Error("has no same resource, can not restart", ctx.Data["MsgID"])
  305. resultCode = "-1"
  306. errorMsg = "the job's version is too old and can not be restarted"
  307. break
  308. }
  309. count, err := models.GetCloudbrainCountByUserID(ctx.User.ID, string(models.JobTypeDebug))
  310. if err != nil {
  311. log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])
  312. resultCode = "-1"
  313. errorMsg = "system error"
  314. break
  315. } else {
  316. if count >= 1 {
  317. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  318. resultCode = "-1"
  319. errorMsg = "you have already a running or waiting task, can not create more"
  320. break
  321. }
  322. }
  323. err = cloudbrain.RestartTask(ctx, task, &ID)
  324. if err != nil {
  325. log.Error("RestartTask failed:%v", err.Error(), ctx.Data["MsgID"])
  326. resultCode = "-1"
  327. errorMsg = "system error"
  328. break
  329. }
  330. break
  331. }
  332. ctx.JSON(200, map[string]string{
  333. "result_code": resultCode,
  334. "error_msg": errorMsg,
  335. "status": status,
  336. "id": ID,
  337. })
  338. }
  339. func CloudBrainBenchMarkShow(ctx *context.Context) {
  340. cloudBrainShow(ctx, tplCloudBrainBenchmarkShow, models.JobTypeBenchmark)
  341. }
  342. func CloudBrainShow(ctx *context.Context) {
  343. cloudBrainShow(ctx, tplCloudBrainShow, models.JobTypeDebug)
  344. }
  345. func CloudBrainTrainJobShow(ctx *context.Context) {
  346. cloudBrainShow(ctx, tplCloudBrainTrainJobShow, models.JobTypeTrain)
  347. }
  348. func cloudBrainShow(ctx *context.Context, tpName base.TplName, jobType models.JobType) {
  349. ctx.Data["PageIsCloudBrain"] = true
  350. debugListType := ctx.Query("debugListType")
  351. var task *models.Cloudbrain
  352. var err error
  353. if jobType == models.JobTypeTrain {
  354. task, err = models.GetCloudbrainByJobID(ctx.Params(":jobid"))
  355. } else {
  356. task, err = models.GetCloudbrainByIDWithDeleted(ctx.Params(":id"))
  357. }
  358. if err != nil {
  359. log.Info("error:" + err.Error())
  360. ctx.Data["error"] = err.Error()
  361. return
  362. }
  363. result, err := cloudbrain.GetJob(task.JobID)
  364. if err != nil {
  365. log.Info("error:" + err.Error())
  366. ctx.Data["error"] = err.Error()
  367. return
  368. }
  369. if cloudbrain.ResourceSpecs == nil {
  370. json.Unmarshal([]byte(setting.ResourceSpecs), &cloudbrain.ResourceSpecs)
  371. }
  372. for _, tmp := range cloudbrain.ResourceSpecs.ResourceSpec {
  373. if tmp.Id == task.ResourceSpecId {
  374. ctx.Data["GpuNum"] = tmp.GpuNum
  375. ctx.Data["CpuNum"] = tmp.CpuNum
  376. ctx.Data["MemMiB"] = tmp.MemMiB
  377. ctx.Data["ShareMemMiB"] = tmp.ShareMemMiB
  378. }
  379. }
  380. if result != nil {
  381. jobRes, _ := models.ConvertToJobResultPayload(result.Payload)
  382. jobRes.Resource.Memory = strings.ReplaceAll(jobRes.Resource.Memory, "Mi", "MB")
  383. spec := "GPU数:" + strconv.Itoa(jobRes.Resource.NvidiaComGpu) + ",CPU数:" + strconv.Itoa(jobRes.Resource.CPU) + ",内存(MB):" + jobRes.Resource.Memory
  384. ctx.Data["resource_spec"] = spec
  385. if task.JobType == string(models.JobTypeTrain) {
  386. if trainGpuInfos == nil {
  387. json.Unmarshal([]byte(setting.TrainGpuTypes), &trainGpuInfos)
  388. }
  389. for _, resourceType := range trainGpuInfos.GpuInfo {
  390. if resourceType.Queue == jobRes.Config.GpuType {
  391. ctx.Data["resource_type"] = resourceType.Value
  392. }
  393. }
  394. } else {
  395. if gpuInfos == nil {
  396. json.Unmarshal([]byte(setting.GpuTypes), &gpuInfos)
  397. }
  398. for _, resourceType := range gpuInfos.GpuInfo {
  399. if resourceType.Queue == jobRes.Config.GpuType {
  400. ctx.Data["resource_type"] = resourceType.Value
  401. }
  402. }
  403. }
  404. taskRoles := jobRes.TaskRoles
  405. if jobRes.JobStatus.State != string(models.JobFailed) {
  406. taskRes, _ := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
  407. ctx.Data["taskRes"] = taskRes
  408. task.Status = taskRes.TaskStatuses[0].State
  409. task.ContainerID = taskRes.TaskStatuses[0].ContainerID
  410. task.ContainerIp = taskRes.TaskStatuses[0].ContainerIP
  411. models.ParseAndSetDurationFromCloudBrainOne(jobRes, task)
  412. if task.DeletedAt.IsZero() { //normal record
  413. err = models.UpdateJob(task)
  414. if err != nil {
  415. ctx.Data["error"] = err.Error()
  416. return
  417. }
  418. } else { //deleted record
  419. }
  420. } else {
  421. task.Status = jobRes.JobStatus.State
  422. taskRes := models.TaskPod{TaskStatuses: []models.TaskStatuses{
  423. {
  424. State: jobRes.JobStatus.State,
  425. },
  426. }}
  427. ctx.Data["taskRes"] = taskRes
  428. jobRes.JobStatus.StartTime = time.Unix(int64(task.CreatedUnix), 0).Format("2006-01-02 15:04:05")
  429. jobRes.JobStatus.EndTime = time.Unix(int64(task.UpdatedUnix), 0).Format("2006-01-02 15:04:05")
  430. }
  431. ctx.Data["result"] = jobRes
  432. } else {
  433. log.Info("error:" + err.Error())
  434. return
  435. }
  436. user, err := models.GetUserByID(task.UserID)
  437. if err == nil {
  438. task.User = user
  439. }
  440. if task.BenchmarkTypeID > 0 {
  441. for _, benchmarkType := range GetBenchmarkTypes(ctx).BenchmarkType {
  442. if task.BenchmarkTypeID == benchmarkType.Id {
  443. ctx.Data["BenchmarkTypeName"] = benchmarkType.First
  444. for _, benchmarkChildType := range benchmarkType.Second {
  445. if task.BenchmarkChildTypeID == benchmarkChildType.Id {
  446. ctx.Data["BenchmarkChildTypeName"] = benchmarkChildType.Value
  447. break
  448. }
  449. }
  450. break
  451. }
  452. }
  453. }
  454. if task.TrainJobDuration == "" {
  455. if task.Duration == 0 {
  456. var duration int64
  457. if task.Status == string(models.JobRunning) {
  458. duration = time.Now().Unix() - int64(task.CreatedUnix)
  459. } else {
  460. duration = int64(task.UpdatedUnix) - int64(task.CreatedUnix)
  461. }
  462. task.Duration = duration
  463. }
  464. task.TrainJobDuration = models.ConvertDurationToStr(task.Duration)
  465. }
  466. ctx.Data["duration"] = task.TrainJobDuration
  467. if len(task.Parameters) > 0 {
  468. var parameters models.Parameters
  469. err := json.Unmarshal([]byte(task.Parameters), &parameters)
  470. if err != nil {
  471. log.Error("Failed to Unmarshal Parameters: %s (%v)", task.Parameters, err)
  472. task.Parameters = ""
  473. } else {
  474. if len(parameters.Parameter) > 0 {
  475. paramTemp := ""
  476. for _, Parameter := range parameters.Parameter {
  477. param := Parameter.Label + " = " + Parameter.Value + "; "
  478. paramTemp = paramTemp + param
  479. }
  480. task.Parameters = paramTemp[:len(paramTemp)-2]
  481. } else {
  482. task.Parameters = ""
  483. }
  484. }
  485. }
  486. attachment, err := models.GetAttachmentByUUID(task.Uuid)
  487. if err == nil {
  488. ctx.Data["datasetname"] = attachment.Name
  489. } else {
  490. ctx.Data["datasetname"] = ""
  491. }
  492. ctx.Data["task"] = task
  493. ctx.Data["jobName"] = task.JobName
  494. ctx.Data["displayJobName"] = task.DisplayJobName
  495. version_list_task := make([]*models.Cloudbrain, 0)
  496. version_list_task = append(version_list_task, task)
  497. ctx.Data["version_list_task"] = version_list_task
  498. ctx.Data["debugListType"] = debugListType
  499. ctx.Data["code_path"] = cloudbrain.CodeMountPath
  500. ctx.Data["dataset_path"] = cloudbrain.DataSetMountPath
  501. ctx.Data["model_path"] = cloudbrain.ModelMountPath
  502. ctx.Data["canDownload"] = cloudbrain.CanModifyJob(ctx, task)
  503. ctx.HTML(200, tpName)
  504. }
  505. func CloudBrainDebug(ctx *context.Context) {
  506. task := ctx.Cloudbrain
  507. debugUrl := setting.DebugServerHost + "jpylab_" + task.JobID + "_" + task.SubTaskName
  508. ctx.Redirect(debugUrl)
  509. }
  510. func CloudBrainCommitImage(ctx *context.Context, form auth.CommitImageCloudBrainForm) {
  511. err := cloudbrain.CommitImage(ctx.Cloudbrain.JobID, models.CommitImageParams{
  512. Ip: ctx.Cloudbrain.ContainerIp,
  513. TaskContainerId: ctx.Cloudbrain.ContainerID,
  514. ImageDescription: form.Description,
  515. ImageTag: form.Tag,
  516. })
  517. if err != nil {
  518. log.Error("CommitImage(%s) failed:%v", ctx.Cloudbrain.JobName, err.Error(), ctx.Data["msgID"])
  519. ctx.JSON(200, map[string]string{
  520. "result_code": "-1",
  521. "error_msg": "CommitImage failed",
  522. })
  523. return
  524. }
  525. ctx.JSON(200, map[string]string{
  526. "result_code": "0",
  527. "error_msg": "",
  528. })
  529. }
  530. func CloudBrainStop(ctx *context.Context) {
  531. var ID = ctx.Params(":id")
  532. var resultCode = "0"
  533. var errorMsg = ""
  534. var status = ""
  535. task := ctx.Cloudbrain
  536. for {
  537. if task.Status == string(models.JobStopped) || task.Status == string(models.JobFailed) || task.Status == string(models.JobSucceeded) {
  538. log.Error("the job(%s) has been stopped", task.JobName, ctx.Data["msgID"])
  539. resultCode = "-1"
  540. errorMsg = "system error"
  541. break
  542. }
  543. err := cloudbrain.StopJob(task.JobID)
  544. if err != nil {
  545. log.Error("StopJob(%s) failed:%v", task.JobName, err, ctx.Data["msgID"])
  546. resultCode = "-1"
  547. errorMsg = "system error"
  548. break
  549. }
  550. task.Status = string(models.JobStopped)
  551. if task.EndTime == 0 {
  552. task.EndTime = timeutil.TimeStampNow()
  553. }
  554. task.ComputeAndSetDuration()
  555. err = models.UpdateJob(task)
  556. if err != nil {
  557. log.Error("UpdateJob(%s) failed:%v", task.JobName, err, ctx.Data["msgID"])
  558. resultCode = "-1"
  559. errorMsg = "system error"
  560. break
  561. }
  562. status = task.Status
  563. break
  564. }
  565. ctx.JSON(200, map[string]interface{}{
  566. "result_code": resultCode,
  567. "error_msg": errorMsg,
  568. "status": status,
  569. "id": ID,
  570. "StatusOK": 0,
  571. })
  572. }
  573. func StopJobsByUserID(userID int64) {
  574. cloudBrains, err := models.GetCloudbrainsNeededStopByUserID(userID)
  575. if err != nil {
  576. log.Warn("Failed to get cloudBrain info", err)
  577. return
  578. }
  579. StopJobs(cloudBrains)
  580. }
  581. func StopJobsByRepoID(repoID int64) {
  582. cloudBrains, err := models.GetCloudbrainsNeededStopByRepoID(repoID)
  583. if err != nil {
  584. log.Warn("Failed to get cloudBrain info", err)
  585. return
  586. }
  587. StopJobs(cloudBrains)
  588. }
  589. /**
  590. */
  591. func StopJobs(cloudBrains []*models.Cloudbrain) {
  592. for _, taskInfo := range cloudBrains {
  593. if taskInfo.Type == models.TypeCloudBrainOne {
  594. err := retry(3, time.Second*30, func() error {
  595. return cloudbrain.StopJob(taskInfo.JobID)
  596. })
  597. logErrorAndUpdateJobStatus(err, taskInfo)
  598. } else {
  599. if taskInfo.JobType == string(models.JobTypeTrain) {
  600. err := retry(3, time.Second*30, func() error {
  601. _, err := modelarts.StopTrainJob(taskInfo.JobID, strconv.FormatInt(taskInfo.VersionID, 10))
  602. return err
  603. })
  604. logErrorAndUpdateJobStatus(err, taskInfo)
  605. } else {
  606. param := models.NotebookAction{
  607. Action: models.ActionStop,
  608. }
  609. err := retry(3, time.Second*30, func() error {
  610. _, err := modelarts.ManageNotebook(taskInfo.JobID, param)
  611. return err
  612. })
  613. logErrorAndUpdateJobStatus(err, taskInfo)
  614. }
  615. }
  616. }
  617. }
  618. func retry(attempts int, sleep time.Duration, f func() error) (err error) {
  619. for i := 0; i < attempts; i++ {
  620. if i > 0 {
  621. log.Warn("retrying after error:", err)
  622. time.Sleep(sleep)
  623. }
  624. err = f()
  625. if err == nil {
  626. return nil
  627. }
  628. }
  629. return fmt.Errorf("after %d attempts, last error: %s", attempts, err)
  630. }
  631. func logErrorAndUpdateJobStatus(err error, taskInfo *models.Cloudbrain) {
  632. if err != nil {
  633. log.Warn("Failed to stop cloudBrain job:"+taskInfo.JobID, err)
  634. } else {
  635. taskInfo.Status = string(models.JobStopped)
  636. if taskInfo.EndTime == 0 {
  637. taskInfo.EndTime = timeutil.TimeStampNow()
  638. }
  639. taskInfo.ComputeAndSetDuration()
  640. err = models.UpdateJob(taskInfo)
  641. if err != nil {
  642. log.Warn("UpdateJob failed", err)
  643. }
  644. }
  645. }
  646. func CloudBrainDel(ctx *context.Context) {
  647. var listType = ctx.Query("debugListType")
  648. if err := deleteCloudbrainJob(ctx); err != nil {
  649. log.Error("deleteCloudbrainJob failed: %v", err, ctx.Data["msgID"])
  650. ctx.ServerError(err.Error(), err)
  651. return
  652. }
  653. var isAdminPage = ctx.Query("isadminpage")
  654. var isHomePage = ctx.Query("isHomepage")
  655. if ctx.IsUserSiteAdmin() && isAdminPage == "true" {
  656. ctx.Redirect(setting.AppSubURL + "/admin" + "/cloudbrains")
  657. } else if isHomePage == "true" {
  658. ctx.Redirect(setting.AppSubURL + "/cloudbrains")
  659. } else {
  660. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=" + listType)
  661. }
  662. }
  663. func deleteCloudbrainJob(ctx *context.Context) error {
  664. task := ctx.Cloudbrain
  665. if task.Status != string(models.JobStopped) && task.Status != string(models.JobFailed) && task.Status != string(models.JobSucceeded) {
  666. log.Error("the job(%s) has not been stopped", task.JobName, ctx.Data["msgID"])
  667. return errors.New("the job has not been stopped")
  668. }
  669. err := models.DeleteJob(task)
  670. if err != nil {
  671. log.Error("DeleteJob failed: %v", err, ctx.Data["msgID"])
  672. return err
  673. }
  674. deleteJobStorage(task.JobName, models.TypeCloudBrainOne)
  675. return nil
  676. }
  677. func CloudBrainShowModels(ctx *context.Context) {
  678. ctx.Data["PageIsCloudBrain"] = true
  679. ID := ctx.Params(":id")
  680. parentDir := ctx.Query("parentDir")
  681. dirArray := strings.Split(parentDir, "/")
  682. task, err := models.GetCloudbrainByID(ID)
  683. if err != nil {
  684. log.Error("no such job!", ctx.Data["msgID"])
  685. ctx.ServerError("no such job:", err)
  686. return
  687. }
  688. //get dirs
  689. dirs, err := GetModelDirs(task.JobName, parentDir)
  690. if err != nil {
  691. log.Error("GetModelDirs failed:%v", err.Error(), ctx.Data["msgID"])
  692. ctx.ServerError("GetModelDirs failed:", err)
  693. return
  694. }
  695. var fileInfos []storage.FileInfo
  696. err = json.Unmarshal([]byte(dirs), &fileInfos)
  697. if err != nil {
  698. log.Error("json.Unmarshal failed:%v", err.Error(), ctx.Data["msgID"])
  699. ctx.ServerError("json.Unmarshal failed:", err)
  700. return
  701. }
  702. for i, fileInfo := range fileInfos {
  703. temp, _ := time.Parse("2006-01-02 15:04:05", fileInfo.ModTime)
  704. fileInfos[i].ModTime = temp.Local().Format("2006-01-02 15:04:05")
  705. }
  706. sort.Slice(fileInfos, func(i, j int) bool {
  707. return fileInfos[i].ModTime > fileInfos[j].ModTime
  708. })
  709. ctx.Data["Path"] = dirArray
  710. ctx.Data["Dirs"] = fileInfos
  711. ctx.Data["task"] = task
  712. ctx.Data["ID"] = ID
  713. ctx.HTML(200, tplCloudBrainShowModels)
  714. }
  715. func GetPublicImages(ctx *context.Context) {
  716. getImages(ctx, cloudbrain.Public)
  717. }
  718. func GetCustomImages(ctx *context.Context) {
  719. getImages(ctx, cloudbrain.Custom)
  720. }
  721. func getImages(ctx *context.Context, imageType string) {
  722. log.Info("Get images begin")
  723. page := ctx.QueryInt("page")
  724. size := ctx.QueryInt("size")
  725. name := ctx.Query("name")
  726. getImagesResult, err := cloudbrain.GetImagesPageable(page, size, imageType, name)
  727. if err != nil {
  728. log.Error("Can not get images:%v", err)
  729. ctx.JSON(http.StatusOK, models.GetImagesPayload{
  730. Count: 0,
  731. TotalPages: 0,
  732. ImageInfo: []*models.ImageInfo{},
  733. })
  734. } else {
  735. ctx.JSON(http.StatusOK, getImagesResult.Payload)
  736. }
  737. log.Info("Get images end")
  738. }
  739. func GetModelDirs(jobName string, parentDir string) (string, error) {
  740. var req string
  741. modelActualPath := storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/")
  742. if parentDir == "" {
  743. req = "baseDir=" + modelActualPath
  744. } else {
  745. req = "baseDir=" + modelActualPath + "&parentDir=" + parentDir
  746. }
  747. return getDirs(req)
  748. }
  749. func CloudBrainDownloadModel(ctx *context.Context) {
  750. parentDir := ctx.Query("parentDir")
  751. fileName := ctx.Query("fileName")
  752. jobName := ctx.Query("jobName")
  753. filePath := "jobs/" + jobName + "/model/" + parentDir
  754. url, err := storage.Attachments.PresignedGetURL(filePath, fileName)
  755. if err != nil {
  756. log.Error("PresignedGetURL failed: %v", err.Error(), ctx.Data["msgID"])
  757. ctx.ServerError("PresignedGetURL", err)
  758. return
  759. }
  760. http.Redirect(ctx.Resp, ctx.Req.Request, url, http.StatusMovedPermanently)
  761. }
  762. func GetRate(ctx *context.Context) {
  763. isObjectDetcionAll := ctx.QueryBool("isObjectDetcionAll")
  764. if isObjectDetcionAll {
  765. ctx.Redirect(setting.BenchmarkServerHost + "?username=admin")
  766. return
  767. }
  768. var ID = ctx.Params(":id")
  769. job, err := models.GetCloudbrainByID(ID)
  770. if err != nil {
  771. ctx.ServerError("GetCloudbrainByJobID failed", err)
  772. return
  773. }
  774. if job.JobType == string(models.JobTypeBenchmark) {
  775. log.Info("url=" + setting.BenchmarkServerHost + "?username=" + ctx.User.Name)
  776. ctx.Redirect(setting.BenchmarkServerHost + "?username=" + ctx.User.Name)
  777. } else if job.JobType == string(models.JobTypeSnn4imagenet) {
  778. ctx.Redirect(setting.Snn4imagenetServerHost)
  779. } else if job.JobType == string(models.JobTypeBrainScore) {
  780. ctx.Redirect(setting.BrainScoreServerHost)
  781. } else {
  782. log.Error("JobType error:%s", job.JobType, ctx.Data["msgID"])
  783. }
  784. }
  785. func downloadCode(repo *models.Repository, codePath, branchName string) error {
  786. if err := git.Clone(repo.RepoPath(), codePath, git.CloneRepoOptions{Branch: branchName}); err != nil {
  787. log.Error("Failed to clone repository: %s (%v)", repo.FullName(), err)
  788. return err
  789. }
  790. configFile, err := os.OpenFile(codePath+"/.git/config", os.O_RDWR, 0666)
  791. if err != nil {
  792. log.Error("open file(%s) failed:%v", codePath+"/,git/config", err)
  793. return err
  794. }
  795. defer configFile.Close()
  796. pos := int64(0)
  797. reader := bufio.NewReader(configFile)
  798. for {
  799. line, err := reader.ReadString('\n')
  800. if err != nil {
  801. if err == io.EOF {
  802. log.Error("not find the remote-url")
  803. return nil
  804. } else {
  805. log.Error("read error: %v", err)
  806. return err
  807. }
  808. }
  809. if strings.Contains(line, "url") && strings.Contains(line, ".git") {
  810. originUrl := "\turl = " + repo.CloneLink().HTTPS + "\n"
  811. if len(line) > len(originUrl) {
  812. originUrl += strings.Repeat(" ", len(line)-len(originUrl))
  813. }
  814. bytes := []byte(originUrl)
  815. _, err := configFile.WriteAt(bytes, pos)
  816. if err != nil {
  817. log.Error("WriteAt failed:%v", err)
  818. return err
  819. }
  820. break
  821. }
  822. pos += int64(len(line))
  823. }
  824. return nil
  825. }
  826. func downloadRateCode(repo *models.Repository, taskName, rateOwnerName, rateRepoName, codePath, benchmarkCategory, gpuType string) error {
  827. err := os.MkdirAll(codePath, os.ModePerm)
  828. if err != nil {
  829. log.Error("mkdir codePath failed", err.Error())
  830. return err
  831. }
  832. repoExt, err := models.GetRepositoryByOwnerAndName(rateOwnerName, rateRepoName)
  833. if err != nil {
  834. log.Error("GetRepositoryByOwnerAndName(%s) failed", rateRepoName, err.Error())
  835. return err
  836. }
  837. if err := git.Clone(repoExt.RepoPath(), codePath, git.CloneRepoOptions{}); err != nil {
  838. log.Error("Failed to clone repository: %s (%v)", repoExt.FullName(), err)
  839. return err
  840. }
  841. fileName := codePath + cloudbrain.TaskInfoName
  842. f, err := os.OpenFile(fileName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, os.ModePerm)
  843. if err != nil {
  844. log.Error("OpenFile failed", err.Error())
  845. return err
  846. }
  847. defer f.Close()
  848. data, err := json.Marshal(models.TaskInfo{
  849. Username: repo.Owner.Name,
  850. TaskName: taskName,
  851. CodeName: repo.Name,
  852. BenchmarkCategory: strings.Split(benchmarkCategory, ","),
  853. CodeLink: strings.TrimSuffix(repo.CloneLink().HTTPS, ".git"),
  854. GpuType: gpuType,
  855. })
  856. if err != nil {
  857. log.Error("json.Marshal failed", err.Error())
  858. return err
  859. }
  860. _, err = f.Write(data)
  861. if err != nil {
  862. log.Error("WriteString failed", err.Error())
  863. return err
  864. }
  865. return nil
  866. }
  867. func uploadCodeToMinio(codePath, jobName, parentDir string) error {
  868. files, err := readDir(codePath)
  869. if err != nil {
  870. log.Error("readDir(%s) failed: %s", codePath, err.Error())
  871. return err
  872. }
  873. for _, file := range files {
  874. if file.IsDir() {
  875. if err = uploadCodeToMinio(codePath+file.Name()+"/", jobName, parentDir+file.Name()+"/"); err != nil {
  876. log.Error("uploadCodeToMinio(%s) failed: %s", file.Name(), err.Error())
  877. return err
  878. }
  879. } else {
  880. destObject := setting.CBCodePathPrefix + jobName + parentDir + file.Name()
  881. sourceFile := codePath + file.Name()
  882. err = storage.Attachments.UploadObject(destObject, sourceFile)
  883. if err != nil {
  884. log.Error("UploadObject(%s) failed: %s", file.Name(), err.Error())
  885. return err
  886. }
  887. }
  888. }
  889. return nil
  890. }
  891. func mkModelPath(modelPath string) error {
  892. err := os.MkdirAll(modelPath, os.ModePerm)
  893. if err != nil {
  894. log.Error("MkdirAll(%s) failed:%v", modelPath, err)
  895. return err
  896. }
  897. fileName := modelPath + "README"
  898. f, err := os.OpenFile(fileName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, os.ModePerm)
  899. if err != nil {
  900. log.Error("OpenFile failed", err.Error())
  901. return err
  902. }
  903. defer f.Close()
  904. _, err = f.WriteString("You can put the model file into this directory and download it by the web page.")
  905. if err != nil {
  906. log.Error("WriteString failed", err.Error())
  907. return err
  908. }
  909. return nil
  910. }
  911. func deleteJobStorage(jobName string, cloudbrainType int) error {
  912. //delete local
  913. localJobPath := setting.JobPath + jobName
  914. err := os.RemoveAll(localJobPath)
  915. if err != nil {
  916. log.Error("RemoveAll(%s) failed:%v", localJobPath, err)
  917. }
  918. //delete oss
  919. if cloudbrainType == models.TypeCloudBrainOne {
  920. dirPath := setting.CBCodePathPrefix + jobName + "/"
  921. err = storage.Attachments.DeleteDir(dirPath)
  922. if err != nil {
  923. log.Error("DeleteDir(%s) failed:%v", localJobPath, err)
  924. }
  925. } else if cloudbrainType == models.TypeCloudBrainTwo {
  926. //dirPath := setting.CodePathPrefix + jobName + "/"
  927. //err = storage.ObsRemoveObject(setting.Bucket, dirPath)
  928. //if err != nil {
  929. // log.Error("ObsRemoveObject(%s) failed:%v", localJobPath, err)
  930. //}
  931. log.Info("no need to delete")
  932. } else {
  933. log.Error("cloudbrainType(%d) error", cloudbrainType)
  934. }
  935. return nil
  936. }
  937. func SyncCloudbrainStatus() {
  938. cloudBrains, err := models.GetCloudBrainUnStoppedJob()
  939. if err != nil {
  940. log.Error("GetCloudBrainUnStoppedJob failed:", err.Error())
  941. return
  942. }
  943. for _, task := range cloudBrains {
  944. if task.Type == models.TypeCloudBrainOne {
  945. result, err := cloudbrain.GetJob(task.JobID)
  946. if err != nil {
  947. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  948. continue
  949. }
  950. if result != nil {
  951. jobRes, _ := models.ConvertToJobResultPayload(result.Payload)
  952. taskRoles := jobRes.TaskRoles
  953. taskRes, _ := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
  954. task.Status = taskRes.TaskStatuses[0].State
  955. if task.Status != string(models.JobWaiting) {
  956. models.ParseAndSetDurationFromCloudBrainOne(jobRes, task)
  957. err = models.UpdateJob(task)
  958. if err != nil {
  959. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  960. }
  961. var maxDuration int64
  962. if task.JobType == string(models.JobTypeBenchmark) {
  963. maxDuration = setting.BenchmarkMaxDuration
  964. } else {
  965. maxDuration = setting.MaxDuration
  966. }
  967. if task.Duration >= maxDuration {
  968. log.Info("begin to stop job(%s), because of the duration", task.JobName)
  969. err = cloudbrain.StopJob(task.JobID)
  970. if err != nil {
  971. log.Error("StopJob(%s) failed:%v", task.JobName, err)
  972. continue
  973. }
  974. task.Status = string(models.JobStopped)
  975. if task.EndTime == 0 {
  976. task.EndTime = timeutil.TimeStampNow()
  977. }
  978. task.ComputeAndSetDuration()
  979. err = models.UpdateJob(task)
  980. if err != nil {
  981. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  982. }
  983. }
  984. }
  985. }
  986. } else if task.Type == models.TypeCloudBrainTwo {
  987. if task.JobType == string(models.JobTypeDebug) {
  988. //result, err := modelarts.GetJob(task.JobID)
  989. result, err := modelarts.GetNotebook2(task.JobID)
  990. if err != nil {
  991. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  992. continue
  993. }
  994. if result != nil {
  995. task.Status = result.Status
  996. if task.StartTime == 0 && result.Lease.UpdateTime > 0 {
  997. task.StartTime = timeutil.TimeStamp(result.Lease.UpdateTime / 1000)
  998. }
  999. if task.EndTime == 0 && models.IsModelArtsDebugJobTerminal(task.Status) {
  1000. task.EndTime = timeutil.TimeStampNow()
  1001. }
  1002. task.ComputeAndSetDuration()
  1003. err = models.UpdateJob(task)
  1004. if err != nil {
  1005. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  1006. continue
  1007. }
  1008. }
  1009. } else if task.JobType == string(models.JobTypeTrain) || task.JobType == string(models.JobTypeInference) {
  1010. result, err := modelarts.GetTrainJob(task.JobID, strconv.FormatInt(task.VersionID, 10))
  1011. if err != nil {
  1012. log.Error("GetTrainJob(%s) failed:%v", task.JobName, err)
  1013. continue
  1014. }
  1015. if result != nil {
  1016. task.Status = modelarts.TransTrainJobStatus(result.IntStatus)
  1017. task.Duration = result.Duration / 1000
  1018. task.TrainJobDuration = result.TrainJobDuration
  1019. if task.StartTime == 0 && result.StartTime > 0 {
  1020. task.StartTime = timeutil.TimeStamp(result.StartTime / 1000)
  1021. }
  1022. task.TrainJobDuration = models.ConvertDurationToStr(task.Duration)
  1023. if task.EndTime == 0 && models.IsTrainJobTerminal(task.Status) && task.StartTime > 0 {
  1024. task.EndTime = task.StartTime.Add(task.Duration)
  1025. }
  1026. err = models.UpdateJob(task)
  1027. if err != nil {
  1028. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  1029. continue
  1030. }
  1031. }
  1032. } else {
  1033. log.Error("task.JobType(%s) is error:%s", task.JobName, task.JobType)
  1034. }
  1035. } else {
  1036. log.Error("task.Type(%s) is error:%d", task.JobName, task.Type)
  1037. }
  1038. }
  1039. return
  1040. }
  1041. func HandleTaskWithNoDuration(ctx *context.Context) {
  1042. log.Info("HandleTaskWithNoDuration start")
  1043. count := 0
  1044. start := time.Now().Unix()
  1045. for {
  1046. cloudBrains, err := models.GetStoppedJobWithNoDurationJob()
  1047. if err != nil {
  1048. log.Error("HandleTaskWithNoTrainJobDuration failed:", err.Error())
  1049. break
  1050. }
  1051. if len(cloudBrains) == 0 {
  1052. log.Info("HandleTaskWithNoTrainJobDuration:no task need handle")
  1053. break
  1054. }
  1055. handleNoDurationTask(cloudBrains)
  1056. count += len(cloudBrains)
  1057. if len(cloudBrains) < 100 {
  1058. log.Info("HandleTaskWithNoTrainJobDuration:task less than 100")
  1059. break
  1060. }
  1061. if time.Now().Unix()-start > 600 {
  1062. log.Info("HandleTaskWithNoDuration : time out")
  1063. ctx.JSON(200, fmt.Sprintf("task stop for time out,count=%d", count))
  1064. return
  1065. }
  1066. }
  1067. log.Info("HandleTaskWithNoTrainJobDuration:count=%d", count)
  1068. ctx.JSON(200, fmt.Sprintf("success,count=%d", count))
  1069. }
  1070. func handleNoDurationTask(cloudBrains []*models.Cloudbrain) {
  1071. for _, task := range cloudBrains {
  1072. time.Sleep(time.Millisecond * 100)
  1073. log.Info("Handle job ,%+v", task)
  1074. if task.Type == models.TypeCloudBrainOne {
  1075. result, err := cloudbrain.GetJob(task.JobID)
  1076. if err != nil {
  1077. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  1078. updateDefaultDuration(task)
  1079. continue
  1080. }
  1081. if result != nil {
  1082. if result.Msg != "success" {
  1083. updateDefaultDuration(task)
  1084. continue
  1085. }
  1086. jobRes, err := models.ConvertToJobResultPayload(result.Payload)
  1087. if err != nil || len(jobRes.TaskRoles) == 0 {
  1088. updateDefaultDuration(task)
  1089. continue
  1090. }
  1091. taskRoles := jobRes.TaskRoles
  1092. taskRes, err := models.ConvertToTaskPod(taskRoles[cloudbrain.SubTaskName].(map[string]interface{}))
  1093. if err != nil || len(taskRes.TaskStatuses) == 0 {
  1094. updateDefaultDuration(task)
  1095. continue
  1096. }
  1097. task.Status = taskRes.TaskStatuses[0].State
  1098. log.Info("task startTime = %v endTime= %v ,jobId=%d", jobRes.JobStatus.StartTime, jobRes.JobStatus.EndTime, task.ID)
  1099. if jobRes.JobStatus.CreatedTime > 0 {
  1100. task.StartTime = timeutil.TimeStamp(jobRes.JobStatus.CreatedTime / 1000)
  1101. if jobRes.JobStatus.CompletedTime > 0 {
  1102. task.EndTime = timeutil.TimeStamp(jobRes.JobStatus.CompletedTime / 1000)
  1103. } else {
  1104. task.EndTime = task.UpdatedUnix
  1105. }
  1106. } else {
  1107. task.StartTime = 0
  1108. task.EndTime = 0
  1109. }
  1110. if task.EndTime < task.StartTime {
  1111. log.Info("endTime[%v] is less than starTime[%v],jobId=%d", task.EndTime, task.StartTime, task.ID)
  1112. st := task.StartTime
  1113. task.StartTime = task.EndTime
  1114. task.EndTime = st
  1115. }
  1116. task.Duration = task.EndTime.AsTime().Unix() - task.StartTime.AsTime().Unix()
  1117. task.TrainJobDuration = models.ConvertDurationToStr(task.Duration)
  1118. err = models.UpdateJob(task)
  1119. if err != nil {
  1120. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  1121. }
  1122. }
  1123. } else if task.Type == models.TypeCloudBrainTwo {
  1124. if task.JobType == string(models.JobTypeDebug) {
  1125. //result, err := modelarts.GetJob(task.JobID)
  1126. result, err := modelarts.GetNotebook2(task.JobID)
  1127. if err != nil {
  1128. log.Error("GetJob(%s) failed:%v", task.JobName, err)
  1129. updateDefaultDuration(task)
  1130. continue
  1131. }
  1132. if result != nil {
  1133. task.Status = result.Status
  1134. startTime := result.Lease.CreateTime
  1135. duration := result.Lease.Duration / 1000
  1136. if startTime > 0 {
  1137. task.StartTime = timeutil.TimeStamp(startTime / 1000)
  1138. task.EndTime = task.StartTime.Add(duration)
  1139. }
  1140. task.ComputeAndSetDuration()
  1141. err = models.UpdateJob(task)
  1142. if err != nil {
  1143. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  1144. continue
  1145. }
  1146. }
  1147. } else if task.JobType == string(models.JobTypeTrain) || task.JobType == string(models.JobTypeInference) {
  1148. result, err := modelarts.GetTrainJob(task.JobID, strconv.FormatInt(task.VersionID, 10))
  1149. if err != nil {
  1150. log.Error("GetTrainJob(%s) failed:%v", task.JobName, err)
  1151. updateDefaultDuration(task)
  1152. continue
  1153. }
  1154. if result != nil {
  1155. startTime := result.StartTime / 1000
  1156. if startTime > 0 {
  1157. task.StartTime = timeutil.TimeStamp(startTime)
  1158. task.EndTime = task.StartTime.Add(result.Duration / 1000)
  1159. }
  1160. task.ComputeAndSetDuration()
  1161. err = models.UpdateJob(task)
  1162. if err != nil {
  1163. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  1164. continue
  1165. }
  1166. }
  1167. } else {
  1168. log.Error("task.JobType(%s) is error:%s", task.JobName, task.JobType)
  1169. }
  1170. } else {
  1171. log.Error("task.Type(%s) is error:%d", task.JobName, task.Type)
  1172. }
  1173. }
  1174. }
  1175. func updateDefaultDuration(task *models.Cloudbrain) {
  1176. log.Info("updateDefaultDuration: taskId=%d", task.ID)
  1177. task.StartTime = task.CreatedUnix
  1178. task.EndTime = task.UpdatedUnix
  1179. task.ComputeAndSetDuration()
  1180. err := models.UpdateJob(task)
  1181. if err != nil {
  1182. log.Error("UpdateJob(%s) failed:%v", task.JobName, err)
  1183. }
  1184. }
  1185. func CloudBrainBenchmarkIndex(ctx *context.Context) {
  1186. MustEnableCloudbrain(ctx)
  1187. repo := ctx.Repo.Repository
  1188. page := ctx.QueryInt("page")
  1189. if page <= 0 {
  1190. page = 1
  1191. }
  1192. var jobTypes []string
  1193. jobTypes = append(jobTypes, string(models.JobTypeBenchmark))
  1194. ciTasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  1195. ListOptions: models.ListOptions{
  1196. Page: page,
  1197. PageSize: setting.UI.IssuePagingNum,
  1198. },
  1199. RepoID: repo.ID,
  1200. Type: models.TypeCloudBrainOne,
  1201. JobTypes: jobTypes,
  1202. })
  1203. if err != nil {
  1204. ctx.ServerError("Get debugjob faild:", err)
  1205. return
  1206. }
  1207. for i, task := range ciTasks {
  1208. ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  1209. ciTasks[i].Cloudbrain.ComputeResource = task.ComputeResource
  1210. if ciTasks[i].TrainJobDuration == "" {
  1211. if ciTasks[i].Duration == 0 {
  1212. var duration int64
  1213. if task.Status == string(models.JobRunning) {
  1214. duration = time.Now().Unix() - int64(task.Cloudbrain.CreatedUnix)
  1215. } else {
  1216. duration = int64(task.Cloudbrain.UpdatedUnix) - int64(task.Cloudbrain.CreatedUnix)
  1217. }
  1218. ciTasks[i].Duration = duration
  1219. }
  1220. ciTasks[i].TrainJobDuration = models.ConvertDurationToStr(ciTasks[i].Duration)
  1221. }
  1222. ciTasks[i].BenchmarkTypeName = ""
  1223. if task.BenchmarkTypeID > 0 {
  1224. for _, benchmarkType := range GetBenchmarkTypes(ctx).BenchmarkType {
  1225. if task.BenchmarkTypeID == benchmarkType.Id {
  1226. ciTasks[i].BenchmarkTypeRankLink = benchmarkType.RankLink
  1227. ciTasks[i].BenchmarkTypeName = benchmarkType.First
  1228. break
  1229. }
  1230. }
  1231. }
  1232. }
  1233. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  1234. ctx.Data["Page"] = pager
  1235. ctx.Data["PageIsCloudBrain"] = true
  1236. ctx.Data["Tasks"] = ciTasks
  1237. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  1238. ctx.Data["RepoIsEmpty"] = repo.IsEmpty
  1239. ctx.HTML(200, tplCloudBrainBenchmarkIndex)
  1240. }
  1241. func GetChildTypes(ctx *context.Context) {
  1242. benchmarkTypeID := ctx.QueryInt("benchmark_type_id")
  1243. re := make(map[string]interface{})
  1244. for {
  1245. var isExist bool
  1246. for _, benchmarkType := range GetBenchmarkTypes(ctx).BenchmarkType {
  1247. if benchmarkTypeID == benchmarkType.Id {
  1248. isExist = true
  1249. re["child_types"] = benchmarkType.Second
  1250. re["result_code"] = "0"
  1251. break
  1252. }
  1253. }
  1254. if !isExist {
  1255. re["result_code"] = "1"
  1256. log.Error("no such benchmark_type_id", ctx.Data["MsgID"])
  1257. re["errMsg"] = "system error"
  1258. break
  1259. }
  1260. break
  1261. }
  1262. ctx.JSON(200, re)
  1263. }
  1264. func CloudBrainBenchmarkNew(ctx *context.Context) {
  1265. ctx.Data["description"] = ""
  1266. ctx.Data["benchmarkTypeID"] = -1
  1267. ctx.Data["benchmark_child_types_id_hidden"] = -1
  1268. err := cloudBrainNewDataPrepare(ctx)
  1269. if err != nil {
  1270. ctx.ServerError("get new cloudbrain info failed", err)
  1271. return
  1272. }
  1273. ctx.HTML(200, tplCloudBrainBenchmarkNew)
  1274. }
  1275. func getBenchmarkAttachment(benchmarkTypeID, benchmarkChildTypeID int, ctx *context.Context) (*models.BenchmarkDataset, error) {
  1276. var childInfo *models.BenchmarkDataset
  1277. var isExist bool
  1278. for _, benchmarkType := range GetBenchmarkTypes(ctx).BenchmarkType {
  1279. if benchmarkType.Id == benchmarkTypeID {
  1280. for _, childType := range benchmarkType.Second {
  1281. if childType.Id == benchmarkChildTypeID {
  1282. childInfo = childType
  1283. isExist = true
  1284. break
  1285. }
  1286. }
  1287. break
  1288. }
  1289. }
  1290. if !isExist {
  1291. log.Error("no such benchmark_type_id&benchmark_child_type_id")
  1292. return childInfo, errors.New("no such benchmark_type_id&benchmark_child_type_id")
  1293. }
  1294. return childInfo, nil
  1295. }
  1296. func getBenchmarkGpuQueue(gpuQueue string) (string, error) {
  1297. queue := ""
  1298. if benchmarkGpuInfos == nil {
  1299. if err := json.Unmarshal([]byte(setting.BenchmarkGpuTypes), &benchmarkGpuInfos); err != nil {
  1300. log.Error("json.Unmarshal BenchmarkGpuTypes(%s) failed:%v", setting.BenchmarkGpuTypes, err)
  1301. return queue, err
  1302. }
  1303. }
  1304. var isExist bool
  1305. for _, gpuInfo := range benchmarkGpuInfos.GpuInfo {
  1306. if gpuQueue == gpuInfo.Queue {
  1307. isExist = true
  1308. queue = gpuQueue
  1309. break
  1310. }
  1311. }
  1312. if !isExist {
  1313. log.Error("no such gpuQueue, %s", gpuQueue)
  1314. return queue, errors.New("no such gpuQueue")
  1315. }
  1316. return queue, nil
  1317. }
  1318. func getBenchmarkResourceSpec(resourceSpecID int) (int, error) {
  1319. var id int
  1320. if benchmarkResourceSpecs == nil {
  1321. if err := json.Unmarshal([]byte(setting.BenchmarkResourceSpecs), &benchmarkResourceSpecs); err != nil {
  1322. log.Error("json.Unmarshal BenchmarkResourceSpecs(%s) failed:%v", setting.BenchmarkResourceSpecs, err)
  1323. return id, err
  1324. }
  1325. }
  1326. var isExist bool
  1327. for _, resourceSpec := range benchmarkResourceSpecs.ResourceSpec {
  1328. if resourceSpecID == resourceSpec.Id {
  1329. isExist = true
  1330. id = resourceSpecID
  1331. break
  1332. }
  1333. }
  1334. if !isExist {
  1335. log.Error("no such resourceSpecID, %d", resourceSpecID)
  1336. return id, errors.New("no such resourceSpec")
  1337. }
  1338. return id, nil
  1339. }
  1340. func CloudBrainBenchmarkCreate(ctx *context.Context, form auth.CreateCloudBrainForm) {
  1341. ctx.Data["PageIsCloudBrain"] = true
  1342. displayJobName := form.DisplayJobName
  1343. jobName := util.ConvertDisplayJobNameToJobName(displayJobName)
  1344. image := form.Image
  1345. gpuQueue := form.GpuType
  1346. command := cloudbrain.CommandBenchmark
  1347. codePath := setting.JobPath + jobName + cloudbrain.CodeMountPath
  1348. resourceSpecId := cloudbrain.BenchMarkResourceID
  1349. benchmarkTypeID := form.BenchmarkTypeID
  1350. benchmarkChildTypeID := form.BenchmarkChildTypeID
  1351. ctx.Data["description"] = form.Description
  1352. ctx.Data["benchmarkTypeID"] = benchmarkTypeID
  1353. ctx.Data["benchmark_child_types_id_hidden"] = benchmarkChildTypeID
  1354. repo := ctx.Repo.Repository
  1355. tasks, err := models.GetCloudbrainsByDisplayJobName(repo.ID, string(models.JobTypeBenchmark), displayJobName)
  1356. if err == nil {
  1357. if len(tasks) != 0 {
  1358. log.Error("the job name did already exist", ctx.Data["MsgID"])
  1359. cloudBrainNewDataPrepare(ctx)
  1360. ctx.RenderWithErr("the job name did already exist", tplCloudBrainBenchmarkNew, &form)
  1361. return
  1362. }
  1363. } else {
  1364. if !models.IsErrJobNotExist(err) {
  1365. log.Error("system error, %v", err, ctx.Data["MsgID"])
  1366. cloudBrainNewDataPrepare(ctx)
  1367. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1368. return
  1369. }
  1370. }
  1371. if !jobNamePattern.MatchString(jobName) {
  1372. cloudBrainNewDataPrepare(ctx)
  1373. ctx.RenderWithErr(ctx.Tr("repo.cloudbrain_jobname_err"), tplCloudBrainBenchmarkNew, &form)
  1374. return
  1375. }
  1376. childInfo, err := getBenchmarkAttachment(benchmarkTypeID, benchmarkChildTypeID, ctx)
  1377. if err != nil {
  1378. log.Error("getBenchmarkAttachment failed:%v", err, ctx.Data["MsgID"])
  1379. cloudBrainNewDataPrepare(ctx)
  1380. ctx.RenderWithErr("benchmark type error", tplCloudBrainBenchmarkNew, &form)
  1381. return
  1382. }
  1383. _, err = getBenchmarkGpuQueue(gpuQueue)
  1384. if err != nil {
  1385. log.Error("getBenchmarkGpuQueue failed:%v", err, ctx.Data["MsgID"])
  1386. cloudBrainNewDataPrepare(ctx)
  1387. ctx.RenderWithErr("gpu queue error", tplCloudBrainBenchmarkNew, &form)
  1388. return
  1389. }
  1390. _, err = getBenchmarkResourceSpec(resourceSpecId)
  1391. if err != nil {
  1392. log.Error("getBenchmarkResourceSpec failed:%v", err, ctx.Data["MsgID"])
  1393. cloudBrainNewDataPrepare(ctx)
  1394. ctx.RenderWithErr("resource spec error", tplCloudBrainBenchmarkNew, &form)
  1395. return
  1396. }
  1397. count, err := models.GetCloudbrainCountByUserID(ctx.User.ID, string(models.JobTypeBenchmark))
  1398. if err != nil {
  1399. log.Error("GetCloudbrainCountByUserID failed:%v", err, ctx.Data["MsgID"])
  1400. cloudBrainNewDataPrepare(ctx)
  1401. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1402. return
  1403. } else {
  1404. if count >= 1 {
  1405. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  1406. cloudBrainNewDataPrepare(ctx)
  1407. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplCloudBrainBenchmarkNew, &form)
  1408. return
  1409. }
  1410. }
  1411. os.RemoveAll(codePath)
  1412. if err := downloadCode(repo, codePath, cloudbrain.DefaultBranchName); err != nil {
  1413. log.Error("downloadCode failed, %v", err, ctx.Data["MsgID"])
  1414. cloudBrainNewDataPrepare(ctx)
  1415. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1416. return
  1417. }
  1418. if _, err := os.Stat(codePath + "/train.py"); err != nil {
  1419. if os.IsNotExist(err) {
  1420. // file does not exist
  1421. log.Error("train.py does not exist, %v", err, ctx.Data["MsgID"])
  1422. cloudBrainNewDataPrepare(ctx)
  1423. ctx.RenderWithErr("train.py does not exist", tplCloudBrainBenchmarkNew, &form)
  1424. } else {
  1425. log.Error("Stat failed, %v", err, ctx.Data["MsgID"])
  1426. cloudBrainNewDataPrepare(ctx)
  1427. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1428. }
  1429. return
  1430. } else if _, err := os.Stat(codePath + "/test.py"); err != nil {
  1431. if os.IsNotExist(err) {
  1432. // file does not exist
  1433. log.Error("test.py does not exist, %v", err, ctx.Data["MsgID"])
  1434. cloudBrainNewDataPrepare(ctx)
  1435. ctx.RenderWithErr("test.py does not exist", tplCloudBrainBenchmarkNew, &form)
  1436. } else {
  1437. log.Error("Stat failed, %v", err, ctx.Data["MsgID"])
  1438. cloudBrainNewDataPrepare(ctx)
  1439. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1440. }
  1441. return
  1442. }
  1443. if err := uploadCodeToMinio(codePath+"/", jobName, cloudbrain.CodeMountPath+"/"); err != nil {
  1444. log.Error("uploadCodeToMinio failed, %v", err, ctx.Data["MsgID"])
  1445. cloudBrainNewDataPrepare(ctx)
  1446. ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1447. return
  1448. }
  1449. benchmarkPath := setting.JobPath + jobName + cloudbrain.BenchMarkMountPath
  1450. var gpuType string
  1451. for _, gpuInfo := range gpuInfos.GpuInfo {
  1452. if gpuInfo.Queue == gpuQueue {
  1453. gpuType = gpuInfo.Value
  1454. }
  1455. }
  1456. if err := downloadRateCode(repo, jobName, childInfo.Owner, childInfo.RepoName, benchmarkPath, form.BenchmarkCategory, gpuType); err != nil {
  1457. log.Error("downloadRateCode failed, %v", err, ctx.Data["MsgID"])
  1458. //cloudBrainNewDataPrepare(ctx)
  1459. //ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1460. //return
  1461. }
  1462. if err := uploadCodeToMinio(benchmarkPath+"/", jobName, cloudbrain.BenchMarkMountPath+"/"); err != nil {
  1463. log.Error("uploadCodeToMinio failed, %v", err, ctx.Data["MsgID"])
  1464. //cloudBrainNewDataPrepare(ctx)
  1465. //ctx.RenderWithErr("system error", tplCloudBrainBenchmarkNew, &form)
  1466. //return
  1467. }
  1468. err = cloudbrain.GenerateTask(ctx, displayJobName, jobName, image, command, childInfo.Attachment, storage.GetMinioPath(jobName, cloudbrain.CodeMountPath+"/"),
  1469. storage.GetMinioPath(jobName, cloudbrain.ModelMountPath+"/"),
  1470. storage.GetMinioPath(jobName, cloudbrain.BenchMarkMountPath+"/"), storage.GetMinioPath(jobName, cloudbrain.Snn4imagenetMountPath+"/"),
  1471. storage.GetMinioPath(jobName, cloudbrain.BrainScoreMountPath+"/"), string(models.JobTypeBenchmark), gpuQueue, form.Description, cloudbrain.DefaultBranchName, "", "",
  1472. benchmarkTypeID, benchmarkChildTypeID, resourceSpecId)
  1473. if err != nil {
  1474. cloudBrainNewDataPrepare(ctx)
  1475. ctx.RenderWithErr(err.Error(), tplCloudBrainBenchmarkNew, &form)
  1476. return
  1477. }
  1478. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/cloudbrain/benchmark")
  1479. }
  1480. func BenchmarkDel(ctx *context.Context) {
  1481. if err := deleteCloudbrainJob(ctx); err != nil {
  1482. log.Error("deleteCloudbrainJob failed: %v", err, ctx.Data["msgID"])
  1483. ctx.ServerError(err.Error(), err)
  1484. return
  1485. }
  1486. var isAdminPage = ctx.Query("isadminpage")
  1487. var isHomePage = ctx.Query("isHomepage")
  1488. if ctx.IsUserSiteAdmin() && isAdminPage == "true" {
  1489. ctx.Redirect(setting.AppSubURL + "/admin" + "/cloudbrains")
  1490. } else if isHomePage == "true" {
  1491. ctx.Redirect(setting.AppSubURL + "/cloudbrains")
  1492. } else {
  1493. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/cloudbrain/benchmark")
  1494. }
  1495. }
  1496. func CloudBrainTrainJobNew(ctx *context.Context) {
  1497. err := cloudBrainNewDataPrepare(ctx)
  1498. if err != nil {
  1499. ctx.ServerError("get new train-job info failed", err)
  1500. return
  1501. }
  1502. ctx.HTML(http.StatusOK, tplCloudBrainTrainJobNew)
  1503. }
  1504. func getTrainJobCommand(form auth.CreateCloudBrainForm) (string, error) {
  1505. var command string
  1506. bootFile := form.BootFile
  1507. params := form.Params
  1508. if !strings.HasSuffix(bootFile, ".py") {
  1509. log.Error("bootFile(%s) format error", bootFile)
  1510. return command, errors.New("bootFile format error")
  1511. }
  1512. var parameters models.Parameters
  1513. var param string
  1514. if len(params) != 0 {
  1515. err := json.Unmarshal([]byte(params), &parameters)
  1516. if err != nil {
  1517. log.Error("Failed to Unmarshal params: %s (%v)", params, err)
  1518. return command, err
  1519. }
  1520. for _, parameter := range parameters.Parameter {
  1521. param += " --" + parameter.Label + "=" + parameter.Value
  1522. }
  1523. }
  1524. command += "python /code/" + bootFile + param + " > " + cloudbrain.ModelMountPath + "/" + form.DisplayJobName + "-" + cloudbrain.LogFile
  1525. return command, nil
  1526. }
  1527. func CloudBrainTrainJobDel(ctx *context.Context) {
  1528. var listType = ctx.Query("listType")
  1529. if err := deleteCloudbrainJob(ctx); err != nil {
  1530. log.Error("deleteCloudbrainJob failed: %v", err, ctx.Data["msgID"])
  1531. ctx.ServerError(err.Error(), err)
  1532. return
  1533. }
  1534. var isAdminPage = ctx.Query("isadminpage")
  1535. var isHomePage = ctx.Query("isHomepage")
  1536. if ctx.IsUserSiteAdmin() && isAdminPage == "true" {
  1537. ctx.Redirect(setting.AppSubURL + "/admin" + "/cloudbrains")
  1538. } else if isHomePage == "true" {
  1539. ctx.Redirect(setting.AppSubURL + "/cloudbrains")
  1540. } else {
  1541. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job?listType=" + listType)
  1542. }
  1543. }
  1544. func GetBenchmarkTypes(ctx *context.Context) *models.BenchmarkTypes {
  1545. var lang = ctx.Locale.Language()
  1546. if benchmarkTypesMap[lang] == nil {
  1547. var val = i18n.Tr(lang, BENCHMARK_TYPE_CODE)
  1548. //use config
  1549. val = setting.BenchmarkTypes
  1550. var tempType *models.BenchmarkTypes
  1551. if err := json.Unmarshal([]byte(val), &tempType); err != nil {
  1552. log.Error("json.Unmarshal BenchmarkTypes(%s) failed:%v", val, err, ctx.Data["MsgID"])
  1553. return &models.BenchmarkTypes{}
  1554. }
  1555. benchmarkTypesMap[lang] = tempType
  1556. }
  1557. return benchmarkTypesMap[lang]
  1558. }