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.

modelarts.go 49 kB

4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558
  1. package repo
  2. import (
  3. "encoding/json"
  4. "errors"
  5. "io"
  6. "net/http"
  7. "os"
  8. "path"
  9. "strconv"
  10. "strings"
  11. "time"
  12. "code.gitea.io/gitea/models"
  13. "code.gitea.io/gitea/modules/auth"
  14. "code.gitea.io/gitea/modules/base"
  15. "code.gitea.io/gitea/modules/cloudbrain"
  16. "code.gitea.io/gitea/modules/context"
  17. "code.gitea.io/gitea/modules/git"
  18. "code.gitea.io/gitea/modules/log"
  19. "code.gitea.io/gitea/modules/modelarts"
  20. "code.gitea.io/gitea/modules/obs"
  21. "code.gitea.io/gitea/modules/setting"
  22. "code.gitea.io/gitea/modules/storage"
  23. "github.com/unknwon/com"
  24. )
  25. const (
  26. tplDebugJobIndex base.TplName = "repo/debugjob/index"
  27. tplModelArtsNotebookIndex base.TplName = "repo/modelarts/notebook/index"
  28. tplModelArtsNotebookNew base.TplName = "repo/modelarts/notebook/new"
  29. tplModelArtsNotebookShow base.TplName = "repo/modelarts/notebook/show"
  30. tplModelArtsTrainJobIndex base.TplName = "repo/modelarts/trainjob/index"
  31. tplModelArtsTrainJobNew base.TplName = "repo/modelarts/trainjob/new"
  32. tplModelArtsTrainJobShow base.TplName = "repo/modelarts/trainjob/show"
  33. tplModelArtsTrainJobVersionNew base.TplName = "repo/modelarts/trainjob/version_new"
  34. )
  35. func DebugJobIndex(ctx *context.Context) {
  36. debugListType := ctx.Query("debugListType")
  37. ctx.Data["ListType"] = debugListType
  38. MustEnableCloudbrain(ctx)
  39. repo := ctx.Repo.Repository
  40. page := ctx.QueryInt("page")
  41. if page <= 0 {
  42. page = 1
  43. }
  44. debugType := modelarts.DebugType
  45. if debugListType == models.GPUResource {
  46. debugType = models.TypeCloudBrainOne
  47. } else if debugListType == models.NPUResource {
  48. debugType = models.TypeCloudBrainTwo
  49. }
  50. ciTasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  51. ListOptions: models.ListOptions{
  52. Page: page,
  53. PageSize: setting.UI.IssuePagingNum,
  54. },
  55. RepoID: repo.ID,
  56. Type: debugType,
  57. })
  58. if err != nil {
  59. ctx.ServerError("Get debugjob faild:", err)
  60. return
  61. }
  62. for i, task := range ciTasks {
  63. ciTasks[i].CanDebug = cloudbrain.CanModifyJob(ctx, &task.Cloudbrain)
  64. ciTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  65. ciTasks[i].Cloudbrain.ComputeResource = task.ComputeResource
  66. }
  67. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  68. pager.AddParam(ctx, "debugListType", "ListType")
  69. ctx.Data["Page"] = pager
  70. ctx.Data["PageIsCloudBrain"] = true
  71. ctx.Data["Tasks"] = ciTasks
  72. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  73. ctx.Data["RepoIsEmpty"] = repo.IsEmpty
  74. ctx.HTML(200, tplDebugJobIndex)
  75. }
  76. // MustEnableDataset check if repository enable internal cb
  77. func MustEnableModelArts(ctx *context.Context) {
  78. if !ctx.Repo.CanRead(models.UnitTypeCloudBrain) {
  79. ctx.NotFound("MustEnableCloudbrain", nil)
  80. return
  81. }
  82. }
  83. func NotebookNew(ctx *context.Context) {
  84. ctx.Data["PageIsCloudBrain"] = true
  85. t := time.Now()
  86. var jobName = jobNamePrefixValid(cutString(ctx.User.Name, 5)) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  87. ctx.Data["job_name"] = jobName
  88. attachs, err := models.GetModelArtsUserAttachments(ctx.User.ID)
  89. if err != nil {
  90. ctx.ServerError("GetAllUserAttachments failed:", err)
  91. return
  92. }
  93. ctx.Data["attachments"] = attachs
  94. ctx.Data["dataset_path"] = modelarts.DataSetMountPath
  95. ctx.Data["env"] = modelarts.NotebookEnv
  96. ctx.Data["notebook_type"] = modelarts.NotebookType
  97. if modelarts.FlavorInfos == nil {
  98. json.Unmarshal([]byte(setting.FlavorInfos), &modelarts.FlavorInfos)
  99. }
  100. ctx.Data["flavors"] = modelarts.FlavorInfos.FlavorInfo
  101. ctx.HTML(200, tplModelArtsNotebookNew)
  102. }
  103. func NotebookCreate(ctx *context.Context, form auth.CreateModelArtsNotebookForm) {
  104. ctx.Data["PageIsNotebook"] = true
  105. jobName := form.JobName
  106. uuid := form.Attachment
  107. description := form.Description
  108. flavor := form.Flavor
  109. count, err := models.GetCloudbrainNotebookCountByUserID(ctx.User.ID)
  110. if err != nil {
  111. log.Error("GetCloudbrainNotebookCountByUserID failed:%v", err, ctx.Data["MsgID"])
  112. cloudBrainNewDataPrepare(ctx)
  113. ctx.RenderWithErr("system error", tplModelArtsNotebookNew, &form)
  114. return
  115. } else {
  116. if count >= 1 {
  117. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  118. cloudBrainNewDataPrepare(ctx)
  119. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsNotebookNew, &form)
  120. return
  121. }
  122. }
  123. err = modelarts.GenerateTask(ctx, jobName, uuid, description, flavor)
  124. if err != nil {
  125. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookNew, &form)
  126. return
  127. }
  128. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=all")
  129. }
  130. func NotebookShow(ctx *context.Context) {
  131. ctx.Data["PageIsCloudBrain"] = true
  132. var jobID = ctx.Params(":jobid")
  133. task, err := models.GetCloudbrainByJobID(jobID)
  134. if err != nil {
  135. ctx.Data["error"] = err.Error()
  136. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil)
  137. return
  138. }
  139. result, err := modelarts.GetJob(jobID)
  140. if err != nil {
  141. ctx.Data["error"] = err.Error()
  142. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil)
  143. return
  144. }
  145. if result != nil {
  146. task.Status = result.Status
  147. err = models.UpdateJob(task)
  148. if err != nil {
  149. ctx.Data["error"] = err.Error()
  150. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookShow, nil)
  151. return
  152. }
  153. createTime, _ := com.StrTo(result.CreationTimestamp).Int64()
  154. result.CreateTime = time.Unix(int64(createTime/1000), 0).Format("2006-01-02 15:04:05")
  155. endTime, _ := com.StrTo(result.LatestUpdateTimestamp).Int64()
  156. result.LatestUpdateTime = time.Unix(int64(endTime/1000), 0).Format("2006-01-02 15:04:05")
  157. result.QueuingInfo.BeginTime = time.Unix(int64(result.QueuingInfo.BeginTimestamp/1000), 0).Format("2006-01-02 15:04:05")
  158. result.QueuingInfo.EndTime = time.Unix(int64(result.QueuingInfo.EndTimestamp/1000), 0).Format("2006-01-02 15:04:05")
  159. }
  160. ctx.Data["task"] = task
  161. ctx.Data["jobID"] = jobID
  162. ctx.Data["result"] = result
  163. ctx.HTML(200, tplModelArtsNotebookShow)
  164. }
  165. func NotebookDebug(ctx *context.Context) {
  166. var jobID = ctx.Params(":jobid")
  167. _, err := models.GetCloudbrainByJobID(jobID)
  168. if err != nil {
  169. ctx.ServerError("GetCloudbrainByJobID failed", err)
  170. return
  171. }
  172. result, err := modelarts.GetJob(jobID)
  173. if err != nil {
  174. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil)
  175. return
  176. }
  177. res, err := modelarts.GetJobToken(jobID)
  178. if err != nil {
  179. ctx.RenderWithErr(err.Error(), tplModelArtsNotebookIndex, nil)
  180. return
  181. }
  182. urls := strings.Split(result.Spec.Annotations.Url, "/")
  183. urlPrefix := result.Spec.Annotations.TargetDomain
  184. for i, url := range urls {
  185. if i > 2 {
  186. urlPrefix += "/" + url
  187. }
  188. }
  189. debugUrl := urlPrefix + "?token=" + res.Token
  190. ctx.Redirect(debugUrl)
  191. }
  192. func NotebookManage(ctx *context.Context) {
  193. var jobID = ctx.Params(":jobid")
  194. var action = ctx.Params(":action")
  195. var resultCode = "0"
  196. var errorMsg = ""
  197. var status = ""
  198. for {
  199. task, err := models.GetCloudbrainByJobID(jobID)
  200. if err != nil {
  201. log.Error("GetCloudbrainByJobID failed:%v", err, ctx.Data["MsgID"])
  202. resultCode = "-1"
  203. errorMsg = "system error"
  204. break
  205. }
  206. if action == models.ActionStop {
  207. if task.Status != string(models.ModelArtsRunning) {
  208. log.Error("the job(%s) is not running", task.JobName, ctx.Data["MsgID"])
  209. resultCode = "-1"
  210. errorMsg = "the job is not running"
  211. break
  212. }
  213. if !ctx.IsSigned || (ctx.User.ID != task.UserID && !ctx.IsUserSiteAdmin() && !ctx.IsUserRepoOwner()) {
  214. log.Error("the user has no right ro stop the job", task.JobName, ctx.Data["MsgID"])
  215. resultCode = "-1"
  216. errorMsg = "you have no right to stop the job"
  217. break
  218. }
  219. } else if action == models.ActionRestart {
  220. if task.Status != string(models.ModelArtsStopped) && task.Status != string(models.ModelArtsStartFailed) && task.Status != string(models.ModelArtsCreateFailed) {
  221. log.Error("the job(%s) is not stopped", task.JobName, ctx.Data["MsgID"])
  222. resultCode = "-1"
  223. errorMsg = "the job is not stopped"
  224. break
  225. }
  226. if !ctx.IsSigned || (ctx.User.ID != task.UserID && !ctx.IsUserSiteAdmin()) {
  227. log.Error("the user has no right ro restart the job", task.JobName, ctx.Data["MsgID"])
  228. resultCode = "-1"
  229. errorMsg = "you have no right to restart the job"
  230. break
  231. }
  232. count, err := models.GetCloudbrainNotebookCountByUserID(ctx.User.ID)
  233. if err != nil {
  234. log.Error("GetCloudbrainNotebookCountByUserID failed:%v", err, ctx.Data["MsgID"])
  235. resultCode = "-1"
  236. errorMsg = "system error"
  237. break
  238. } else {
  239. if count >= 1 {
  240. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  241. resultCode = "-1"
  242. errorMsg = "you have already a running or waiting task, can not create more"
  243. break
  244. }
  245. }
  246. action = models.ActionStart
  247. } else {
  248. log.Error("the action(%s) is illegal", action, ctx.Data["MsgID"])
  249. resultCode = "-1"
  250. errorMsg = "非法操作"
  251. break
  252. }
  253. param := models.NotebookAction{
  254. Action: action,
  255. }
  256. res, err := modelarts.ManageNotebook(jobID, param)
  257. if err != nil {
  258. log.Error("ManageNotebook(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"])
  259. resultCode = "-1"
  260. errorMsg = "启动失败"
  261. break
  262. }
  263. task.Status = res.CurrentStatus
  264. err = models.UpdateJob(task)
  265. if err != nil {
  266. log.Error("UpdateJob(%s) failed:%v", task.JobName, err.Error(), ctx.Data["MsgID"])
  267. resultCode = "-1"
  268. errorMsg = "system error"
  269. break
  270. }
  271. status = task.Status
  272. break
  273. }
  274. ctx.JSON(200, map[string]string{
  275. "result_code": resultCode,
  276. "error_msg": errorMsg,
  277. "status": status,
  278. "job_id": jobID,
  279. })
  280. }
  281. func NotebookDel(ctx *context.Context) {
  282. var jobID = ctx.Params(":jobid")
  283. task, err := models.GetCloudbrainByJobID(jobID)
  284. if err != nil {
  285. ctx.ServerError("GetCloudbrainByJobID failed", err)
  286. return
  287. }
  288. if task.Status != string(models.ModelArtsCreateFailed) && task.Status != string(models.ModelArtsStartFailed) && task.Status != string(models.ModelArtsStopped) {
  289. log.Error("the job(%s) has not been stopped", task.JobName)
  290. ctx.ServerError("the job has not been stopped", errors.New("the job has not been stopped"))
  291. return
  292. }
  293. _, err = modelarts.DelNotebook(jobID)
  294. if err != nil {
  295. log.Error("DelJob(%s) failed:%v", task.JobName, err.Error())
  296. ctx.ServerError("DelJob failed", err)
  297. return
  298. }
  299. err = models.DeleteJob(task)
  300. if err != nil {
  301. ctx.ServerError("DeleteJob failed", err)
  302. return
  303. }
  304. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/debugjob?debugListType=all")
  305. }
  306. func TrainJobIndex(ctx *context.Context) {
  307. MustEnableModelArts(ctx)
  308. repo := ctx.Repo.Repository
  309. page := ctx.QueryInt("page")
  310. if page <= 0 {
  311. page = 1
  312. }
  313. tasks, count, err := models.Cloudbrains(&models.CloudbrainsOptions{
  314. ListOptions: models.ListOptions{
  315. Page: page,
  316. PageSize: setting.UI.IssuePagingNum,
  317. },
  318. RepoID: repo.ID,
  319. Type: models.TypeCloudBrainTwo,
  320. JobType: string(models.JobTypeTrain),
  321. IsLatestVersion: modelarts.IsLatestVersion,
  322. })
  323. if err != nil {
  324. ctx.ServerError("Cloudbrain", err)
  325. return
  326. }
  327. for i, task := range tasks {
  328. tasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  329. tasks[i].CanModify = cloudbrain.CanModifyJob(ctx, &task.Cloudbrain)
  330. tasks[i].ComputeResource = models.NPUResource
  331. }
  332. pager := context.NewPagination(int(count), setting.UI.IssuePagingNum, page, 5)
  333. pager.SetDefaultParams(ctx)
  334. ctx.Data["Page"] = pager
  335. ctx.Data["PageIsCloudBrain"] = true
  336. ctx.Data["Tasks"] = tasks
  337. ctx.Data["CanCreate"] = cloudbrain.CanCreateOrDebugJob(ctx)
  338. ctx.Data["RepoIsEmpty"] = repo.IsEmpty
  339. ctx.HTML(200, tplModelArtsTrainJobIndex)
  340. }
  341. func TrainJobNew(ctx *context.Context) {
  342. err := trainJobNewDataPrepare(ctx)
  343. if err != nil {
  344. ctx.ServerError("get new train-job info failed", err)
  345. return
  346. }
  347. ctx.HTML(200, tplModelArtsTrainJobNew)
  348. }
  349. func trainJobNewDataPrepare(ctx *context.Context) error {
  350. ctx.Data["PageIsCloudBrain"] = true
  351. //can, err := canUserCreateTrainJob(ctx.User.ID)
  352. //if err != nil {
  353. // ctx.ServerError("canUserCreateTrainJob", err)
  354. // return
  355. //}
  356. //
  357. //if !can {
  358. // log.Error("the user can not create train-job")
  359. // ctx.ServerError("the user can not create train-job", fmt.Errorf("the user can not create train-job"))
  360. // return
  361. //}
  362. t := time.Now()
  363. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  364. ctx.Data["job_name"] = jobName
  365. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  366. if err != nil {
  367. ctx.ServerError("GetAllUserAttachments failed:", err)
  368. return err
  369. }
  370. ctx.Data["attachments"] = attachs
  371. var resourcePools modelarts.ResourcePool
  372. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  373. ctx.ServerError("json.Unmarshal failed:", err)
  374. return err
  375. }
  376. ctx.Data["resource_pools"] = resourcePools.Info
  377. var engines modelarts.Engine
  378. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  379. ctx.ServerError("json.Unmarshal failed:", err)
  380. return err
  381. }
  382. ctx.Data["engines"] = engines.Info
  383. var versionInfos modelarts.VersionInfo
  384. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  385. ctx.ServerError("json.Unmarshal failed:", err)
  386. return err
  387. }
  388. ctx.Data["engine_versions"] = versionInfos.Version
  389. var flavorInfos modelarts.Flavor
  390. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  391. ctx.ServerError("json.Unmarshal failed:", err)
  392. return err
  393. }
  394. ctx.Data["flavor_infos"] = flavorInfos.Info
  395. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  396. ctx.Data["train_url"] = outputObsPath
  397. ctx.Data["params"] = ""
  398. ctx.Data["branchName"] = ctx.Repo.BranchName
  399. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  400. if err != nil {
  401. ctx.ServerError("getConfigList failed:", err)
  402. return err
  403. }
  404. ctx.Data["config_list"] = configList.ParaConfigs
  405. return nil
  406. }
  407. func trainJobErrorNewDataPrepare(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) error {
  408. ctx.Data["PageIsCloudBrain"] = true
  409. //can, err := canUserCreateTrainJob(ctx.User.ID)
  410. //if err != nil {
  411. // ctx.ServerError("canUserCreateTrainJob", err)
  412. // return
  413. //}
  414. //
  415. //if !can {
  416. // log.Error("the user can not create train-job")
  417. // ctx.ServerError("the user can not create train-job", fmt.Errorf("the user can not create train-job"))
  418. // return
  419. //}
  420. t := time.Now()
  421. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  422. ctx.Data["job_name"] = jobName
  423. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  424. if err != nil {
  425. ctx.ServerError("GetAllUserAttachments failed:", err)
  426. return err
  427. }
  428. ctx.Data["attachments"] = attachs
  429. var resourcePools modelarts.ResourcePool
  430. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  431. ctx.ServerError("json.Unmarshal failed:", err)
  432. return err
  433. }
  434. ctx.Data["resource_pools"] = resourcePools.Info
  435. var engines modelarts.Engine
  436. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  437. ctx.ServerError("json.Unmarshal failed:", err)
  438. return err
  439. }
  440. ctx.Data["engines"] = engines.Info
  441. var versionInfos modelarts.VersionInfo
  442. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  443. ctx.ServerError("json.Unmarshal failed:", err)
  444. return err
  445. }
  446. ctx.Data["engine_versions"] = versionInfos.Version
  447. var flavorInfos modelarts.Flavor
  448. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  449. ctx.ServerError("json.Unmarshal failed:", err)
  450. return err
  451. }
  452. ctx.Data["flavor_infos"] = flavorInfos.Info
  453. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  454. ctx.Data["train_url"] = outputObsPath
  455. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  456. if err != nil {
  457. ctx.ServerError("getConfigList failed:", err)
  458. return err
  459. }
  460. var Parameters modelarts.Parameters
  461. if err = json.Unmarshal([]byte(form.Params), &Parameters); err != nil {
  462. ctx.ServerError("json.Unmarshal failed:", err)
  463. return err
  464. }
  465. ctx.Data["params"] = Parameters.Parameter
  466. ctx.Data["config_list"] = configList.ParaConfigs
  467. ctx.Data["bootFile"] = form.BootFile
  468. ctx.Data["uuid"] = form.Attachment
  469. ctx.Data["branch_name"] = form.BranchName
  470. return nil
  471. }
  472. func TrainJobNewVersion(ctx *context.Context) {
  473. err := trainJobNewVersionDataPrepare(ctx)
  474. if err != nil {
  475. ctx.ServerError("get new train-job info failed", err)
  476. return
  477. }
  478. ctx.HTML(200, tplModelArtsTrainJobVersionNew)
  479. }
  480. func trainJobNewVersionDataPrepare(ctx *context.Context) error {
  481. ctx.Data["PageIsCloudBrain"] = true
  482. var jobID = ctx.Params(":jobid")
  483. var versionName = ctx.Query("version_name")
  484. // canNewJob, err := canUserCreateTrainJobVersion(ctx, jobID, versionName)
  485. // if err != nil {
  486. // ctx.ServerError("canNewJob can info failed", err)
  487. // return err
  488. // }
  489. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  490. if err != nil {
  491. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  492. return err
  493. }
  494. t := time.Now()
  495. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  496. ctx.Data["job_name"] = task.JobName
  497. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  498. if err != nil {
  499. ctx.ServerError("GetAllUserAttachments failed:", err)
  500. return err
  501. }
  502. ctx.Data["attachments"] = attachs
  503. var resourcePools modelarts.ResourcePool
  504. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  505. ctx.ServerError("json.Unmarshal failed:", err)
  506. return err
  507. }
  508. ctx.Data["resource_pools"] = resourcePools.Info
  509. var engines modelarts.Engine
  510. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  511. ctx.ServerError("json.Unmarshal failed:", err)
  512. return err
  513. }
  514. ctx.Data["engines"] = engines.Info
  515. var versionInfos modelarts.VersionInfo
  516. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  517. ctx.ServerError("json.Unmarshal failed:", err)
  518. return err
  519. }
  520. ctx.Data["engine_versions"] = versionInfos.Version
  521. var flavorInfos modelarts.Flavor
  522. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  523. ctx.ServerError("json.Unmarshal failed:", err)
  524. return err
  525. }
  526. ctx.Data["flavor_infos"] = flavorInfos.Info
  527. var Parameters modelarts.Parameters
  528. if err = json.Unmarshal([]byte(task.Parameters), &Parameters); err != nil {
  529. ctx.ServerError("json.Unmarshal failed:", err)
  530. return err
  531. }
  532. ctx.Data["params"] = Parameters.Parameter
  533. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  534. ctx.Data["train_url"] = outputObsPath
  535. branches, _, err := ctx.Repo.GitRepo.GetBranches(0, 0)
  536. if err != nil {
  537. ctx.ServerError("GetBranches error:", err)
  538. return err
  539. }
  540. ctx.Data["branches"] = branches
  541. ctx.Data["branch_name"] = task.BranchName
  542. ctx.Data["description"] = task.Description
  543. ctx.Data["boot_file"] = task.BootFile
  544. ctx.Data["dataset_name"] = task.DatasetName
  545. ctx.Data["work_server_number"] = task.WorkServerNumber
  546. ctx.Data["flavor_name"] = task.FlavorName
  547. ctx.Data["engine_name"] = task.EngineName
  548. ctx.Data["uuid"] = task.Uuid
  549. ctx.Data["flavor_code"] = task.FlavorCode
  550. ctx.Data["engine_id"] = task.EngineID
  551. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  552. if err != nil {
  553. ctx.ServerError("getConfigList failed:", err)
  554. return err
  555. }
  556. ctx.Data["config_list"] = configList.ParaConfigs
  557. return nil
  558. }
  559. func versionErrorDataPrepare(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) error {
  560. ctx.Data["PageIsCloudBrain"] = true
  561. var jobID = ctx.Params(":jobid")
  562. // var versionName = ctx.Params(":version-name")
  563. var versionName = ctx.Query("version_name")
  564. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  565. if err != nil {
  566. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  567. return err
  568. }
  569. t := time.Now()
  570. var jobName = cutString(ctx.User.Name, 5) + t.Format("2006010215") + strconv.Itoa(int(t.Unix()))[5:]
  571. ctx.Data["job_name"] = task.JobName
  572. attachs, err := models.GetModelArtsTrainAttachments(ctx.User.ID)
  573. if err != nil {
  574. ctx.ServerError("GetAllUserAttachments failed:", err)
  575. return err
  576. }
  577. ctx.Data["attachments"] = attachs
  578. var resourcePools modelarts.ResourcePool
  579. if err = json.Unmarshal([]byte(setting.ResourcePools), &resourcePools); err != nil {
  580. ctx.ServerError("json.Unmarshal failed:", err)
  581. return err
  582. }
  583. ctx.Data["resource_pools"] = resourcePools.Info
  584. var engines modelarts.Engine
  585. if err = json.Unmarshal([]byte(setting.Engines), &engines); err != nil {
  586. ctx.ServerError("json.Unmarshal failed:", err)
  587. return err
  588. }
  589. ctx.Data["engines"] = engines.Info
  590. var versionInfos modelarts.VersionInfo
  591. if err = json.Unmarshal([]byte(setting.EngineVersions), &versionInfos); err != nil {
  592. ctx.ServerError("json.Unmarshal failed:", err)
  593. return err
  594. }
  595. ctx.Data["engine_versions"] = versionInfos.Version
  596. var flavorInfos modelarts.Flavor
  597. if err = json.Unmarshal([]byte(setting.TrainJobFLAVORINFOS), &flavorInfos); err != nil {
  598. ctx.ServerError("json.Unmarshal failed:", err)
  599. return err
  600. }
  601. ctx.Data["flavor_infos"] = flavorInfos.Info
  602. var Parameters modelarts.Parameters
  603. if err = json.Unmarshal([]byte(form.Params), &Parameters); err != nil {
  604. ctx.ServerError("json.Unmarshal failed:", err)
  605. return err
  606. }
  607. ctx.Data["params"] = Parameters.Parameter
  608. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath
  609. ctx.Data["train_url"] = outputObsPath
  610. branches, _, err := ctx.Repo.GitRepo.GetBranches(0, 0)
  611. if err != nil {
  612. ctx.ServerError("GetBranches error:", err)
  613. return err
  614. }
  615. ctx.Data["branches"] = branches
  616. ctx.Data["description"] = form.Description
  617. ctx.Data["dataset_name"] = task.DatasetName
  618. ctx.Data["work_server_number"] = form.WorkServerNumber
  619. ctx.Data["flavor_name"] = form.FlavorName
  620. ctx.Data["engine_name"] = form.EngineName
  621. ctx.Data["flavor_code"] = task.FlavorCode
  622. ctx.Data["engine_id"] = task.EngineID
  623. ctx.Data["version_name"] = form.VersionName
  624. ctx.Data["bootFile"] = form.BootFile
  625. ctx.Data["uuid"] = form.Attachment
  626. ctx.Data["branch_name"] = form.BranchName
  627. configList, err := getConfigList(modelarts.PerPage, 1, modelarts.SortByCreateTime, "desc", "", modelarts.ConfigTypeCustom)
  628. if err != nil {
  629. ctx.ServerError("getConfigList failed:", err)
  630. return err
  631. }
  632. ctx.Data["config_list"] = configList.ParaConfigs
  633. return nil
  634. }
  635. func TrainJobCreate(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) {
  636. ctx.Data["PageIsTrainJob"] = true
  637. VersionOutputPath := modelarts.GetVersionOutputPathByTotalVersionCount(modelarts.TotalVersionCount)
  638. jobName := form.JobName
  639. uuid := form.Attachment
  640. description := form.Description
  641. workServerNumber := form.WorkServerNumber
  642. engineID := form.EngineID
  643. bootFile := form.BootFile
  644. flavorCode := form.Flavor
  645. params := form.Params
  646. poolID := form.PoolID
  647. isSaveParam := form.IsSaveParam
  648. repo := ctx.Repo.Repository
  649. codeLocalPath := setting.JobPath + jobName + modelarts.CodePath
  650. codeObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.CodePath
  651. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath + VersionOutputPath + "/"
  652. logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath + VersionOutputPath + "/"
  653. dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/"
  654. branch_name := form.BranchName
  655. isLatestVersion := modelarts.IsLatestVersion
  656. FlavorName := form.FlavorName
  657. VersionCount := modelarts.VersionCount
  658. EngineName := form.EngineName
  659. count, err := models.GetCloudbrainTrainJobCountByUserID(ctx.User.ID)
  660. if err != nil {
  661. log.Error("GetCloudbrainTrainJobCountByUserID failed:%v", err, ctx.Data["MsgID"])
  662. trainJobErrorNewDataPrepare(ctx, form)
  663. ctx.RenderWithErr("system error", tplModelArtsTrainJobNew, &form)
  664. return
  665. } else {
  666. if count >= 1 {
  667. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  668. trainJobErrorNewDataPrepare(ctx, form)
  669. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsTrainJobNew, &form)
  670. return
  671. }
  672. }
  673. if err := paramCheckCreateTrainJob(form); err != nil {
  674. log.Error("paramCheckCreateTrainJob failed:(%v)", err)
  675. trainJobErrorNewDataPrepare(ctx, form)
  676. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobNew, &form)
  677. return
  678. }
  679. // attach, err := models.GetAttachmentByUUID(uuid)
  680. // if err != nil {
  681. // log.Error("GetAttachmentByUUID(%s) failed:%v", uuid, err.Error())
  682. // return
  683. // }
  684. //todo: del the codeLocalPath
  685. // _, err := ioutil.ReadDir(codeLocalPath)
  686. // if err == nil {
  687. // os.RemoveAll(codeLocalPath)
  688. // }
  689. os.RemoveAll(codeLocalPath)
  690. gitRepo, _ := git.OpenRepository(repo.RepoPath())
  691. commitID, _ := gitRepo.GetBranchCommitID(branch_name)
  692. if err := git.Clone(repo.RepoPath(), codeLocalPath, git.CloneRepoOptions{
  693. Branch: branch_name,
  694. }); err != nil {
  695. log.Error("创建任务失败,服务器超时!: %s (%v)", repo.FullName(), err)
  696. trainJobErrorNewDataPrepare(ctx, form)
  697. ctx.RenderWithErr("创建任务失败,服务器超时!", tplModelArtsTrainJobNew, &form)
  698. return
  699. }
  700. //todo: upload code (send to file_server todo this work?)
  701. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.OutputPath + VersionOutputPath + "/"); err != nil {
  702. log.Error("Failed to obsMkdir_output: %s (%v)", repo.FullName(), err)
  703. trainJobErrorNewDataPrepare(ctx, form)
  704. ctx.RenderWithErr("Failed to obsMkdir_output", tplModelArtsTrainJobNew, &form)
  705. return
  706. }
  707. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.LogPath + VersionOutputPath + "/"); err != nil {
  708. log.Error("Failed to obsMkdir_log: %s (%v)", repo.FullName(), err)
  709. trainJobErrorNewDataPrepare(ctx, form)
  710. ctx.RenderWithErr("Failed to obsMkdir_log", tplModelArtsTrainJobNew, &form)
  711. return
  712. }
  713. // parentDir := VersionOutputPath + "/"
  714. if err := uploadCodeToObs(codeLocalPath, jobName, ""); err != nil {
  715. // if err := uploadCodeToObs(codeLocalPath, jobName, parentDir); err != nil {
  716. log.Error("Failed to uploadCodeToObs: %s (%v)", repo.FullName(), err)
  717. trainJobErrorNewDataPrepare(ctx, form)
  718. ctx.RenderWithErr("Failed to uploadCodeToObs", tplModelArtsTrainJobNew, &form)
  719. return
  720. }
  721. //todo: del local code?
  722. var parameters models.Parameters
  723. param := make([]models.Parameter, 0)
  724. param = append(param, models.Parameter{
  725. Label: modelarts.TrainUrl,
  726. Value: outputObsPath,
  727. }, models.Parameter{
  728. Label: modelarts.DataUrl,
  729. Value: dataPath,
  730. })
  731. if len(params) != 0 {
  732. err := json.Unmarshal([]byte(params), &parameters)
  733. if err != nil {
  734. log.Error("Failed to Unmarshal params: %s (%v)", params, err)
  735. trainJobErrorNewDataPrepare(ctx, form)
  736. ctx.RenderWithErr("运行参数错误", tplModelArtsTrainJobNew, &form)
  737. return
  738. }
  739. for _, parameter := range parameters.Parameter {
  740. if parameter.Label != modelarts.TrainUrl && parameter.Label != modelarts.DataUrl {
  741. param = append(param, models.Parameter{
  742. Label: parameter.Label,
  743. Value: parameter.Value,
  744. })
  745. }
  746. }
  747. }
  748. //save param config
  749. if isSaveParam == "on" {
  750. if form.ParameterTemplateName == "" {
  751. log.Error("ParameterTemplateName is empty")
  752. trainJobNewDataPrepare(ctx)
  753. ctx.RenderWithErr("保存作业参数时,作业参数名称不能为空", tplModelArtsTrainJobNew, &form)
  754. return
  755. }
  756. _, err := modelarts.CreateTrainJobConfig(models.CreateConfigParams{
  757. ConfigName: form.ParameterTemplateName,
  758. Description: form.PrameterDescription,
  759. DataUrl: dataPath,
  760. AppUrl: codeObsPath,
  761. BootFileUrl: codeObsPath + bootFile,
  762. TrainUrl: outputObsPath,
  763. Flavor: models.Flavor{
  764. Code: flavorCode,
  765. },
  766. WorkServerNum: workServerNumber,
  767. EngineID: int64(engineID),
  768. LogUrl: logObsPath,
  769. PoolID: poolID,
  770. Parameter: param,
  771. })
  772. if err != nil {
  773. log.Error("Failed to CreateTrainJobConfig: %v", err)
  774. trainJobErrorNewDataPrepare(ctx, form)
  775. ctx.RenderWithErr("保存作业参数失败:"+err.Error(), tplModelArtsTrainJobNew, &form)
  776. return
  777. }
  778. }
  779. req := &modelarts.GenerateTrainJobReq{
  780. JobName: jobName,
  781. DataUrl: dataPath,
  782. Description: description,
  783. CodeObsPath: codeObsPath,
  784. BootFileUrl: codeObsPath + bootFile,
  785. BootFile: bootFile,
  786. TrainUrl: outputObsPath,
  787. FlavorCode: flavorCode,
  788. WorkServerNumber: workServerNumber,
  789. EngineID: int64(engineID),
  790. LogUrl: logObsPath,
  791. PoolID: poolID,
  792. Uuid: uuid,
  793. Parameters: parameters.Parameter,
  794. CommitID: commitID,
  795. IsLatestVersion: isLatestVersion,
  796. BranchName: branch_name,
  797. Params: form.Params,
  798. FlavorName: FlavorName,
  799. EngineName: EngineName,
  800. VersionCount: VersionCount,
  801. TotalVersionCount: modelarts.TotalVersionCount,
  802. }
  803. //将params转换Parameters.Parameter,出错时返回给前端
  804. var Parameters modelarts.Parameters
  805. if err := json.Unmarshal([]byte(params), &Parameters); err != nil {
  806. ctx.ServerError("json.Unmarshal failed:", err)
  807. return
  808. }
  809. err = modelarts.GenerateTrainJob(ctx, req)
  810. if err != nil {
  811. log.Error("GenerateTrainJob failed:%v", err.Error())
  812. trainJobErrorNewDataPrepare(ctx, form)
  813. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobNew, &form)
  814. return
  815. }
  816. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  817. }
  818. func TrainJobCreateVersion(ctx *context.Context, form auth.CreateModelArtsTrainJobForm) {
  819. ctx.Data["PageIsTrainJob"] = true
  820. var jobID = ctx.Params(":jobid")
  821. count, err := models.GetCloudbrainTrainJobCountByUserID(ctx.User.ID)
  822. if err != nil {
  823. log.Error("GetCloudbrainTrainJobCountByUserID failed:%v", err, ctx.Data["MsgID"])
  824. versionErrorDataPrepare(ctx, form)
  825. ctx.RenderWithErr("system error", tplModelArtsTrainJobVersionNew, &form)
  826. return
  827. } else {
  828. if count >= 1 {
  829. log.Error("the user already has running or waiting task", ctx.Data["MsgID"])
  830. versionErrorDataPrepare(ctx, form)
  831. ctx.RenderWithErr("you have already a running or waiting task, can not create more", tplModelArtsTrainJobVersionNew, &form)
  832. return
  833. }
  834. }
  835. latestTask, err := models.GetCloudbrainByJobIDAndIsLatestVersion(jobID, modelarts.IsLatestVersion)
  836. if err != nil {
  837. ctx.ServerError("GetCloudbrainByJobIDAndIsLatestVersion faild:", err)
  838. return
  839. }
  840. VersionOutputPath := modelarts.GetVersionOutputPathByTotalVersionCount(latestTask.TotalVersionCount + 1)
  841. jobName := form.JobName
  842. uuid := form.Attachment
  843. description := form.Description
  844. workServerNumber := form.WorkServerNumber
  845. engineID := form.EngineID
  846. bootFile := form.BootFile
  847. flavorCode := form.Flavor
  848. params := form.Params
  849. poolID := form.PoolID
  850. isSaveParam := form.IsSaveParam
  851. repo := ctx.Repo.Repository
  852. codeLocalPath := setting.JobPath + jobName + modelarts.CodePath
  853. codeObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.CodePath + VersionOutputPath + "/"
  854. outputObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.OutputPath + VersionOutputPath + "/"
  855. logObsPath := "/" + setting.Bucket + modelarts.JobPath + jobName + modelarts.LogPath + VersionOutputPath + "/"
  856. dataPath := "/" + setting.Bucket + "/" + setting.BasePath + path.Join(uuid[0:1], uuid[1:2]) + "/" + uuid + uuid + "/"
  857. branch_name := form.BranchName
  858. PreVersionName := form.VersionName
  859. FlavorName := form.FlavorName
  860. EngineName := form.EngineName
  861. isLatestVersion := modelarts.IsLatestVersion
  862. //判断权限
  863. canNewJob, _ := canUserCreateTrainJobVersion(ctx, latestTask.UserID)
  864. if !canNewJob {
  865. ctx.RenderWithErr("user cann't new trainjob", tplModelArtsTrainJobVersionNew, &form)
  866. return
  867. }
  868. if err := paramCheckCreateTrainJob(form); err != nil {
  869. log.Error("paramCheckCreateTrainJob failed:(%v)", err)
  870. versionErrorDataPrepare(ctx, form)
  871. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  872. return
  873. }
  874. // attach, err := models.GetAttachmentByUUID(uuid)
  875. // if err != nil {
  876. // log.Error("GetAttachmentByUUID(%s) failed:%v", uuid, err.Error())
  877. // return
  878. // }
  879. //todo: del the codeLocalPath
  880. // _, err = ioutil.ReadDir(codeLocalPath)
  881. // if err == nil {
  882. // os.RemoveAll(codeLocalPath)
  883. // }
  884. os.RemoveAll(codeLocalPath)
  885. gitRepo, _ := git.OpenRepository(repo.RepoPath())
  886. commitID, _ := gitRepo.GetBranchCommitID(branch_name)
  887. if err := git.Clone(repo.RepoPath(), codeLocalPath, git.CloneRepoOptions{
  888. Branch: branch_name,
  889. }); err != nil {
  890. log.Error("创建任务失败,任务名称已存在!: %s (%v)", repo.FullName(), err)
  891. versionErrorDataPrepare(ctx, form)
  892. ctx.RenderWithErr("创建任务失败,任务名称已存在!", tplModelArtsTrainJobVersionNew, &form)
  893. return
  894. }
  895. //todo: upload code (send to file_server todo this work?)
  896. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.OutputPath + VersionOutputPath + "/"); err != nil {
  897. log.Error("Failed to obsMkdir_output: %s (%v)", repo.FullName(), err)
  898. versionErrorDataPrepare(ctx, form)
  899. ctx.RenderWithErr("Failed to obsMkdir_output", tplModelArtsTrainJobVersionNew, &form)
  900. return
  901. }
  902. if err := obsMkdir(setting.CodePathPrefix + jobName + modelarts.LogPath + VersionOutputPath + "/"); err != nil {
  903. log.Error("Failed to obsMkdir_log: %s (%v)", repo.FullName(), err)
  904. versionErrorDataPrepare(ctx, form)
  905. ctx.RenderWithErr("Failed to obsMkdir_log", tplModelArtsTrainJobVersionNew, &form)
  906. return
  907. }
  908. parentDir := VersionOutputPath + "/"
  909. // parentDir := ""
  910. // if err := uploadCodeToObs(codeLocalPath, jobName, ""); err != nil {
  911. if err := uploadCodeToObs(codeLocalPath, jobName, parentDir); err != nil {
  912. log.Error("Failed to uploadCodeToObs: %s (%v)", repo.FullName(), err)
  913. versionErrorDataPrepare(ctx, form)
  914. ctx.RenderWithErr("Failed to uploadCodeToObs", tplModelArtsTrainJobVersionNew, &form)
  915. return
  916. }
  917. //todo: del local code?
  918. var parameters models.Parameters
  919. param := make([]models.Parameter, 0)
  920. param = append(param, models.Parameter{
  921. Label: modelarts.TrainUrl,
  922. Value: outputObsPath,
  923. }, models.Parameter{
  924. Label: modelarts.DataUrl,
  925. Value: dataPath,
  926. })
  927. if len(params) != 0 {
  928. err := json.Unmarshal([]byte(params), &parameters)
  929. if err != nil {
  930. log.Error("Failed to Unmarshal params: %s (%v)", params, err)
  931. versionErrorDataPrepare(ctx, form)
  932. ctx.RenderWithErr("运行参数错误", tplModelArtsTrainJobVersionNew, &form)
  933. return
  934. }
  935. for _, parameter := range parameters.Parameter {
  936. if parameter.Label != modelarts.TrainUrl && parameter.Label != modelarts.DataUrl {
  937. param = append(param, models.Parameter{
  938. Label: parameter.Label,
  939. Value: parameter.Value,
  940. })
  941. }
  942. }
  943. }
  944. //save param config
  945. if isSaveParam == "on" {
  946. if form.ParameterTemplateName == "" {
  947. log.Error("ParameterTemplateName is empty")
  948. versionErrorDataPrepare(ctx, form)
  949. ctx.RenderWithErr("保存作业参数时,作业参数名称不能为空", tplModelArtsTrainJobVersionNew, &form)
  950. return
  951. }
  952. _, err := modelarts.CreateTrainJobConfig(models.CreateConfigParams{
  953. ConfigName: form.ParameterTemplateName,
  954. Description: form.PrameterDescription,
  955. DataUrl: dataPath,
  956. AppUrl: codeObsPath,
  957. BootFileUrl: codeObsPath + bootFile,
  958. TrainUrl: outputObsPath,
  959. Flavor: models.Flavor{
  960. Code: flavorCode,
  961. },
  962. WorkServerNum: workServerNumber,
  963. EngineID: int64(engineID),
  964. LogUrl: logObsPath,
  965. PoolID: poolID,
  966. Parameter: parameters.Parameter,
  967. })
  968. if err != nil {
  969. log.Error("Failed to CreateTrainJobConfig: %v", err)
  970. versionErrorDataPrepare(ctx, form)
  971. ctx.RenderWithErr("保存作业参数失败:"+err.Error(), tplModelArtsTrainJobVersionNew, &form)
  972. return
  973. }
  974. }
  975. if err != nil {
  976. log.Error("getFlavorNameByEngineID(%s) failed:%v", engineID, err.Error())
  977. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  978. return
  979. }
  980. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, PreVersionName)
  981. if err != nil {
  982. log.Error("GetCloudbrainByJobIDAndVersionName(%s) failed:%v", jobID, err.Error())
  983. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  984. return
  985. }
  986. req := &modelarts.GenerateTrainJobReq{
  987. JobName: task.JobName,
  988. DataUrl: dataPath,
  989. Description: description,
  990. CodeObsPath: codeObsPath,
  991. BootFileUrl: codeObsPath + bootFile,
  992. BootFile: bootFile,
  993. TrainUrl: outputObsPath,
  994. FlavorCode: flavorCode,
  995. WorkServerNumber: workServerNumber,
  996. IsLatestVersion: isLatestVersion,
  997. EngineID: int64(engineID),
  998. LogUrl: logObsPath,
  999. PoolID: poolID,
  1000. Uuid: uuid,
  1001. Params: form.Params,
  1002. Parameters: parameters.Parameter,
  1003. PreVersionId: task.VersionID,
  1004. CommitID: commitID,
  1005. BranchName: branch_name,
  1006. FlavorName: FlavorName,
  1007. EngineName: EngineName,
  1008. PreVersionName: PreVersionName,
  1009. TotalVersionCount: latestTask.TotalVersionCount + 1,
  1010. }
  1011. err = modelarts.GenerateTrainJobVersion(ctx, req, jobID)
  1012. if err != nil {
  1013. log.Error("GenerateTrainJob failed:%v", err.Error())
  1014. versionErrorDataPrepare(ctx, form)
  1015. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobVersionNew, &form)
  1016. return
  1017. }
  1018. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job/" + jobID)
  1019. // ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1020. }
  1021. // readDir reads the directory named by dirname and returns
  1022. // a list of directory entries sorted by filename.
  1023. func readDir(dirname string) ([]os.FileInfo, error) {
  1024. f, err := os.Open(dirname)
  1025. if err != nil {
  1026. return nil, err
  1027. }
  1028. list, err := f.Readdir(100)
  1029. f.Close()
  1030. if err != nil {
  1031. //todo: can not upload empty folder
  1032. if err == io.EOF {
  1033. return nil, nil
  1034. }
  1035. return nil, err
  1036. }
  1037. //sort.Slice(list, func(i, j int) bool { return list[i].Name() < list[j].Name() })
  1038. return list, nil
  1039. }
  1040. func uploadCodeToObs(codePath, jobName, parentDir string) error {
  1041. files, err := readDir(codePath)
  1042. if err != nil {
  1043. log.Error("readDir(%s) failed: %s", codePath, err.Error())
  1044. return err
  1045. }
  1046. for _, file := range files {
  1047. if file.IsDir() {
  1048. input := &obs.PutObjectInput{}
  1049. input.Bucket = setting.Bucket
  1050. input.Key = parentDir + file.Name() + "/"
  1051. _, err = storage.ObsCli.PutObject(input)
  1052. if err != nil {
  1053. log.Error("PutObject(%s) failed: %s", input.Key, err.Error())
  1054. return err
  1055. }
  1056. if err = uploadCodeToObs(codePath+file.Name()+"/", jobName, parentDir+file.Name()+"/"); err != nil {
  1057. log.Error("uploadCodeToObs(%s) failed: %s", file.Name(), err.Error())
  1058. return err
  1059. }
  1060. } else {
  1061. input := &obs.PutFileInput{}
  1062. input.Bucket = setting.Bucket
  1063. input.Key = setting.CodePathPrefix + jobName + "/code/" + parentDir + file.Name()
  1064. input.SourceFile = codePath + file.Name()
  1065. _, err = storage.ObsCli.PutFile(input)
  1066. if err != nil {
  1067. log.Error("PutFile(%s) failed: %s", input.SourceFile, err.Error())
  1068. return err
  1069. }
  1070. }
  1071. }
  1072. return nil
  1073. }
  1074. func obsMkdir(dir string) error {
  1075. input := &obs.PutObjectInput{}
  1076. input.Bucket = setting.Bucket
  1077. input.Key = dir
  1078. _, err := storage.ObsCli.PutObject(input)
  1079. if err != nil {
  1080. log.Error("PutObject(%s) failed: %s", input.Key, err.Error())
  1081. return err
  1082. }
  1083. return nil
  1084. }
  1085. func paramCheckCreateTrainJob(form auth.CreateModelArtsTrainJobForm) error {
  1086. if !strings.HasSuffix(form.BootFile, ".py") {
  1087. log.Error("the boot file(%s) must be a python file", form.BootFile)
  1088. return errors.New("启动文件必须是python文件")
  1089. }
  1090. if form.WorkServerNumber > 25 || form.WorkServerNumber < 1 {
  1091. log.Error("the WorkServerNumber(%d) must be in (1,25)", form.WorkServerNumber)
  1092. return errors.New("计算节点数必须在1-25之间")
  1093. }
  1094. return nil
  1095. }
  1096. func TrainJobShow(ctx *context.Context) {
  1097. ctx.Data["PageIsCloudBrain"] = true
  1098. var jobID = ctx.Params(":jobid")
  1099. repo := ctx.Repo.Repository
  1100. page := ctx.QueryInt("page")
  1101. if page <= 0 {
  1102. page = 1
  1103. }
  1104. VersionListTasks, VersionListCount, err := models.CloudbrainsVersionList(&models.CloudbrainsOptions{
  1105. ListOptions: models.ListOptions{
  1106. Page: page,
  1107. PageSize: setting.UI.IssuePagingNum,
  1108. },
  1109. RepoID: repo.ID,
  1110. Type: models.TypeCloudBrainTwo,
  1111. JobType: string(models.JobTypeTrain),
  1112. JobID: jobID,
  1113. })
  1114. if err != nil {
  1115. log.Error("GetVersionListTasks(%s) failed:%v", jobID, err.Error())
  1116. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1117. return
  1118. }
  1119. //设置权限
  1120. canNewJob, err := canUserCreateTrainJobVersion(ctx, VersionListTasks[0].UserID)
  1121. if err != nil {
  1122. ctx.ServerError("canNewJob failed", err)
  1123. return
  1124. }
  1125. ctx.Data["canNewJob"] = canNewJob
  1126. //将运行参数转化为epoch_size = 3, device_target = Ascend的格式
  1127. for i, task := range VersionListTasks {
  1128. var parameters models.Parameters
  1129. err := json.Unmarshal([]byte(VersionListTasks[i].Parameters), &parameters)
  1130. if err != nil {
  1131. log.Error("Failed to Unmarshal Parameters: %s (%v)", VersionListTasks[i].Parameters, err)
  1132. trainJobNewDataPrepare(ctx)
  1133. return
  1134. }
  1135. if len(parameters.Parameter) > 0 {
  1136. paramTemp := ""
  1137. for _, Parameter := range parameters.Parameter {
  1138. param := Parameter.Label + " = " + Parameter.Value + "; "
  1139. paramTemp = paramTemp + param
  1140. }
  1141. VersionListTasks[i].Parameters = paramTemp[:len(paramTemp)-2]
  1142. } else {
  1143. VersionListTasks[i].Parameters = ""
  1144. }
  1145. VersionListTasks[i].CanDel = cloudbrain.CanDeleteJob(ctx, &task.Cloudbrain)
  1146. VersionListTasks[i].CanModify = cloudbrain.CanModifyJob(ctx, &task.Cloudbrain)
  1147. }
  1148. pager := context.NewPagination(VersionListCount, setting.UI.IssuePagingNum, page, 5)
  1149. pager.SetDefaultParams(ctx)
  1150. ctx.Data["Page"] = pager
  1151. ctx.Data["jobID"] = jobID
  1152. ctx.Data["jobName"] = VersionListTasks[0].JobName
  1153. ctx.Data["version_list_task"] = VersionListTasks
  1154. ctx.Data["version_list_count"] = VersionListCount
  1155. ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1156. }
  1157. func TrainJobGetLog(ctx *context.Context) {
  1158. ctx.Data["PageIsTrainJob"] = true
  1159. var jobID = ctx.Params(":jobid")
  1160. var logFileName = ctx.Query("file_name")
  1161. var baseLine = ctx.Query("base_line")
  1162. var order = ctx.Query("order")
  1163. if order != modelarts.OrderDesc && order != modelarts.OrderAsc {
  1164. log.Error("order(%s) check failed", order)
  1165. ctx.HTML(http.StatusBadRequest, tplModelArtsTrainJobShow)
  1166. return
  1167. }
  1168. task, err := models.GetCloudbrainByJobID(jobID)
  1169. if err != nil {
  1170. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  1171. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1172. return
  1173. }
  1174. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), baseLine, logFileName, order, modelarts.Lines)
  1175. if err != nil {
  1176. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  1177. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1178. return
  1179. }
  1180. ctx.Data["log"] = result
  1181. //ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1182. }
  1183. func trainJobGetLog(jobID string) (*models.GetTrainJobLogFileNamesResult, *models.GetTrainJobLogResult, error) {
  1184. task, err := models.GetCloudbrainByJobID(jobID)
  1185. if err != nil {
  1186. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  1187. return nil, nil, err
  1188. }
  1189. resultLogFile, err := modelarts.GetTrainJobLogFileNames(jobID, strconv.FormatInt(task.VersionID, 10))
  1190. if err != nil {
  1191. log.Error("GetTrainJobLogFileNames(%s) failed:%v", jobID, err.Error())
  1192. return nil, nil, err
  1193. }
  1194. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), "", resultLogFile.LogFileList[0], modelarts.OrderDesc, modelarts.Lines)
  1195. if err != nil {
  1196. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  1197. return nil, nil, err
  1198. }
  1199. return resultLogFile, result, err
  1200. }
  1201. func TrainJobDel(ctx *context.Context) {
  1202. var jobID = ctx.Params(":jobid")
  1203. repo := ctx.Repo.Repository
  1204. VersionListTasks, _, err := models.CloudbrainsVersionList(&models.CloudbrainsOptions{
  1205. RepoID: repo.ID,
  1206. Type: models.TypeCloudBrainTwo,
  1207. JobType: string(models.JobTypeTrain),
  1208. JobID: jobID,
  1209. })
  1210. if err != nil {
  1211. ctx.ServerError("get VersionListTasks failed", err)
  1212. return
  1213. }
  1214. //删除modelarts上的任务记录
  1215. _, err = modelarts.DelTrainJob(jobID)
  1216. if err != nil {
  1217. log.Error("DelTrainJob(%s) failed:%v", jobID, err.Error())
  1218. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil)
  1219. return
  1220. }
  1221. //删除数据库Cloudbrain表的记录
  1222. for _, task := range VersionListTasks {
  1223. err = models.DeleteJob(&task.Cloudbrain)
  1224. if err != nil {
  1225. ctx.ServerError("DeleteJob failed", err)
  1226. return
  1227. }
  1228. }
  1229. //删除存储
  1230. if len(VersionListTasks) > 0 {
  1231. DeleteJobStorage(VersionListTasks[0].JobName)
  1232. }
  1233. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  1234. }
  1235. func TrainJobStop(ctx *context.Context) {
  1236. var jobID = ctx.Params(":jobid")
  1237. task, err := models.GetCloudbrainByJobID(jobID)
  1238. if err != nil {
  1239. log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
  1240. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil)
  1241. return
  1242. }
  1243. _, err = modelarts.StopTrainJob(jobID, strconv.FormatInt(task.VersionID, 10))
  1244. if err != nil {
  1245. log.Error("StopTrainJob(%s) failed:%v", task.JobName, err.Error())
  1246. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobIndex, nil)
  1247. return
  1248. }
  1249. ctx.Redirect(setting.AppSubURL + ctx.Repo.RepoLink + "/modelarts/train-job")
  1250. }
  1251. func canUserCreateTrainJob(uid int64) (bool, error) {
  1252. org, err := models.GetOrgByName(setting.AllowedOrg)
  1253. if err != nil {
  1254. log.Error("get allowed org failed: ", setting.AllowedOrg)
  1255. return false, err
  1256. }
  1257. return org.IsOrgMember(uid)
  1258. }
  1259. func canUserCreateTrainJobVersion(ctx *context.Context, userID int64) (bool, error) {
  1260. if ctx == nil || ctx.User == nil {
  1261. log.Error("user unlogin!")
  1262. return false, nil
  1263. }
  1264. if userID == ctx.User.ID || ctx.User.IsAdmin {
  1265. return true, nil
  1266. } else {
  1267. log.Error("Only user itself and admin can new trainjob!")
  1268. return false, nil
  1269. }
  1270. }
  1271. func TrainJobGetConfigList(ctx *context.Context) {
  1272. ctx.Data["PageIsTrainJob"] = true
  1273. var jobID = ctx.Params(":jobid")
  1274. var logFileName = ctx.Query("file_name")
  1275. var baseLine = ctx.Query("base_line")
  1276. var order = ctx.Query("order")
  1277. if order != modelarts.OrderDesc && order != modelarts.OrderAsc {
  1278. log.Error("order(%s) check failed", order)
  1279. ctx.HTML(http.StatusBadRequest, tplModelArtsTrainJobShow)
  1280. return
  1281. }
  1282. task, err := models.GetCloudbrainByJobID(jobID)
  1283. if err != nil {
  1284. log.Error("GetCloudbrainByJobID(%s) failed:%v", jobID, err.Error())
  1285. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1286. return
  1287. }
  1288. result, err := modelarts.GetTrainJobLog(jobID, strconv.FormatInt(task.VersionID, 10), baseLine, logFileName, order, modelarts.Lines)
  1289. if err != nil {
  1290. log.Error("GetTrainJobLog(%s) failed:%v", jobID, err.Error())
  1291. ctx.RenderWithErr(err.Error(), tplModelArtsTrainJobShow, nil)
  1292. return
  1293. }
  1294. ctx.Data["log"] = result
  1295. //ctx.HTML(http.StatusOK, tplModelArtsTrainJobShow)
  1296. }
  1297. func getConfigList(perPage, page int, sortBy, order, searchContent, configType string) (*models.GetConfigListResult, error) {
  1298. var result models.GetConfigListResult
  1299. list, err := modelarts.GetConfigList(perPage, page, sortBy, order, searchContent, configType)
  1300. if err != nil {
  1301. log.Error("GetConfigList failed:", err)
  1302. return &result, err
  1303. }
  1304. for _, config := range list.ParaConfigs {
  1305. paraConfig, err := modelarts.GetParaConfig(config.ConfigName, configType)
  1306. if err != nil {
  1307. log.Error("GetParaConfig failed:", err)
  1308. return &result, err
  1309. }
  1310. config.Result = paraConfig
  1311. }
  1312. return list, nil
  1313. }
  1314. func ModelDownload(ctx *context.Context) {
  1315. var (
  1316. err error
  1317. )
  1318. var jobID = ctx.Params(":jobid")
  1319. versionName := ctx.Query("version_name")
  1320. parentDir := ctx.Query("parent_dir")
  1321. fileName := ctx.Query("file_name")
  1322. log.Info("DownloadSingleModelFile start.")
  1323. task, err := models.GetCloudbrainByJobIDAndVersionName(jobID, versionName)
  1324. if err != nil {
  1325. log.Error("GetCloudbrainByJobID(%s) failed:%v", task.JobName, err.Error())
  1326. return
  1327. }
  1328. path := strings.TrimPrefix(path.Join(setting.TrainJobModelPath, task.JobName, setting.OutPutPath, versionName, parentDir, fileName), "/")
  1329. log.Info("Download path is:%s", path)
  1330. url, err := storage.GetObsCreateSignedUrlByBucketAndKey(setting.Bucket, path)
  1331. if err != nil {
  1332. log.Error("GetObsCreateSignedUrl failed: %v", err.Error(), ctx.Data["msgID"])
  1333. ctx.ServerError("GetObsCreateSignedUrl", err)
  1334. return
  1335. }
  1336. http.Redirect(ctx.Resp, ctx.Req.Request, url, http.StatusMovedPermanently)
  1337. }
  1338. func DeleteJobStorage(jobName string) error {
  1339. //delete local
  1340. localJobPath := setting.JobPath + jobName
  1341. err := os.RemoveAll(localJobPath)
  1342. if err != nil {
  1343. log.Error("RemoveAll(%s) failed:%v", localJobPath, err)
  1344. }
  1345. //delete oss
  1346. dirPath := setting.CodePathPrefix + jobName + "/"
  1347. err = storage.ObsRemoveObject(setting.Bucket, dirPath)
  1348. if err != nil {
  1349. log.Error("ObsRemoveObject(%s) failed:%v", localJobPath, err)
  1350. }
  1351. return nil
  1352. }