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.

resty.go 46 kB

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
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
2 years ago
4 years ago
4 years ago
4 years ago
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
4 years ago
4 years ago
3 years ago
3 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
2 years ago
2 years ago
4 years ago
4 years ago
3 years ago
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
2 years ago
2 years ago
2 years ago
2 years ago
2 years ago
2 years ago
4 years ago
2 years ago
2 years ago
2 years ago
2 years ago
2 years ago
2 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
2 years ago
3 years ago
3 years ago
3 years ago
3 years ago
2 years ago
3 years ago
2 years ago
3 years ago
3 years ago
3 years ago
2 years ago
1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489
  1. package modelarts
  2. import (
  3. "crypto/tls"
  4. "encoding/json"
  5. "fmt"
  6. "net/http"
  7. "strconv"
  8. "code.gitea.io/gitea/models"
  9. "code.gitea.io/gitea/modules/log"
  10. "code.gitea.io/gitea/modules/setting"
  11. "github.com/go-resty/resty/v2"
  12. )
  13. var (
  14. restyClient *resty.Client
  15. HOST string
  16. TOKEN string
  17. )
  18. const (
  19. methodPassword = "password"
  20. urlGetToken = "/v3/auth/tokens"
  21. urlNotebook = "/demanager/instances"
  22. urlTrainJob = "/training-jobs"
  23. urlResourceSpecs = "/job/resource-specs"
  24. urlTrainJobConfig = "/training-job-configs"
  25. errorCodeExceedLimit = "ModelArts.0118"
  26. //notebook 2.0
  27. urlNotebook2 = "/notebooks"
  28. //error code
  29. modelartsIllegalToken = "ModelArts.6401"
  30. NotebookNotFound = "ModelArts.6404"
  31. NotebookNoPermission = "ModelArts.6407"
  32. NotebookInvalid = "ModelArts.6400"
  33. UnknownErrorPrefix = "UNKNOWN:"
  34. )
  35. func getRestyClient() *resty.Client {
  36. if restyClient == nil {
  37. restyClient = resty.New()
  38. restyClient.SetTLSClientConfig(&tls.Config{InsecureSkipVerify: true})
  39. }
  40. return restyClient
  41. }
  42. func checkSetting() {
  43. if len(HOST) != 0 && len(TOKEN) != 0 && restyClient != nil {
  44. return
  45. }
  46. err := getToken()
  47. if err != nil {
  48. log.Error("getToken failed:%v", err)
  49. }
  50. }
  51. func getToken() error {
  52. HOST = setting.ModelArtsHost
  53. client := getRestyClient()
  54. params := models.GetTokenParams{
  55. Auth: models.Auth{
  56. Identity: models.Identity{
  57. Methods: []string{methodPassword},
  58. Password: models.Password{
  59. User: models.NotebookUser{
  60. Name: setting.ModelArtsUsername,
  61. Password: setting.ModelArtsPassword,
  62. Domain: models.Domain{
  63. Name: setting.ModelArtsDomain,
  64. },
  65. },
  66. },
  67. },
  68. Scope: models.Scope{
  69. Project: models.Project{
  70. Name: setting.ProjectName,
  71. },
  72. },
  73. },
  74. }
  75. res, err := client.R().
  76. SetHeader("Content-Type", "application/json").
  77. SetBody(params).
  78. Post(setting.IamHost + urlGetToken)
  79. if err != nil {
  80. return fmt.Errorf("resty getToken: %v", err)
  81. }
  82. if res.StatusCode() != http.StatusCreated {
  83. return fmt.Errorf("getToken failed:%s", res.String())
  84. }
  85. TOKEN = res.Header().Get("X-Subject-Token")
  86. return nil
  87. }
  88. func CreateJob(createJobParams models.CreateNotebookParams) (*models.CreateNotebookResult, error) {
  89. checkSetting()
  90. client := getRestyClient()
  91. var result models.CreateNotebookResult
  92. retry := 0
  93. sendjob:
  94. res, err := client.R().
  95. SetHeader("Content-Type", "application/json").
  96. SetAuthToken(TOKEN).
  97. SetBody(createJobParams).
  98. SetResult(&result).
  99. Post(HOST + "/v1/" + setting.ProjectID + urlNotebook)
  100. if err != nil {
  101. return nil, fmt.Errorf("resty create notebook: %s", err)
  102. }
  103. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  104. retry++
  105. _ = getToken()
  106. goto sendjob
  107. }
  108. var response models.NotebookResult
  109. err = json.Unmarshal(res.Body(), &response)
  110. if err != nil {
  111. log.Error("json.Unmarshal failed: %s", err.Error())
  112. return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
  113. }
  114. if len(response.ErrorCode) != 0 {
  115. log.Error("createNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  116. if response.ErrorCode == errorCodeExceedLimit {
  117. response.ErrorMsg = "所选规格使用数量已超过最大配额限制。"
  118. }
  119. return &result, fmt.Errorf("createNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  120. }
  121. return &result, nil
  122. }
  123. func GetJob(jobID string) (*models.GetNotebookResult, error) {
  124. checkSetting()
  125. client := getRestyClient()
  126. var result models.GetNotebookResult
  127. retry := 0
  128. sendjob:
  129. res, err := client.R().
  130. SetHeader("Content-Type", "application/json").
  131. SetAuthToken(TOKEN).
  132. SetResult(&result).
  133. Get(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID)
  134. if err != nil {
  135. return nil, fmt.Errorf("resty GetJob: %v", err)
  136. }
  137. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  138. retry++
  139. _ = getToken()
  140. goto sendjob
  141. }
  142. var response models.NotebookResult
  143. err = json.Unmarshal(res.Body(), &response)
  144. if err != nil {
  145. log.Error("json.Unmarshal failed: %s", err.Error())
  146. return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
  147. }
  148. if len(response.ErrorCode) != 0 {
  149. log.Error("GetJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  150. return &result, fmt.Errorf("GetJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  151. }
  152. return &result, nil
  153. }
  154. func GetNotebook2(jobID string) (*models.GetNotebook2Result, error) {
  155. checkSetting()
  156. client := getRestyClient()
  157. var result models.GetNotebook2Result
  158. retry := 0
  159. sendjob:
  160. res, err := client.R().
  161. SetHeader("Content-Type", "application/json").
  162. SetAuthToken(TOKEN).
  163. SetResult(&result).
  164. Get(HOST + "/v1/" + setting.ProjectID + urlNotebook2 + "/" + jobID)
  165. if err != nil {
  166. return nil, fmt.Errorf("resty GetJob: %v", err)
  167. }
  168. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  169. retry++
  170. _ = getToken()
  171. goto sendjob
  172. }
  173. var response models.NotebookResult
  174. err = json.Unmarshal(res.Body(), &response)
  175. if err != nil {
  176. log.Error("json.Unmarshal failed: %s", err.Error())
  177. return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
  178. }
  179. if len(response.ErrorCode) != 0 {
  180. log.Error("GetJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  181. if response.ErrorCode == modelartsIllegalToken && retry < 1 {
  182. retry++
  183. _ = getToken()
  184. goto sendjob
  185. }
  186. return &result, fmt.Errorf("GetJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  187. }
  188. return &result, nil
  189. }
  190. func ManageNotebook(jobID string, param models.NotebookAction) (*models.NotebookActionResult, error) {
  191. checkSetting()
  192. client := getRestyClient()
  193. var result models.NotebookActionResult
  194. retry := 0
  195. sendjob:
  196. res, err := client.R().
  197. SetHeader("Content-Type", "application/json").
  198. SetBody(param).
  199. SetAuthToken(TOKEN).
  200. SetResult(&result).
  201. Post(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID + "/action")
  202. if err != nil {
  203. return &result, fmt.Errorf("resty StopJob: %v", err)
  204. }
  205. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  206. retry++
  207. _ = getToken()
  208. goto sendjob
  209. }
  210. var response models.NotebookResult
  211. err = json.Unmarshal(res.Body(), &response)
  212. if err != nil {
  213. log.Error("json.Unmarshal failed: %s", err.Error())
  214. return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
  215. }
  216. if len(response.ErrorCode) != 0 {
  217. log.Error("ManageNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  218. return &result, fmt.Errorf("ManageNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  219. }
  220. return &result, nil
  221. }
  222. func ManageNotebook2(jobID string, param models.NotebookAction) (*models.NotebookActionResult, error) {
  223. checkSetting()
  224. client := getRestyClient()
  225. var result models.NotebookActionResult
  226. retry := 0
  227. sendjob:
  228. res, err := client.R().
  229. SetHeader("Content-Type", "application/json").
  230. SetAuthToken(TOKEN).
  231. SetResult(&result).
  232. Post(HOST + "/v1/" + setting.ProjectID + urlNotebook2 + "/" + jobID + "/" + param.Action + "?duration=" + strconv.Itoa(AutoStopDurationMs))
  233. if err != nil {
  234. return &result, fmt.Errorf("resty ManageNotebook2: %v", err)
  235. }
  236. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  237. retry++
  238. _ = getToken()
  239. goto sendjob
  240. }
  241. var response models.NotebookResult
  242. err = json.Unmarshal(res.Body(), &response)
  243. if err != nil {
  244. log.Error("json.Unmarshal failed: %s", err.Error())
  245. return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
  246. }
  247. if res.StatusCode() == http.StatusBadGateway {
  248. return &result, fmt.Errorf(UnknownErrorPrefix+"createNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  249. }
  250. if len(response.ErrorCode) != 0 {
  251. log.Error("ManageNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  252. if response.ErrorCode == modelartsIllegalToken && retry < 1 {
  253. retry++
  254. _ = getToken()
  255. goto sendjob
  256. }
  257. return &result, fmt.Errorf("ManageNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  258. }
  259. return &result, nil
  260. }
  261. func DelNotebook(jobID string) (*models.NotebookDelResult, error) {
  262. checkSetting()
  263. client := getRestyClient()
  264. var result models.NotebookDelResult
  265. retry := 0
  266. sendjob:
  267. res, err := client.R().
  268. SetHeader("Content-Type", "application/json").
  269. SetAuthToken(TOKEN).
  270. SetResult(&result).
  271. Delete(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID)
  272. if err != nil {
  273. return &result, fmt.Errorf("resty DelJob: %v", err)
  274. }
  275. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  276. retry++
  277. _ = getToken()
  278. goto sendjob
  279. }
  280. var response models.NotebookResult
  281. err = json.Unmarshal(res.Body(), &response)
  282. if err != nil {
  283. log.Error("json.Unmarshal failed: %s", err.Error())
  284. return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
  285. }
  286. if len(response.ErrorCode) != 0 {
  287. log.Error("DelJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  288. return &result, fmt.Errorf("DelJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  289. }
  290. return &result, nil
  291. }
  292. func DelNotebook2(jobID string) (*models.NotebookDelResult, error) {
  293. checkSetting()
  294. client := getRestyClient()
  295. var result models.NotebookDelResult
  296. retry := 0
  297. sendjob:
  298. res, err := client.R().
  299. SetHeader("Content-Type", "application/json").
  300. SetAuthToken(TOKEN).
  301. SetResult(&result).
  302. Delete(HOST + "/v1/" + setting.ProjectID + urlNotebook2 + "/" + jobID)
  303. if err != nil {
  304. return &result, fmt.Errorf("resty DelJob: %v", err)
  305. }
  306. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  307. retry++
  308. _ = getToken()
  309. goto sendjob
  310. }
  311. var response models.NotebookResult
  312. err = json.Unmarshal(res.Body(), &response)
  313. if err != nil {
  314. log.Error("json.Unmarshal failed: %s", err.Error())
  315. return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
  316. }
  317. if len(response.ErrorCode) != 0 {
  318. log.Error("DelNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  319. if response.ErrorCode == modelartsIllegalToken && retry < 1 {
  320. retry++
  321. _ = getToken()
  322. goto sendjob
  323. }
  324. return &result, fmt.Errorf("DelNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  325. }
  326. return &result, nil
  327. }
  328. func DelJob(jobID string) (*models.NotebookDelResult, error) {
  329. checkSetting()
  330. client := getRestyClient()
  331. var result models.NotebookDelResult
  332. retry := 0
  333. sendjob:
  334. res, err := client.R().
  335. SetHeader("Content-Type", "application/json").
  336. SetAuthToken(TOKEN).
  337. SetResult(&result).
  338. Delete(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID)
  339. if err != nil {
  340. return &result, fmt.Errorf("resty DelJob: %v", err)
  341. }
  342. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  343. retry++
  344. _ = getToken()
  345. goto sendjob
  346. }
  347. var response models.NotebookResult
  348. err = json.Unmarshal(res.Body(), &response)
  349. if err != nil {
  350. log.Error("json.Unmarshal failed: %s", err.Error())
  351. return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
  352. }
  353. if len(response.ErrorCode) != 0 {
  354. log.Error("DelJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  355. return &result, fmt.Errorf("DelJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  356. }
  357. return &result, nil
  358. }
  359. func GetJobToken(jobID string) (*models.NotebookGetJobTokenResult, error) {
  360. checkSetting()
  361. client := getRestyClient()
  362. var result models.NotebookGetJobTokenResult
  363. retry := 0
  364. sendjob:
  365. res, err := client.R().
  366. SetHeader("Content-Type", "application/json").
  367. SetAuthToken(TOKEN).
  368. SetResult(&result).
  369. Get(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID + "/token")
  370. if err != nil {
  371. return &result, fmt.Errorf("resty GetJobToken: %v", err)
  372. }
  373. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  374. retry++
  375. _ = getToken()
  376. goto sendjob
  377. }
  378. var response models.NotebookResult
  379. err = json.Unmarshal(res.Body(), &response)
  380. if err != nil {
  381. log.Error("json.Unmarshal failed: %s", err.Error())
  382. return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
  383. }
  384. if len(response.ErrorCode) != 0 {
  385. log.Error("GetJobToken failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  386. return &result, fmt.Errorf("GetJobToken failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  387. }
  388. return &result, nil
  389. }
  390. func createTrainJobUserImage(createJobParams models.CreateUserImageTrainJobParams) (*models.CreateTrainJobResult, error) {
  391. checkSetting()
  392. client := getRestyClient()
  393. var result models.CreateTrainJobResult
  394. retry := 0
  395. sendjob:
  396. res, err := client.R().
  397. SetHeader("Content-Type", "application/json").
  398. SetAuthToken(TOKEN).
  399. SetBody(createJobParams).
  400. SetResult(&result).
  401. Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob)
  402. if err != nil {
  403. return nil, fmt.Errorf("resty create train-job: %s", err)
  404. }
  405. req, _ := json.Marshal(createJobParams)
  406. log.Info("postapi json: %s", req)
  407. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  408. retry++
  409. _ = getToken()
  410. goto sendjob
  411. }
  412. if res.StatusCode() != http.StatusOK {
  413. var temp models.ErrorResult
  414. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  415. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  416. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  417. }
  418. log.Error("createTrainJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  419. bootFileErrorMsg := "Invalid OBS path '" + createJobParams.Config.BootFileUrl + "'."
  420. dataSetErrorMsg := "Invalid OBS path '" + createJobParams.Config.DataUrl + "'."
  421. if temp.ErrorMsg == bootFileErrorMsg {
  422. log.Error("启动文件错误!createTrainJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  423. return &result, fmt.Errorf("启动文件错误!")
  424. }
  425. if temp.ErrorMsg == dataSetErrorMsg {
  426. log.Error("数据集错误!createTrainJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  427. return &result, fmt.Errorf("数据集错误!")
  428. }
  429. if res.StatusCode() == http.StatusBadGateway {
  430. return &result, fmt.Errorf(UnknownErrorPrefix+"createTrainJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  431. } else {
  432. return &result, fmt.Errorf("createTrainJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  433. }
  434. }
  435. if !result.IsSuccess {
  436. log.Error("createTrainJobUserImage failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  437. return &result, fmt.Errorf("createTrainJobUserImage failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  438. }
  439. return &result, nil
  440. }
  441. func createTrainJob(createJobParams models.CreateTrainJobParams) (*models.CreateTrainJobResult, error) {
  442. checkSetting()
  443. client := getRestyClient()
  444. var result models.CreateTrainJobResult
  445. retry := 0
  446. req, _ := json.Marshal(createJobParams)
  447. log.Info("postapi json: %s", req)
  448. sendjob:
  449. res, err := client.R().
  450. SetHeader("Content-Type", "application/json").
  451. SetAuthToken(TOKEN).
  452. SetBody(createJobParams).
  453. SetResult(&result).
  454. Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob)
  455. if err != nil {
  456. return nil, fmt.Errorf("resty create train-job: %s", err)
  457. }
  458. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  459. retry++
  460. _ = getToken()
  461. goto sendjob
  462. }
  463. if res.StatusCode() != http.StatusOK {
  464. var temp models.ErrorResult
  465. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  466. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  467. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  468. }
  469. log.Error("createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  470. bootFileErrorMsg := "Invalid OBS path '" + createJobParams.Config.BootFileUrl + "'."
  471. dataSetErrorMsg := "Invalid OBS path '" + createJobParams.Config.DataUrl + "'."
  472. if temp.ErrorMsg == bootFileErrorMsg {
  473. log.Error("启动文件错误!createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  474. return &result, fmt.Errorf("启动文件错误!")
  475. }
  476. if temp.ErrorMsg == dataSetErrorMsg {
  477. log.Error("数据集错误!createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  478. return &result, fmt.Errorf("数据集错误!")
  479. }
  480. if res.StatusCode() == http.StatusBadGateway {
  481. return &result, fmt.Errorf(UnknownErrorPrefix+"createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  482. } else {
  483. return &result, fmt.Errorf("createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  484. }
  485. }
  486. if !result.IsSuccess {
  487. log.Error("createTrainJob failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  488. return &result, fmt.Errorf("createTrainJob failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  489. }
  490. return &result, nil
  491. }
  492. func createTrainJobVersion(createJobVersionParams models.CreateTrainJobVersionParams, jobID string) (*models.CreateTrainJobResult, error) {
  493. checkSetting()
  494. client := getRestyClient()
  495. var result models.CreateTrainJobResult
  496. retry := 0
  497. sendjob:
  498. res, err := client.R().
  499. SetHeader("Content-Type", "application/json").
  500. SetAuthToken(TOKEN).
  501. SetBody(createJobVersionParams).
  502. SetResult(&result).
  503. Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions")
  504. if err != nil {
  505. return nil, fmt.Errorf("resty create train-job version: %s", err)
  506. }
  507. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  508. retry++
  509. _ = getToken()
  510. goto sendjob
  511. }
  512. if res.StatusCode() != http.StatusOK {
  513. var temp models.ErrorResult
  514. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  515. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  516. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  517. }
  518. log.Error("createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  519. bootFileErrorMsg := "Invalid OBS path '" + createJobVersionParams.Config.BootFileUrl + "'."
  520. dataSetErrorMsg := "Invalid OBS path '" + createJobVersionParams.Config.DataUrl + "'."
  521. if temp.ErrorMsg == bootFileErrorMsg {
  522. log.Error("启动文件错误!createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  523. return &result, fmt.Errorf("启动文件错误!")
  524. }
  525. if temp.ErrorMsg == dataSetErrorMsg {
  526. log.Error("数据集错误!createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  527. return &result, fmt.Errorf("数据集错误!")
  528. }
  529. if res.StatusCode() == http.StatusBadGateway {
  530. return &result, fmt.Errorf(UnknownErrorPrefix+"createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  531. } else {
  532. return &result, fmt.Errorf("createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  533. }
  534. }
  535. if !result.IsSuccess {
  536. log.Error("createTrainJobVersion failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  537. return &result, fmt.Errorf("createTrainJobVersion failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  538. }
  539. return &result, nil
  540. }
  541. func createTrainJobVersionUserImage(createJobVersionParams models.CreateTrainJobVersionUserImageParams, jobID string) (*models.CreateTrainJobResult, error) {
  542. checkSetting()
  543. client := getRestyClient()
  544. var result models.CreateTrainJobResult
  545. retry := 0
  546. sendjob:
  547. res, err := client.R().
  548. SetHeader("Content-Type", "application/json").
  549. SetAuthToken(TOKEN).
  550. SetBody(createJobVersionParams).
  551. SetResult(&result).
  552. Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions")
  553. if err != nil {
  554. return nil, fmt.Errorf("resty create train-job version: %s", err)
  555. }
  556. req, _ := json.Marshal(createJobVersionParams)
  557. log.Info("%s", req)
  558. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  559. retry++
  560. _ = getToken()
  561. goto sendjob
  562. }
  563. if res.StatusCode() != http.StatusOK {
  564. var temp models.ErrorResult
  565. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  566. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  567. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  568. }
  569. BootFileErrorMsg := "Invalid OBS path '" + createJobVersionParams.Config.BootFileUrl + "'."
  570. DataSetErrorMsg := "Invalid OBS path '" + createJobVersionParams.Config.DataUrl + "'."
  571. if temp.ErrorMsg == BootFileErrorMsg {
  572. log.Error("启动文件错误!createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  573. return &result, fmt.Errorf("启动文件错误!")
  574. }
  575. if temp.ErrorMsg == DataSetErrorMsg {
  576. log.Error("数据集错误!createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  577. return &result, fmt.Errorf("数据集错误!")
  578. }
  579. return &result, fmt.Errorf("createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  580. }
  581. if !result.IsSuccess {
  582. log.Error("createTrainJobVersion failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  583. return &result, fmt.Errorf("createTrainJobVersion failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  584. }
  585. return &result, nil
  586. }
  587. func GetResourceSpecs() (*models.GetResourceSpecsResult, error) {
  588. checkSetting()
  589. client := getRestyClient()
  590. var result models.GetResourceSpecsResult
  591. retry := 0
  592. sendjob:
  593. res, err := client.R().
  594. SetHeader("Content-Type", "application/json").
  595. SetAuthToken(TOKEN).
  596. SetResult(&result).
  597. Get(HOST + "/v1/" + setting.ProjectID + urlResourceSpecs)
  598. if err != nil {
  599. return nil, fmt.Errorf("resty GetResourceSpecs: %v", err)
  600. }
  601. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  602. retry++
  603. _ = getToken()
  604. goto sendjob
  605. }
  606. if res.StatusCode() != http.StatusOK {
  607. var temp models.ErrorResult
  608. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  609. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  610. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  611. }
  612. log.Error("GetResourceSpecs failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  613. return &result, fmt.Errorf("GetResourceSpecs failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  614. }
  615. if !result.IsSuccess {
  616. log.Error("GetResourceSpecs failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  617. return &result, fmt.Errorf("GetResourceSpecs failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  618. }
  619. return &result, nil
  620. }
  621. func CreateTrainJobConfig(req models.CreateConfigParams) (*models.CreateTrainJobConfigResult, error) {
  622. checkSetting()
  623. client := getRestyClient()
  624. var result models.CreateTrainJobConfigResult
  625. retry := 0
  626. sendjob:
  627. res, err := client.R().
  628. SetHeader("Content-Type", "application/json").
  629. SetAuthToken(TOKEN).
  630. SetBody(req).
  631. SetResult(&result).
  632. Post(HOST + "/v1/" + setting.ProjectID + urlTrainJobConfig)
  633. if err != nil {
  634. return nil, fmt.Errorf("resty CreateTrainJobConfig: %s", err)
  635. }
  636. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  637. retry++
  638. _ = getToken()
  639. goto sendjob
  640. }
  641. if res.StatusCode() != http.StatusOK {
  642. var temp models.ErrorResult
  643. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  644. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  645. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  646. }
  647. log.Error("CreateTrainJobConfig failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  648. return &result, fmt.Errorf("CreateTrainJobConfig failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  649. }
  650. if !result.IsSuccess {
  651. log.Error("CreateTrainJobConfig failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  652. return &result, fmt.Errorf("CreateTrainJobConfig failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  653. }
  654. return &result, nil
  655. }
  656. func GetConfigList(perPage, page int, sortBy, order, searchContent, configType string) (*models.GetConfigListResult, error) {
  657. checkSetting()
  658. client := getRestyClient()
  659. var result models.GetConfigListResult
  660. retry := 0
  661. sendjob:
  662. res, err := client.R().
  663. SetQueryParams(map[string]string{
  664. "per_page": strconv.Itoa(perPage),
  665. "page": strconv.Itoa(page),
  666. "sortBy": sortBy,
  667. "order": order,
  668. "search_content": searchContent,
  669. "config_type": configType,
  670. }).
  671. SetAuthToken(TOKEN).
  672. SetResult(&result).
  673. Get(HOST + "/v1/" + setting.ProjectID + urlTrainJobConfig)
  674. if err != nil {
  675. return nil, fmt.Errorf("resty GetConfigList: %v", err)
  676. }
  677. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  678. retry++
  679. _ = getToken()
  680. goto sendjob
  681. }
  682. if res.StatusCode() != http.StatusOK {
  683. var temp models.ErrorResult
  684. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  685. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  686. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  687. }
  688. log.Error("GetConfigList failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  689. return &result, fmt.Errorf("获取参数配置列表失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  690. }
  691. if !result.IsSuccess {
  692. log.Error("GetConfigList failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  693. return &result, fmt.Errorf("获取参数配置列表失败(%s): %s", result.ErrorCode, result.ErrorMsg)
  694. }
  695. return &result, nil
  696. }
  697. func GetParaConfig(configName, configType string) (models.GetConfigResult, error) {
  698. checkSetting()
  699. client := getRestyClient()
  700. var result models.GetConfigResult
  701. retry := 0
  702. sendjob:
  703. res, err := client.R().
  704. SetQueryParams(map[string]string{
  705. "config_type": configType,
  706. }).
  707. SetAuthToken(TOKEN).
  708. SetResult(&result).
  709. Get(HOST + "/v1/" + setting.ProjectID + urlTrainJobConfig + "/" + configName)
  710. if err != nil {
  711. return result, fmt.Errorf("resty GetParaConfig: %v", err)
  712. }
  713. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  714. retry++
  715. _ = getToken()
  716. goto sendjob
  717. }
  718. if res.StatusCode() != http.StatusOK {
  719. var temp models.ErrorResult
  720. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  721. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  722. return result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  723. }
  724. log.Error("GetParaConfig failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  725. return result, fmt.Errorf("获取参数配置详情失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  726. }
  727. if !result.IsSuccess {
  728. log.Error("GetParaConfig failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  729. return result, fmt.Errorf("获取参数配置详情失败(%s): %s", result.ErrorCode, result.ErrorMsg)
  730. }
  731. return result, nil
  732. }
  733. func GetTrainJob(jobID, versionID string) (*models.GetTrainJobResult, error) {
  734. checkSetting()
  735. client := getRestyClient()
  736. var result models.GetTrainJobResult
  737. retry := 0
  738. sendjob:
  739. res, err := client.R().
  740. SetAuthToken(TOKEN).
  741. SetResult(&result).
  742. Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID)
  743. if err != nil {
  744. return nil, fmt.Errorf("resty GetTrainJob: %v", err)
  745. }
  746. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  747. retry++
  748. _ = getToken()
  749. goto sendjob
  750. }
  751. if res.StatusCode() != http.StatusOK {
  752. var temp models.ErrorResult
  753. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  754. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  755. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  756. }
  757. log.Error("GetTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  758. return &result, fmt.Errorf("获取作业详情失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  759. }
  760. if !result.IsSuccess {
  761. log.Error("GetTrainJob(%s) failed", jobID)
  762. return &result, fmt.Errorf("获取作业详情失败")
  763. }
  764. return &result, nil
  765. }
  766. func GetTrainJobLog(jobID, versionID, baseLine, logFile, order string, lines int) (*models.GetTrainJobLogResult, error) {
  767. checkSetting()
  768. client := getRestyClient()
  769. var result models.GetTrainJobLogResult
  770. retry := 0
  771. sendjob:
  772. res, err := client.R().
  773. SetQueryParams(map[string]string{
  774. "base_line": baseLine,
  775. "lines": strconv.Itoa(lines),
  776. "log_file": logFile,
  777. "order": order,
  778. }).
  779. SetAuthToken(TOKEN).
  780. SetResult(&result).
  781. Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/aom-log")
  782. if err != nil {
  783. return nil, fmt.Errorf("resty GetTrainJobLog: %v", err)
  784. }
  785. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  786. retry++
  787. _ = getToken()
  788. goto sendjob
  789. }
  790. if res.StatusCode() != http.StatusOK {
  791. var temp models.ErrorResult
  792. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  793. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  794. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  795. }
  796. log.Error("GetTrainJobLog failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  797. return &result, fmt.Errorf("获取作业日志失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  798. }
  799. if !result.IsSuccess {
  800. log.Error("GetTrainJobLog(%s) failed", jobID)
  801. return &result, fmt.Errorf("获取作业日志失败:%s", result.ErrorMsg)
  802. }
  803. return &result, nil
  804. }
  805. func GetTrainJobLogFileNames(jobID, versionID string) (*models.GetTrainJobLogFileNamesResult, error) {
  806. checkSetting()
  807. client := getRestyClient()
  808. var result models.GetTrainJobLogFileNamesResult
  809. retry := 0
  810. sendjob:
  811. res, err := client.R().
  812. SetAuthToken(TOKEN).
  813. SetResult(&result).
  814. Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/log/file-names")
  815. if err != nil {
  816. return nil, fmt.Errorf("resty GetTrainJobLogFileNames: %v", err)
  817. }
  818. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  819. retry++
  820. _ = getToken()
  821. goto sendjob
  822. }
  823. if res.StatusCode() != http.StatusOK {
  824. var temp models.ErrorResult
  825. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  826. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  827. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  828. }
  829. log.Error("GetTrainJobLogFileNames failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  830. return &result, fmt.Errorf("GetTrainJobLogFileNames failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  831. }
  832. if !result.IsSuccess {
  833. log.Error("GetTrainJobLogFileNames(%s) failed", jobID)
  834. return &result, fmt.Errorf("获取作业日志文件失败:%s", result.ErrorMsg)
  835. }
  836. return &result, nil
  837. }
  838. func DelTrainJob(jobID string) (*models.TrainJobResult, error) {
  839. checkSetting()
  840. client := getRestyClient()
  841. var result models.TrainJobResult
  842. retry := 0
  843. sendjob:
  844. res, err := client.R().
  845. SetAuthToken(TOKEN).
  846. SetResult(&result).
  847. Delete(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID)
  848. if err != nil {
  849. return &result, fmt.Errorf("resty DelTrainJob: %v", err)
  850. }
  851. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  852. retry++
  853. _ = getToken()
  854. goto sendjob
  855. }
  856. if res.StatusCode() != http.StatusOK {
  857. var temp models.ErrorResult
  858. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  859. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  860. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  861. }
  862. log.Error("DelTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  863. return &result, fmt.Errorf("删除训练作业失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  864. }
  865. if !result.IsSuccess {
  866. log.Error("DelTrainJob(%s) failed", jobID)
  867. return &result, fmt.Errorf("删除训练作业失败:%s", result.ErrorMsg)
  868. }
  869. return &result, nil
  870. }
  871. func StopTrainJob(jobID, versionID string) (*models.TrainJobResult, error) {
  872. checkSetting()
  873. client := getRestyClient()
  874. var result models.TrainJobResult
  875. retry := 0
  876. sendjob:
  877. res, err := client.R().
  878. SetAuthToken(TOKEN).
  879. SetResult(&result).
  880. Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/stop")
  881. if err != nil {
  882. return &result, fmt.Errorf("resty StopTrainJob: %v", err)
  883. }
  884. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  885. retry++
  886. _ = getToken()
  887. goto sendjob
  888. }
  889. if res.StatusCode() != http.StatusOK {
  890. var temp models.ErrorResult
  891. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  892. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  893. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  894. }
  895. log.Error("StopTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  896. return &result, fmt.Errorf("停止训练作业失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  897. }
  898. if !result.IsSuccess {
  899. log.Error("StopTrainJob(%s) failed", jobID)
  900. return &result, fmt.Errorf("停止训练作业失败:%s", result.ErrorMsg)
  901. }
  902. return &result, nil
  903. }
  904. func DelTrainJobVersion(jobID string, versionID string) (*models.TrainJobResult, error) {
  905. checkSetting()
  906. client := getRestyClient()
  907. var result models.TrainJobResult
  908. retry := 0
  909. sendjob:
  910. res, err := client.R().
  911. SetAuthToken(TOKEN).
  912. SetResult(&result).
  913. Delete(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID)
  914. if err != nil {
  915. return &result, fmt.Errorf("resty DelTrainJobVersion: %v", err)
  916. }
  917. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  918. retry++
  919. _ = getToken()
  920. goto sendjob
  921. }
  922. if res.StatusCode() != http.StatusOK {
  923. var temp models.ErrorResult
  924. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  925. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  926. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  927. }
  928. log.Error("DelTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  929. return &result, fmt.Errorf("删除训练作业版本失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  930. }
  931. if !result.IsSuccess {
  932. log.Error("DelTrainJob(%s) failed", jobID)
  933. return &result, fmt.Errorf("删除训练作业版本失败:%s", result.ErrorMsg)
  934. }
  935. return &result, nil
  936. }
  937. func createInferenceJob(createJobParams models.CreateInferenceJobParams) (*models.CreateTrainJobResult, error) {
  938. checkSetting()
  939. client := getRestyClient()
  940. var result models.CreateTrainJobResult
  941. retry := 0
  942. sendjob:
  943. res, err := client.R().
  944. SetHeader("Content-Type", "application/json").
  945. SetAuthToken(TOKEN).
  946. SetBody(createJobParams).
  947. SetResult(&result).
  948. Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob)
  949. if err != nil {
  950. return nil, fmt.Errorf("resty create inference-job: %s", err)
  951. }
  952. req, _ := json.Marshal(createJobParams)
  953. log.Info("%s", req)
  954. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  955. retry++
  956. _ = getToken()
  957. goto sendjob
  958. }
  959. if res.StatusCode() != http.StatusOK {
  960. var temp models.ErrorResult
  961. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  962. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  963. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  964. }
  965. log.Error("createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  966. BootFileErrorMsg := "Invalid OBS path '" + createJobParams.InfConfig.BootFileUrl + "'."
  967. DataSetErrorMsg := "Invalid OBS path '" + createJobParams.InfConfig.DataUrl + "'."
  968. if temp.ErrorMsg == BootFileErrorMsg {
  969. log.Error("启动文件错误!createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  970. return &result, fmt.Errorf("启动文件错误!")
  971. }
  972. if temp.ErrorMsg == DataSetErrorMsg {
  973. log.Error("数据集错误!createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  974. return &result, fmt.Errorf("数据集错误!")
  975. }
  976. if res.StatusCode() == http.StatusBadGateway {
  977. return &result, fmt.Errorf(UnknownErrorPrefix+"createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  978. } else {
  979. return &result, fmt.Errorf("createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  980. }
  981. }
  982. if !result.IsSuccess {
  983. log.Error("createInferenceJob failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  984. return &result, fmt.Errorf("createInferenceJob failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  985. }
  986. return &result, nil
  987. }
  988. func createInferenceJobUserImage(createJobParams models.CreateInfUserImageParams) (*models.CreateTrainJobResult, error) {
  989. checkSetting()
  990. client := getRestyClient()
  991. var result models.CreateTrainJobResult
  992. retry := 0
  993. sendjob:
  994. res, err := client.R().
  995. SetHeader("Content-Type", "application/json").
  996. SetAuthToken(TOKEN).
  997. SetBody(createJobParams).
  998. SetResult(&result).
  999. Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob)
  1000. if err != nil {
  1001. return nil, fmt.Errorf("resty create train-job: %s", err)
  1002. }
  1003. req, _ := json.Marshal(createJobParams)
  1004. log.Info("%s", req)
  1005. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  1006. retry++
  1007. _ = getToken()
  1008. goto sendjob
  1009. }
  1010. if res.StatusCode() != http.StatusOK {
  1011. var temp models.ErrorResult
  1012. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  1013. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  1014. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  1015. }
  1016. log.Error("createInferenceJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  1017. bootFileErrorMsg := "Invalid OBS path '" + createJobParams.Config.BootFileUrl + "'."
  1018. dataSetErrorMsg := "Invalid OBS path '" + createJobParams.Config.DataUrl + "'."
  1019. if temp.ErrorMsg == bootFileErrorMsg {
  1020. log.Error("启动文件错误!createInferenceJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  1021. return &result, fmt.Errorf("启动文件错误!")
  1022. }
  1023. if temp.ErrorMsg == dataSetErrorMsg {
  1024. log.Error("数据集错误!createInferenceJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  1025. return &result, fmt.Errorf("数据集错误!")
  1026. }
  1027. if res.StatusCode() == http.StatusBadGateway {
  1028. return &result, fmt.Errorf(UnknownErrorPrefix+"createInferenceJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  1029. } else {
  1030. return &result, fmt.Errorf("createInferenceJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  1031. }
  1032. }
  1033. if !result.IsSuccess {
  1034. log.Error("createInferenceJobUserImage failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  1035. return &result, fmt.Errorf("createInferenceJobUserImage failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  1036. }
  1037. return &result, nil
  1038. }
  1039. func createNotebook2(createJobParams models.CreateNotebook2Params) (*models.CreateNotebookResult, error) {
  1040. checkSetting()
  1041. client := getRestyClient()
  1042. var result models.CreateNotebookResult
  1043. retry := 0
  1044. sendjob:
  1045. res, err := client.R().
  1046. SetHeader("Content-Type", "application/json").
  1047. SetAuthToken(TOKEN).
  1048. SetBody(createJobParams).
  1049. SetResult(&result).
  1050. Post(HOST + "/v1/" + setting.ProjectID + urlNotebook2)
  1051. if err != nil {
  1052. return nil, fmt.Errorf("resty create notebook2: %s", err)
  1053. }
  1054. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  1055. retry++
  1056. _ = getToken()
  1057. goto sendjob
  1058. }
  1059. var response models.NotebookResult
  1060. err = json.Unmarshal(res.Body(), &response)
  1061. if err != nil {
  1062. log.Error("json.Unmarshal failed: %s", err.Error())
  1063. return &result, fmt.Errorf("json.Unmarshal failed: %s", err.Error())
  1064. }
  1065. if res.StatusCode() == http.StatusBadGateway {
  1066. return &result, fmt.Errorf(UnknownErrorPrefix+"createNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  1067. }
  1068. if len(response.ErrorCode) != 0 {
  1069. log.Error("createNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  1070. if response.ErrorCode == errorCodeExceedLimit {
  1071. response.ErrorMsg = "所选规格使用数量已超过最大配额限制。"
  1072. }
  1073. if response.ErrorCode == modelartsIllegalToken && retry < 1 {
  1074. retry++
  1075. _ = getToken()
  1076. goto sendjob
  1077. }
  1078. return &result, fmt.Errorf("createNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  1079. }
  1080. return &result, nil
  1081. }
  1082. func GetTrainJobMetricStatistic(jobID, versionID, podName string) (*models.GetTrainJobMetricStatisticResult, error) {
  1083. checkSetting()
  1084. client := getRestyClient()
  1085. var result models.GetTrainJobMetricStatisticResult
  1086. retry := 0
  1087. sendjob:
  1088. res, err := client.R().
  1089. SetAuthToken(TOKEN).
  1090. SetResult(&result).
  1091. Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/pod/" + podName + "/metric-statistic?statistic_type=each")
  1092. if err != nil {
  1093. return nil, fmt.Errorf("resty GetTrainJobMetricStatistic: %v", err)
  1094. }
  1095. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  1096. retry++
  1097. _ = getToken()
  1098. goto sendjob
  1099. }
  1100. if res.StatusCode() != http.StatusOK {
  1101. var temp models.ErrorResult
  1102. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  1103. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  1104. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  1105. }
  1106. log.Error("GetTrainJobMetricStatistic failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  1107. return &result, fmt.Errorf("GetTrainJobMetricStatistic failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  1108. }
  1109. if !result.IsSuccess {
  1110. log.Error("GetTrainJobMetricStatistic(%s) failed", jobID)
  1111. return &result, fmt.Errorf("获取任务资源占用情况失败:%s", result.ErrorMsg)
  1112. }
  1113. return &result, nil
  1114. }
  1115. func GetTrainJobList(perPage, page int, sortBy, order, searchContent string) (*models.GetTrainJobListResult, error) {
  1116. checkSetting()
  1117. client := getRestyClient()
  1118. var result models.GetTrainJobListResult
  1119. retry := 0
  1120. sendjob:
  1121. res, err := client.R().
  1122. SetQueryParams(map[string]string{
  1123. "per_page": strconv.Itoa(perPage),
  1124. "page": strconv.Itoa(page),
  1125. "sortBy": sortBy,
  1126. "order": order,
  1127. "search_content": searchContent,
  1128. }).
  1129. SetAuthToken(TOKEN).
  1130. SetResult(&result).
  1131. Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob)
  1132. if err != nil {
  1133. return nil, fmt.Errorf("resty GetTrainJobList: %v", err)
  1134. }
  1135. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  1136. retry++
  1137. _ = getToken()
  1138. goto sendjob
  1139. }
  1140. if res.StatusCode() != http.StatusOK {
  1141. var temp models.ErrorResult
  1142. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  1143. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  1144. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  1145. }
  1146. log.Error("GetTrainJobList failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  1147. return &result, fmt.Errorf(temp.ErrorMsg)
  1148. }
  1149. if !result.IsSuccess {
  1150. log.Error("GetTrainJobList failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  1151. return &result, fmt.Errorf(result.ErrorMsg)
  1152. }
  1153. return &result, nil
  1154. }
  1155. func GetTrainJobVersionList(perPage, page int, jobID string) (*models.GetTrainJobVersionListResult, error) {
  1156. checkSetting()
  1157. client := getRestyClient()
  1158. var result models.GetTrainJobVersionListResult
  1159. retry := 0
  1160. sendjob:
  1161. res, err := client.R().
  1162. SetQueryParams(map[string]string{
  1163. "per_page": strconv.Itoa(perPage),
  1164. "page": strconv.Itoa(page),
  1165. }).
  1166. SetAuthToken(TOKEN).
  1167. SetResult(&result).
  1168. Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions")
  1169. if err != nil {
  1170. return nil, fmt.Errorf("resty GetTrainJobVersionList: %v", err)
  1171. }
  1172. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  1173. retry++
  1174. _ = getToken()
  1175. goto sendjob
  1176. }
  1177. if res.StatusCode() != http.StatusOK {
  1178. var temp models.ErrorResult
  1179. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  1180. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  1181. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  1182. }
  1183. log.Error("GetTrainJobVersionList failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  1184. return &result, fmt.Errorf(temp.ErrorMsg)
  1185. }
  1186. if !result.IsSuccess {
  1187. log.Error("GetTrainJobVersionList failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  1188. return &result, fmt.Errorf(result.ErrorMsg)
  1189. }
  1190. return &result, nil
  1191. }
  1192. func GetNotebookList(limit, offset int, sortBy, order, searchContent string) (*models.GetNotebookListResult, error) {
  1193. checkSetting()
  1194. client := getRestyClient()
  1195. var result models.GetNotebookListResult
  1196. retry := 0
  1197. sendjob:
  1198. res, err := client.R().
  1199. SetQueryParams(map[string]string{
  1200. "limit": strconv.Itoa(limit),
  1201. "offset": strconv.Itoa(offset),
  1202. "name": searchContent,
  1203. "sort_key": sortBy,
  1204. "sort_dir": order,
  1205. }).
  1206. SetAuthToken(TOKEN).
  1207. SetResult(&result).
  1208. Get(HOST + "/v1/" + setting.ProjectID + urlNotebook2)
  1209. if err != nil {
  1210. return nil, fmt.Errorf("resty GetNotebookList: %v", err)
  1211. }
  1212. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  1213. retry++
  1214. _ = getToken()
  1215. goto sendjob
  1216. }
  1217. if res.StatusCode() != http.StatusOK {
  1218. var temp models.ErrorResult
  1219. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  1220. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  1221. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  1222. }
  1223. log.Error("GetNotebookList failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  1224. return &result, fmt.Errorf(temp.ErrorMsg)
  1225. }
  1226. return &result, nil
  1227. }