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 28 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
4 years ago
4 years ago
4 years ago
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
4 years ago
4 years ago
4 years ago
4 years ago
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
3 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
3 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932
  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. )
  27. func getRestyClient() *resty.Client {
  28. if restyClient == nil {
  29. restyClient = resty.New()
  30. restyClient.SetTLSClientConfig(&tls.Config{InsecureSkipVerify: true})
  31. }
  32. return restyClient
  33. }
  34. func checkSetting() {
  35. if len(HOST) != 0 && len(TOKEN) != 0 && restyClient != nil {
  36. return
  37. }
  38. err := getToken()
  39. if err != nil {
  40. log.Error("getToken failed:%v", err)
  41. }
  42. }
  43. func getToken() error {
  44. HOST = setting.ModelArtsHost
  45. client := getRestyClient()
  46. params := models.GetTokenParams{
  47. Auth: models.Auth{
  48. Identity: models.Identity{
  49. Methods: []string{methodPassword},
  50. Password: models.Password{
  51. User: models.NotebookUser{
  52. Name: setting.ModelArtsUsername,
  53. Password: setting.ModelArtsPassword,
  54. Domain: models.Domain{
  55. Name: setting.ModelArtsDomain,
  56. },
  57. },
  58. },
  59. },
  60. Scope: models.Scope{
  61. Project: models.Project{
  62. Name: setting.ProjectName,
  63. },
  64. },
  65. },
  66. }
  67. res, err := client.R().
  68. SetHeader("Content-Type", "application/json").
  69. SetBody(params).
  70. Post(setting.IamHost + urlGetToken)
  71. if err != nil {
  72. return fmt.Errorf("resty getToken: %v", err)
  73. }
  74. if res.StatusCode() != http.StatusCreated {
  75. return fmt.Errorf("getToken failed:%s", res.String())
  76. }
  77. TOKEN = res.Header().Get("X-Subject-Token")
  78. return nil
  79. }
  80. func CreateJob(createJobParams models.CreateNotebookParams) (*models.CreateNotebookResult, error) {
  81. checkSetting()
  82. client := getRestyClient()
  83. var result models.CreateNotebookResult
  84. retry := 0
  85. sendjob:
  86. res, err := client.R().
  87. SetHeader("Content-Type", "application/json").
  88. SetAuthToken(TOKEN).
  89. SetBody(createJobParams).
  90. SetResult(&result).
  91. Post(HOST + "/v1/" + setting.ProjectID + urlNotebook)
  92. if err != nil {
  93. return nil, fmt.Errorf("resty create notebook: %s", err)
  94. }
  95. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  96. retry++
  97. _ = getToken()
  98. goto sendjob
  99. }
  100. var response models.NotebookResult
  101. err = json.Unmarshal(res.Body(), &response)
  102. if err != nil {
  103. log.Error("json.Unmarshal failed: %s", err.Error())
  104. return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
  105. }
  106. if len(response.ErrorCode) != 0 {
  107. log.Error("createNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  108. if response.ErrorCode == errorCodeExceedLimit {
  109. response.ErrorMsg = "所选规格使用数量已超过最大配额限制。"
  110. }
  111. return &result, fmt.Errorf("createNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  112. }
  113. return &result, nil
  114. }
  115. func GetJob(jobID string) (*models.GetNotebookResult, error) {
  116. checkSetting()
  117. client := getRestyClient()
  118. var result models.GetNotebookResult
  119. retry := 0
  120. sendjob:
  121. res, err := client.R().
  122. SetHeader("Content-Type", "application/json").
  123. SetAuthToken(TOKEN).
  124. SetResult(&result).
  125. Get(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID)
  126. if err != nil {
  127. return nil, fmt.Errorf("resty GetJob: %v", err)
  128. }
  129. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  130. retry++
  131. _ = getToken()
  132. goto sendjob
  133. }
  134. var response models.NotebookResult
  135. err = json.Unmarshal(res.Body(), &response)
  136. if err != nil {
  137. log.Error("json.Unmarshal failed: %s", err.Error())
  138. return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
  139. }
  140. if len(response.ErrorCode) != 0 {
  141. log.Error("GetJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  142. return &result, fmt.Errorf("GetJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  143. }
  144. return &result, nil
  145. }
  146. func ManageNotebook(jobID string, param models.NotebookAction) (*models.NotebookActionResult, error) {
  147. checkSetting()
  148. client := getRestyClient()
  149. var result models.NotebookActionResult
  150. retry := 0
  151. sendjob:
  152. res, err := client.R().
  153. SetHeader("Content-Type", "application/json").
  154. SetBody(param).
  155. SetAuthToken(TOKEN).
  156. SetResult(&result).
  157. Post(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID + "/action")
  158. if err != nil {
  159. return &result, fmt.Errorf("resty StopJob: %v", err)
  160. }
  161. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  162. retry++
  163. _ = getToken()
  164. goto sendjob
  165. }
  166. var response models.NotebookResult
  167. err = json.Unmarshal(res.Body(), &response)
  168. if err != nil {
  169. log.Error("json.Unmarshal failed: %s", err.Error())
  170. return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
  171. }
  172. if len(response.ErrorCode) != 0 {
  173. log.Error("ManageNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  174. return &result, fmt.Errorf("ManageNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  175. }
  176. return &result, nil
  177. }
  178. func DelNotebook(jobID string) (*models.NotebookDelResult, error) {
  179. checkSetting()
  180. client := getRestyClient()
  181. var result models.NotebookDelResult
  182. retry := 0
  183. sendjob:
  184. res, err := client.R().
  185. SetHeader("Content-Type", "application/json").
  186. SetAuthToken(TOKEN).
  187. SetResult(&result).
  188. Delete(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID)
  189. if err != nil {
  190. return &result, fmt.Errorf("resty DelJob: %v", err)
  191. }
  192. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  193. retry++
  194. _ = getToken()
  195. goto sendjob
  196. }
  197. var response models.NotebookResult
  198. err = json.Unmarshal(res.Body(), &response)
  199. if err != nil {
  200. log.Error("json.Unmarshal failed: %s", err.Error())
  201. return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
  202. }
  203. if len(response.ErrorCode) != 0 {
  204. log.Error("DelJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  205. return &result, fmt.Errorf("DelJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  206. }
  207. return &result, nil
  208. }
  209. func DelJob(jobID string) (*models.NotebookDelResult, error) {
  210. checkSetting()
  211. client := getRestyClient()
  212. var result models.NotebookDelResult
  213. retry := 0
  214. sendjob:
  215. res, err := client.R().
  216. SetHeader("Content-Type", "application/json").
  217. SetAuthToken(TOKEN).
  218. SetResult(&result).
  219. Delete(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID)
  220. if err != nil {
  221. return &result, fmt.Errorf("resty DelJob: %v", err)
  222. }
  223. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  224. retry++
  225. _ = getToken()
  226. goto sendjob
  227. }
  228. var response models.NotebookResult
  229. err = json.Unmarshal(res.Body(), &response)
  230. if err != nil {
  231. log.Error("json.Unmarshal failed: %s", err.Error())
  232. return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
  233. }
  234. if len(response.ErrorCode) != 0 {
  235. log.Error("DelJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  236. return &result, fmt.Errorf("DelJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  237. }
  238. return &result, nil
  239. }
  240. func GetJobToken(jobID string) (*models.NotebookGetJobTokenResult, error) {
  241. checkSetting()
  242. client := getRestyClient()
  243. var result models.NotebookGetJobTokenResult
  244. retry := 0
  245. sendjob:
  246. res, err := client.R().
  247. SetHeader("Content-Type", "application/json").
  248. SetAuthToken(TOKEN).
  249. SetResult(&result).
  250. Get(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID + "/token")
  251. if err != nil {
  252. return &result, fmt.Errorf("resty GetJobToken: %v", err)
  253. }
  254. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  255. retry++
  256. _ = getToken()
  257. goto sendjob
  258. }
  259. var response models.NotebookResult
  260. err = json.Unmarshal(res.Body(), &response)
  261. if err != nil {
  262. log.Error("json.Unmarshal failed: %s", err.Error())
  263. return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
  264. }
  265. if len(response.ErrorCode) != 0 {
  266. log.Error("GetJobToken failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  267. return &result, fmt.Errorf("GetJobToken failed(%s): %s", response.ErrorCode, response.ErrorMsg)
  268. }
  269. return &result, nil
  270. }
  271. func createTrainJob(createJobParams models.CreateTrainJobParams) (*models.CreateTrainJobResult, error) {
  272. checkSetting()
  273. client := getRestyClient()
  274. var result models.CreateTrainJobResult
  275. retry := 0
  276. sendjob:
  277. res, err := client.R().
  278. SetHeader("Content-Type", "application/json").
  279. SetAuthToken(TOKEN).
  280. SetBody(createJobParams).
  281. SetResult(&result).
  282. Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob)
  283. if err != nil {
  284. return nil, fmt.Errorf("resty create train-job: %s", err)
  285. }
  286. req, _ := json.Marshal(createJobParams)
  287. log.Info("%s", req)
  288. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  289. retry++
  290. _ = getToken()
  291. goto sendjob
  292. }
  293. if res.StatusCode() != http.StatusOK {
  294. var temp models.ErrorResult
  295. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  296. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  297. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  298. }
  299. log.Error("createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  300. BootFileErrorMsg := "Invalid OBS path '" + createJobParams.Config.BootFileUrl + "'."
  301. DataSetErrorMsg := "Invalid OBS path '" + createJobParams.Config.DataUrl + "'."
  302. if temp.ErrorMsg == BootFileErrorMsg {
  303. log.Error("启动文件错误!createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  304. return &result, fmt.Errorf("启动文件错误!")
  305. }
  306. if temp.ErrorMsg == DataSetErrorMsg {
  307. log.Error("数据集错误!createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  308. return &result, fmt.Errorf("数据集错误!")
  309. }
  310. return &result, fmt.Errorf("createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  311. }
  312. if !result.IsSuccess {
  313. log.Error("createTrainJob failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  314. return &result, fmt.Errorf("createTrainJob failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  315. }
  316. return &result, nil
  317. }
  318. func createTrainJobVersion(createJobVersionParams models.CreateTrainJobVersionParams, jobID string) (*models.CreateTrainJobResult, error) {
  319. checkSetting()
  320. client := getRestyClient()
  321. var result models.CreateTrainJobResult
  322. retry := 0
  323. sendjob:
  324. res, err := client.R().
  325. SetHeader("Content-Type", "application/json").
  326. SetAuthToken(TOKEN).
  327. SetBody(createJobVersionParams).
  328. SetResult(&result).
  329. Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions")
  330. if err != nil {
  331. return nil, fmt.Errorf("resty create train-job version: %s", err)
  332. }
  333. req, _ := json.Marshal(createJobVersionParams)
  334. log.Info("%s", req)
  335. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  336. retry++
  337. _ = getToken()
  338. goto sendjob
  339. }
  340. if res.StatusCode() != http.StatusOK {
  341. var temp models.ErrorResult
  342. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  343. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  344. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  345. }
  346. BootFileErrorMsg := "Invalid OBS path '" + createJobVersionParams.Config.BootFileUrl + "'."
  347. DataSetErrorMsg := "Invalid OBS path '" + createJobVersionParams.Config.DataUrl + "'."
  348. if temp.ErrorMsg == BootFileErrorMsg {
  349. log.Error("启动文件错误!createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  350. return &result, fmt.Errorf("启动文件错误!")
  351. }
  352. if temp.ErrorMsg == DataSetErrorMsg {
  353. log.Error("数据集错误!createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  354. return &result, fmt.Errorf("数据集错误!")
  355. }
  356. return &result, fmt.Errorf("createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  357. }
  358. if !result.IsSuccess {
  359. log.Error("createTrainJobVersion failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  360. return &result, fmt.Errorf("createTrainJobVersion failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  361. }
  362. return &result, nil
  363. }
  364. func GetResourceSpecs() (*models.GetResourceSpecsResult, error) {
  365. checkSetting()
  366. client := getRestyClient()
  367. var result models.GetResourceSpecsResult
  368. retry := 0
  369. sendjob:
  370. res, err := client.R().
  371. SetHeader("Content-Type", "application/json").
  372. SetAuthToken(TOKEN).
  373. SetResult(&result).
  374. Get(HOST + "/v1/" + setting.ProjectID + urlResourceSpecs)
  375. if err != nil {
  376. return nil, fmt.Errorf("resty GetResourceSpecs: %v", err)
  377. }
  378. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  379. retry++
  380. _ = getToken()
  381. goto sendjob
  382. }
  383. if res.StatusCode() != http.StatusOK {
  384. var temp models.ErrorResult
  385. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  386. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  387. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  388. }
  389. log.Error("GetResourceSpecs failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  390. return &result, fmt.Errorf("GetResourceSpecs failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  391. }
  392. if !result.IsSuccess {
  393. log.Error("GetResourceSpecs failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  394. return &result, fmt.Errorf("GetResourceSpecs failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  395. }
  396. return &result, nil
  397. }
  398. func CreateTrainJobConfig(req models.CreateConfigParams) (*models.CreateTrainJobConfigResult, error) {
  399. checkSetting()
  400. client := getRestyClient()
  401. var result models.CreateTrainJobConfigResult
  402. retry := 0
  403. sendjob:
  404. res, err := client.R().
  405. SetHeader("Content-Type", "application/json").
  406. SetAuthToken(TOKEN).
  407. SetBody(req).
  408. SetResult(&result).
  409. Post(HOST + "/v1/" + setting.ProjectID + urlTrainJobConfig)
  410. if err != nil {
  411. return nil, fmt.Errorf("resty CreateTrainJobConfig: %s", err)
  412. }
  413. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  414. retry++
  415. _ = getToken()
  416. goto sendjob
  417. }
  418. //temp, _ := json.Marshal(req)
  419. //log.Info("%s", temp)
  420. if res.StatusCode() != http.StatusOK {
  421. var temp models.ErrorResult
  422. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  423. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  424. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  425. }
  426. log.Error("CreateTrainJobConfig failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  427. return &result, fmt.Errorf("CreateTrainJobConfig failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  428. }
  429. if !result.IsSuccess {
  430. log.Error("CreateTrainJobConfig failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  431. return &result, fmt.Errorf("CreateTrainJobConfig failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  432. }
  433. return &result, nil
  434. }
  435. func GetConfigList(perPage, page int, sortBy, order, searchContent, configType string) (*models.GetConfigListResult, error) {
  436. checkSetting()
  437. client := getRestyClient()
  438. var result models.GetConfigListResult
  439. retry := 0
  440. sendjob:
  441. res, err := client.R().
  442. SetQueryParams(map[string]string{
  443. "per_page": strconv.Itoa(perPage),
  444. "page": strconv.Itoa(page),
  445. "sortBy": sortBy,
  446. "order": order,
  447. "search_content": searchContent,
  448. "config_type": configType,
  449. }).
  450. SetAuthToken(TOKEN).
  451. SetResult(&result).
  452. Get(HOST + "/v1/" + setting.ProjectID + urlTrainJobConfig)
  453. if err != nil {
  454. return nil, fmt.Errorf("resty GetConfigList: %v", err)
  455. }
  456. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  457. retry++
  458. _ = getToken()
  459. goto sendjob
  460. }
  461. if res.StatusCode() != http.StatusOK {
  462. var temp models.ErrorResult
  463. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  464. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  465. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  466. }
  467. log.Error("GetConfigList failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  468. return &result, fmt.Errorf("获取参数配置列表失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  469. }
  470. if !result.IsSuccess {
  471. log.Error("GetConfigList failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  472. return &result, fmt.Errorf("获取参数配置列表失败(%s): %s", result.ErrorCode, result.ErrorMsg)
  473. }
  474. return &result, nil
  475. }
  476. func GetParaConfig(configName, configType string) (models.GetConfigResult, error) {
  477. checkSetting()
  478. client := getRestyClient()
  479. var result models.GetConfigResult
  480. retry := 0
  481. sendjob:
  482. res, err := client.R().
  483. SetQueryParams(map[string]string{
  484. "config_type": configType,
  485. }).
  486. SetAuthToken(TOKEN).
  487. SetResult(&result).
  488. Get(HOST + "/v1/" + setting.ProjectID + urlTrainJobConfig + "/" + configName)
  489. if err != nil {
  490. return result, fmt.Errorf("resty GetParaConfig: %v", err)
  491. }
  492. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  493. retry++
  494. _ = getToken()
  495. goto sendjob
  496. }
  497. if res.StatusCode() != http.StatusOK {
  498. var temp models.ErrorResult
  499. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  500. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  501. return result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  502. }
  503. log.Error("GetParaConfig failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  504. return result, fmt.Errorf("获取参数配置详情失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  505. }
  506. if !result.IsSuccess {
  507. log.Error("GetParaConfig failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  508. return result, fmt.Errorf("获取参数配置详情失败(%s): %s", result.ErrorCode, result.ErrorMsg)
  509. }
  510. return result, nil
  511. }
  512. func GetTrainJob(jobID, versionID string) (*models.GetTrainJobResult, error) {
  513. checkSetting()
  514. client := getRestyClient()
  515. var result models.GetTrainJobResult
  516. retry := 0
  517. sendjob:
  518. res, err := client.R().
  519. SetAuthToken(TOKEN).
  520. SetResult(&result).
  521. Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID)
  522. if err != nil {
  523. return nil, fmt.Errorf("resty GetTrainJob: %v", err)
  524. }
  525. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  526. retry++
  527. _ = getToken()
  528. goto sendjob
  529. }
  530. if res.StatusCode() != http.StatusOK {
  531. var temp models.ErrorResult
  532. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  533. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  534. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  535. }
  536. log.Error("GetTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  537. return &result, fmt.Errorf("获取作业详情失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  538. }
  539. if !result.IsSuccess {
  540. log.Error("GetTrainJob(%s) failed", jobID)
  541. return &result, fmt.Errorf("获取作业详情失败")
  542. }
  543. return &result, nil
  544. }
  545. func GetTrainJobLog(jobID, versionID, baseLine, logFile, order string, lines int) (*models.GetTrainJobLogResult, error) {
  546. checkSetting()
  547. client := getRestyClient()
  548. var result models.GetTrainJobLogResult
  549. retry := 0
  550. sendjob:
  551. res, err := client.R().
  552. SetQueryParams(map[string]string{
  553. "base_line": baseLine,
  554. "lines": strconv.Itoa(lines),
  555. "log_file": logFile,
  556. "order": order,
  557. }).
  558. SetAuthToken(TOKEN).
  559. SetResult(&result).
  560. Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/aom-log")
  561. if err != nil {
  562. return nil, fmt.Errorf("resty GetTrainJobLog: %v", err)
  563. }
  564. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  565. retry++
  566. _ = getToken()
  567. goto sendjob
  568. }
  569. if res.StatusCode() != http.StatusOK {
  570. var temp models.ErrorResult
  571. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  572. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  573. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  574. }
  575. log.Error("GetTrainJobLog failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  576. return &result, fmt.Errorf("获取作业日志失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  577. }
  578. if !result.IsSuccess {
  579. log.Error("GetTrainJobLog(%s) failed", jobID)
  580. return &result, fmt.Errorf("获取作业日志失败:%s", result.ErrorMsg)
  581. }
  582. return &result, nil
  583. }
  584. func GetTrainJobLogFileNames(jobID, versionID string) (*models.GetTrainJobLogFileNamesResult, error) {
  585. checkSetting()
  586. client := getRestyClient()
  587. var result models.GetTrainJobLogFileNamesResult
  588. retry := 0
  589. sendjob:
  590. res, err := client.R().
  591. SetAuthToken(TOKEN).
  592. SetResult(&result).
  593. Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/log/file-names")
  594. if err != nil {
  595. return nil, fmt.Errorf("resty GetTrainJobLogFileNames: %v", err)
  596. }
  597. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  598. retry++
  599. _ = getToken()
  600. goto sendjob
  601. }
  602. if res.StatusCode() != http.StatusOK {
  603. var temp models.ErrorResult
  604. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  605. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  606. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  607. }
  608. log.Error("GetTrainJobLogFileNames failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  609. return &result, fmt.Errorf("GetTrainJobLogFileNames failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  610. }
  611. if !result.IsSuccess {
  612. log.Error("GetTrainJobLogFileNames(%s) failed", jobID)
  613. return &result, fmt.Errorf("获取作业日志文件失败:%s", result.ErrorMsg)
  614. }
  615. return &result, nil
  616. }
  617. func DelTrainJob(jobID string) (*models.TrainJobResult, error) {
  618. checkSetting()
  619. client := getRestyClient()
  620. var result models.TrainJobResult
  621. retry := 0
  622. sendjob:
  623. res, err := client.R().
  624. SetAuthToken(TOKEN).
  625. SetResult(&result).
  626. Delete(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID)
  627. if err != nil {
  628. return &result, fmt.Errorf("resty DelTrainJob: %v", err)
  629. }
  630. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  631. retry++
  632. _ = getToken()
  633. goto sendjob
  634. }
  635. if res.StatusCode() != http.StatusOK {
  636. var temp models.ErrorResult
  637. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  638. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  639. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  640. }
  641. log.Error("DelTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  642. return &result, fmt.Errorf("删除训练作业失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  643. }
  644. if !result.IsSuccess {
  645. log.Error("DelTrainJob(%s) failed", jobID)
  646. return &result, fmt.Errorf("删除训练作业失败:%s", result.ErrorMsg)
  647. }
  648. return &result, nil
  649. }
  650. func StopTrainJob(jobID, versionID string) (*models.TrainJobResult, error) {
  651. checkSetting()
  652. client := getRestyClient()
  653. var result models.TrainJobResult
  654. retry := 0
  655. sendjob:
  656. res, err := client.R().
  657. SetAuthToken(TOKEN).
  658. SetResult(&result).
  659. Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/stop")
  660. if err != nil {
  661. return &result, fmt.Errorf("resty StopTrainJob: %v", err)
  662. }
  663. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  664. retry++
  665. _ = getToken()
  666. goto sendjob
  667. }
  668. if res.StatusCode() != http.StatusOK {
  669. var temp models.ErrorResult
  670. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  671. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  672. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  673. }
  674. log.Error("StopTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  675. return &result, fmt.Errorf("停止训练作业失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  676. }
  677. if !result.IsSuccess {
  678. log.Error("StopTrainJob(%s) failed", jobID)
  679. return &result, fmt.Errorf("停止训练作业失败:%s", result.ErrorMsg)
  680. }
  681. return &result, nil
  682. }
  683. func DelTrainJobVersion(jobID string, versionID string) (*models.TrainJobResult, error) {
  684. checkSetting()
  685. client := getRestyClient()
  686. var result models.TrainJobResult
  687. retry := 0
  688. sendjob:
  689. res, err := client.R().
  690. SetAuthToken(TOKEN).
  691. SetResult(&result).
  692. Delete(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID)
  693. if err != nil {
  694. return &result, fmt.Errorf("resty DelTrainJobVersion: %v", err)
  695. }
  696. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  697. retry++
  698. _ = getToken()
  699. goto sendjob
  700. }
  701. if res.StatusCode() != http.StatusOK {
  702. var temp models.ErrorResult
  703. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  704. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  705. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  706. }
  707. log.Error("DelTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  708. return &result, fmt.Errorf("删除训练作业版本失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  709. }
  710. if !result.IsSuccess {
  711. log.Error("DelTrainJob(%s) failed", jobID)
  712. return &result, fmt.Errorf("删除训练作业版本失败:%s", result.ErrorMsg)
  713. }
  714. return &result, nil
  715. }
  716. func createInferenceJob(createJobParams models.CreateInferenceJobParams) (*models.CreateTrainJobResult, error) {
  717. checkSetting()
  718. client := getRestyClient()
  719. var result models.CreateTrainJobResult
  720. retry := 0
  721. sendjob:
  722. res, err := client.R().
  723. SetHeader("Content-Type", "application/json").
  724. SetAuthToken(TOKEN).
  725. SetBody(createJobParams).
  726. SetResult(&result).
  727. Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob)
  728. if err != nil {
  729. return nil, fmt.Errorf("resty create inference-job: %s", err)
  730. }
  731. req, _ := json.Marshal(createJobParams)
  732. log.Info("%s", req)
  733. if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
  734. retry++
  735. _ = getToken()
  736. goto sendjob
  737. }
  738. if res.StatusCode() != http.StatusOK {
  739. var temp models.ErrorResult
  740. if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
  741. log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  742. return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
  743. }
  744. log.Error("createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  745. BootFileErrorMsg := "Invalid OBS path '" + createJobParams.InfConfig.BootFileUrl + "'."
  746. DataSetErrorMsg := "Invalid OBS path '" + createJobParams.InfConfig.DataUrl + "'."
  747. if temp.ErrorMsg == BootFileErrorMsg {
  748. log.Error("启动文件错误!createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  749. return &result, fmt.Errorf("启动文件错误!")
  750. }
  751. if temp.ErrorMsg == DataSetErrorMsg {
  752. log.Error("数据集错误!createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  753. return &result, fmt.Errorf("数据集错误!")
  754. }
  755. return &result, fmt.Errorf("createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
  756. }
  757. if !result.IsSuccess {
  758. log.Error("createInferenceJob failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  759. return &result, fmt.Errorf("createInferenceJob failed(%s): %s", result.ErrorCode, result.ErrorMsg)
  760. }
  761. return &result, nil
  762. }