|
1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489 |
- package modelarts
-
- import (
- "crypto/tls"
- "encoding/json"
- "fmt"
- "net/http"
- "strconv"
-
- "code.gitea.io/gitea/models"
- "code.gitea.io/gitea/modules/log"
- "code.gitea.io/gitea/modules/setting"
- "github.com/go-resty/resty/v2"
- )
-
- var (
- restyClient *resty.Client
- HOST string
- TOKEN string
- )
-
- const (
- methodPassword = "password"
-
- urlGetToken = "/v3/auth/tokens"
- urlNotebook = "/demanager/instances"
- urlTrainJob = "/training-jobs"
- urlResourceSpecs = "/job/resource-specs"
- urlTrainJobConfig = "/training-job-configs"
- errorCodeExceedLimit = "ModelArts.0118"
-
- //notebook 2.0
- urlNotebook2 = "/notebooks"
-
- //error code
- modelartsIllegalToken = "ModelArts.6401"
- NotebookNotFound = "ModelArts.6404"
- NotebookNoPermission = "ModelArts.6407"
- NotebookInvalid = "ModelArts.6400"
- UnknownErrorPrefix = "UNKNOWN:"
- )
-
- func getRestyClient() *resty.Client {
- if restyClient == nil {
- restyClient = resty.New()
- restyClient.SetTLSClientConfig(&tls.Config{InsecureSkipVerify: true})
- }
- return restyClient
- }
-
- func checkSetting() {
- if len(HOST) != 0 && len(TOKEN) != 0 && restyClient != nil {
- return
- }
-
- err := getToken()
- if err != nil {
- log.Error("getToken failed:%v", err)
- }
- }
-
- func getToken() error {
- HOST = setting.ModelArtsHost
-
- client := getRestyClient()
- params := models.GetTokenParams{
- Auth: models.Auth{
- Identity: models.Identity{
- Methods: []string{methodPassword},
- Password: models.Password{
- User: models.NotebookUser{
- Name: setting.ModelArtsUsername,
- Password: setting.ModelArtsPassword,
- Domain: models.Domain{
- Name: setting.ModelArtsDomain,
- },
- },
- },
- },
- Scope: models.Scope{
- Project: models.Project{
- Name: setting.ProjectName,
- },
- },
- },
- }
-
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetBody(params).
- Post(setting.IamHost + urlGetToken)
- if err != nil {
- return fmt.Errorf("resty getToken: %v", err)
- }
-
- if res.StatusCode() != http.StatusCreated {
- return fmt.Errorf("getToken failed:%s", res.String())
- }
-
- TOKEN = res.Header().Get("X-Subject-Token")
-
- return nil
- }
-
- func CreateJob(createJobParams models.CreateNotebookParams) (*models.CreateNotebookResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.CreateNotebookResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetBody(createJobParams).
- SetResult(&result).
- Post(HOST + "/v1/" + setting.ProjectID + urlNotebook)
-
- if err != nil {
- return nil, fmt.Errorf("resty create notebook: %s", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- var response models.NotebookResult
- err = json.Unmarshal(res.Body(), &response)
- if err != nil {
- log.Error("json.Unmarshal failed: %s", err.Error())
- return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
- }
-
- if len(response.ErrorCode) != 0 {
- log.Error("createNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- if response.ErrorCode == errorCodeExceedLimit {
- response.ErrorMsg = "所选规格使用数量已超过最大配额限制。"
- }
- return &result, fmt.Errorf("createNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func GetJob(jobID string) (*models.GetNotebookResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.GetNotebookResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetResult(&result).
- Get(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID)
-
- if err != nil {
- return nil, fmt.Errorf("resty GetJob: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- var response models.NotebookResult
- err = json.Unmarshal(res.Body(), &response)
- if err != nil {
- log.Error("json.Unmarshal failed: %s", err.Error())
- return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
- }
-
- if len(response.ErrorCode) != 0 {
- log.Error("GetJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- return &result, fmt.Errorf("GetJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func GetNotebook2(jobID string) (*models.GetNotebook2Result, error) {
- checkSetting()
- client := getRestyClient()
- var result models.GetNotebook2Result
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetResult(&result).
- Get(HOST + "/v1/" + setting.ProjectID + urlNotebook2 + "/" + jobID)
-
- if err != nil {
- return nil, fmt.Errorf("resty GetJob: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- var response models.NotebookResult
- err = json.Unmarshal(res.Body(), &response)
- if err != nil {
- log.Error("json.Unmarshal failed: %s", err.Error())
- return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
- }
-
- if len(response.ErrorCode) != 0 {
- log.Error("GetJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- if response.ErrorCode == modelartsIllegalToken && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
- return &result, fmt.Errorf("GetJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func ManageNotebook(jobID string, param models.NotebookAction) (*models.NotebookActionResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.NotebookActionResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetBody(param).
- SetAuthToken(TOKEN).
- SetResult(&result).
- Post(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID + "/action")
-
- if err != nil {
- return &result, fmt.Errorf("resty StopJob: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- var response models.NotebookResult
- err = json.Unmarshal(res.Body(), &response)
- if err != nil {
- log.Error("json.Unmarshal failed: %s", err.Error())
- return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
- }
-
- if len(response.ErrorCode) != 0 {
- log.Error("ManageNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- return &result, fmt.Errorf("ManageNotebook failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func ManageNotebook2(jobID string, param models.NotebookAction) (*models.NotebookActionResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.NotebookActionResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetResult(&result).
- Post(HOST + "/v1/" + setting.ProjectID + urlNotebook2 + "/" + jobID + "/" + param.Action + "?duration=" + strconv.Itoa(AutoStopDurationMs))
-
- if err != nil {
- return &result, fmt.Errorf("resty ManageNotebook2: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- var response models.NotebookResult
- err = json.Unmarshal(res.Body(), &response)
- if err != nil {
- log.Error("json.Unmarshal failed: %s", err.Error())
- return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
- }
-
- if res.StatusCode() == http.StatusBadGateway {
- return &result, fmt.Errorf(UnknownErrorPrefix+"createNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- }
-
- if len(response.ErrorCode) != 0 {
- log.Error("ManageNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- if response.ErrorCode == modelartsIllegalToken && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
- return &result, fmt.Errorf("ManageNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func DelNotebook(jobID string) (*models.NotebookDelResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.NotebookDelResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetResult(&result).
- Delete(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID)
-
- if err != nil {
- return &result, fmt.Errorf("resty DelJob: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- var response models.NotebookResult
- err = json.Unmarshal(res.Body(), &response)
- if err != nil {
- log.Error("json.Unmarshal failed: %s", err.Error())
- return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
- }
-
- if len(response.ErrorCode) != 0 {
- log.Error("DelJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- return &result, fmt.Errorf("DelJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func DelNotebook2(jobID string) (*models.NotebookDelResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.NotebookDelResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetResult(&result).
- Delete(HOST + "/v1/" + setting.ProjectID + urlNotebook2 + "/" + jobID)
-
- if err != nil {
- return &result, fmt.Errorf("resty DelJob: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- var response models.NotebookResult
- err = json.Unmarshal(res.Body(), &response)
- if err != nil {
- log.Error("json.Unmarshal failed: %s", err.Error())
- return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
- }
-
- if len(response.ErrorCode) != 0 {
- log.Error("DelNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- if response.ErrorCode == modelartsIllegalToken && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
- return &result, fmt.Errorf("DelNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func DelJob(jobID string) (*models.NotebookDelResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.NotebookDelResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetResult(&result).
- Delete(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID)
-
- if err != nil {
- return &result, fmt.Errorf("resty DelJob: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- var response models.NotebookResult
- err = json.Unmarshal(res.Body(), &response)
- if err != nil {
- log.Error("json.Unmarshal failed: %s", err.Error())
- return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
- }
-
- if len(response.ErrorCode) != 0 {
- log.Error("DelJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- return &result, fmt.Errorf("DelJob failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func GetJobToken(jobID string) (*models.NotebookGetJobTokenResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.NotebookGetJobTokenResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetResult(&result).
- Get(HOST + "/v1/" + setting.ProjectID + urlNotebook + "/" + jobID + "/token")
-
- if err != nil {
- return &result, fmt.Errorf("resty GetJobToken: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- var response models.NotebookResult
- err = json.Unmarshal(res.Body(), &response)
- if err != nil {
- log.Error("json.Unmarshal failed: %s", err.Error())
- return &result, fmt.Errorf("son.Unmarshal failed: %s", err.Error())
- }
-
- if len(response.ErrorCode) != 0 {
- log.Error("GetJobToken failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- return &result, fmt.Errorf("GetJobToken failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func createTrainJobUserImage(createJobParams models.CreateUserImageTrainJobParams) (*models.CreateTrainJobResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.CreateTrainJobResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetBody(createJobParams).
- SetResult(&result).
- Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob)
-
- if err != nil {
- return nil, fmt.Errorf("resty create train-job: %s", err)
- }
-
- req, _ := json.Marshal(createJobParams)
- log.Info("postapi json: %s", req)
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("createTrainJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- bootFileErrorMsg := "Invalid OBS path '" + createJobParams.Config.BootFileUrl + "'."
- dataSetErrorMsg := "Invalid OBS path '" + createJobParams.Config.DataUrl + "'."
- if temp.ErrorMsg == bootFileErrorMsg {
- log.Error("启动文件错误!createTrainJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("启动文件错误!")
- }
- if temp.ErrorMsg == dataSetErrorMsg {
- log.Error("数据集错误!createTrainJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("数据集错误!")
- }
- if res.StatusCode() == http.StatusBadGateway {
- return &result, fmt.Errorf(UnknownErrorPrefix+"createTrainJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- } else {
- return &result, fmt.Errorf("createTrainJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
- }
-
- if !result.IsSuccess {
- log.Error("createTrainJobUserImage failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- return &result, fmt.Errorf("createTrainJobUserImage failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func createTrainJob(createJobParams models.CreateTrainJobParams) (*models.CreateTrainJobResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.CreateTrainJobResult
-
- retry := 0
- req, _ := json.Marshal(createJobParams)
- log.Info("postapi json: %s", req)
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetBody(createJobParams).
- SetResult(&result).
- Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob)
-
- if err != nil {
- return nil, fmt.Errorf("resty create train-job: %s", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- bootFileErrorMsg := "Invalid OBS path '" + createJobParams.Config.BootFileUrl + "'."
- dataSetErrorMsg := "Invalid OBS path '" + createJobParams.Config.DataUrl + "'."
- if temp.ErrorMsg == bootFileErrorMsg {
- log.Error("启动文件错误!createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("启动文件错误!")
- }
- if temp.ErrorMsg == dataSetErrorMsg {
- log.Error("数据集错误!createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("数据集错误!")
- }
- if res.StatusCode() == http.StatusBadGateway {
- return &result, fmt.Errorf(UnknownErrorPrefix+"createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- } else {
- return &result, fmt.Errorf("createTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
- }
-
- if !result.IsSuccess {
- log.Error("createTrainJob failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- return &result, fmt.Errorf("createTrainJob failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func createTrainJobVersion(createJobVersionParams models.CreateTrainJobVersionParams, jobID string) (*models.CreateTrainJobResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.CreateTrainJobResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetBody(createJobVersionParams).
- SetResult(&result).
- Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions")
-
- if err != nil {
- return nil, fmt.Errorf("resty create train-job version: %s", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
-
- log.Error("createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- bootFileErrorMsg := "Invalid OBS path '" + createJobVersionParams.Config.BootFileUrl + "'."
- dataSetErrorMsg := "Invalid OBS path '" + createJobVersionParams.Config.DataUrl + "'."
- if temp.ErrorMsg == bootFileErrorMsg {
- log.Error("启动文件错误!createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("启动文件错误!")
- }
- if temp.ErrorMsg == dataSetErrorMsg {
- log.Error("数据集错误!createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("数据集错误!")
- }
- if res.StatusCode() == http.StatusBadGateway {
- return &result, fmt.Errorf(UnknownErrorPrefix+"createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- } else {
- return &result, fmt.Errorf("createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
- }
-
- if !result.IsSuccess {
- log.Error("createTrainJobVersion failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- return &result, fmt.Errorf("createTrainJobVersion failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func createTrainJobVersionUserImage(createJobVersionParams models.CreateTrainJobVersionUserImageParams, jobID string) (*models.CreateTrainJobResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.CreateTrainJobResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetBody(createJobVersionParams).
- SetResult(&result).
- Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions")
-
- if err != nil {
- return nil, fmt.Errorf("resty create train-job version: %s", err)
- }
-
- req, _ := json.Marshal(createJobVersionParams)
- log.Info("%s", req)
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- BootFileErrorMsg := "Invalid OBS path '" + createJobVersionParams.Config.BootFileUrl + "'."
- DataSetErrorMsg := "Invalid OBS path '" + createJobVersionParams.Config.DataUrl + "'."
- if temp.ErrorMsg == BootFileErrorMsg {
- log.Error("启动文件错误!createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("启动文件错误!")
- }
- if temp.ErrorMsg == DataSetErrorMsg {
- log.Error("数据集错误!createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("数据集错误!")
- }
- return &result, fmt.Errorf("createTrainJobVersion failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
-
- if !result.IsSuccess {
- log.Error("createTrainJobVersion failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- return &result, fmt.Errorf("createTrainJobVersion failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func GetResourceSpecs() (*models.GetResourceSpecsResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.GetResourceSpecsResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetResult(&result).
- Get(HOST + "/v1/" + setting.ProjectID + urlResourceSpecs)
-
- if err != nil {
- return nil, fmt.Errorf("resty GetResourceSpecs: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("GetResourceSpecs failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("GetResourceSpecs failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
-
- if !result.IsSuccess {
- log.Error("GetResourceSpecs failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- return &result, fmt.Errorf("GetResourceSpecs failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func CreateTrainJobConfig(req models.CreateConfigParams) (*models.CreateTrainJobConfigResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.CreateTrainJobConfigResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetBody(req).
- SetResult(&result).
- Post(HOST + "/v1/" + setting.ProjectID + urlTrainJobConfig)
-
- if err != nil {
- return nil, fmt.Errorf("resty CreateTrainJobConfig: %s", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("CreateTrainJobConfig failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("CreateTrainJobConfig failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
-
- if !result.IsSuccess {
- log.Error("CreateTrainJobConfig failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- return &result, fmt.Errorf("CreateTrainJobConfig failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func GetConfigList(perPage, page int, sortBy, order, searchContent, configType string) (*models.GetConfigListResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.GetConfigListResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetQueryParams(map[string]string{
- "per_page": strconv.Itoa(perPage),
- "page": strconv.Itoa(page),
- "sortBy": sortBy,
- "order": order,
- "search_content": searchContent,
- "config_type": configType,
- }).
- SetAuthToken(TOKEN).
- SetResult(&result).
- Get(HOST + "/v1/" + setting.ProjectID + urlTrainJobConfig)
-
- if err != nil {
- return nil, fmt.Errorf("resty GetConfigList: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("GetConfigList failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("获取参数配置列表失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
-
- if !result.IsSuccess {
- log.Error("GetConfigList failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- return &result, fmt.Errorf("获取参数配置列表失败(%s): %s", result.ErrorCode, result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func GetParaConfig(configName, configType string) (models.GetConfigResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.GetConfigResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetQueryParams(map[string]string{
- "config_type": configType,
- }).
- SetAuthToken(TOKEN).
- SetResult(&result).
- Get(HOST + "/v1/" + setting.ProjectID + urlTrainJobConfig + "/" + configName)
-
- if err != nil {
- return result, fmt.Errorf("resty GetParaConfig: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("GetParaConfig failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return result, fmt.Errorf("获取参数配置详情失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
-
- if !result.IsSuccess {
- log.Error("GetParaConfig failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- return result, fmt.Errorf("获取参数配置详情失败(%s): %s", result.ErrorCode, result.ErrorMsg)
- }
-
- return result, nil
- }
-
- func GetTrainJob(jobID, versionID string) (*models.GetTrainJobResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.GetTrainJobResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetAuthToken(TOKEN).
- SetResult(&result).
- Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID)
-
- if err != nil {
- return nil, fmt.Errorf("resty GetTrainJob: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("GetTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("获取作业详情失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
-
- if !result.IsSuccess {
- log.Error("GetTrainJob(%s) failed", jobID)
- return &result, fmt.Errorf("获取作业详情失败")
- }
-
- return &result, nil
- }
-
- func GetTrainJobLog(jobID, versionID, baseLine, logFile, order string, lines int) (*models.GetTrainJobLogResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.GetTrainJobLogResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetQueryParams(map[string]string{
- "base_line": baseLine,
- "lines": strconv.Itoa(lines),
- "log_file": logFile,
- "order": order,
- }).
- SetAuthToken(TOKEN).
- SetResult(&result).
- Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/aom-log")
-
- if err != nil {
- return nil, fmt.Errorf("resty GetTrainJobLog: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("GetTrainJobLog failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("获取作业日志失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
-
- if !result.IsSuccess {
- log.Error("GetTrainJobLog(%s) failed", jobID)
- return &result, fmt.Errorf("获取作业日志失败:%s", result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func GetTrainJobLogFileNames(jobID, versionID string) (*models.GetTrainJobLogFileNamesResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.GetTrainJobLogFileNamesResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetAuthToken(TOKEN).
- SetResult(&result).
- Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/log/file-names")
-
- if err != nil {
- return nil, fmt.Errorf("resty GetTrainJobLogFileNames: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("GetTrainJobLogFileNames failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("GetTrainJobLogFileNames failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
-
- if !result.IsSuccess {
- log.Error("GetTrainJobLogFileNames(%s) failed", jobID)
- return &result, fmt.Errorf("获取作业日志文件失败:%s", result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func DelTrainJob(jobID string) (*models.TrainJobResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.TrainJobResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetAuthToken(TOKEN).
- SetResult(&result).
- Delete(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID)
-
- if err != nil {
- return &result, fmt.Errorf("resty DelTrainJob: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("DelTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("删除训练作业失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
-
- if !result.IsSuccess {
- log.Error("DelTrainJob(%s) failed", jobID)
- return &result, fmt.Errorf("删除训练作业失败:%s", result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func StopTrainJob(jobID, versionID string) (*models.TrainJobResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.TrainJobResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetAuthToken(TOKEN).
- SetResult(&result).
- Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/stop")
-
- if err != nil {
- return &result, fmt.Errorf("resty StopTrainJob: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("StopTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("停止训练作业失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
-
- if !result.IsSuccess {
- log.Error("StopTrainJob(%s) failed", jobID)
- return &result, fmt.Errorf("停止训练作业失败:%s", result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func DelTrainJobVersion(jobID string, versionID string) (*models.TrainJobResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.TrainJobResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetAuthToken(TOKEN).
- SetResult(&result).
- Delete(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID)
-
- if err != nil {
- return &result, fmt.Errorf("resty DelTrainJobVersion: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("DelTrainJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("删除训练作业版本失败(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
-
- if !result.IsSuccess {
- log.Error("DelTrainJob(%s) failed", jobID)
- return &result, fmt.Errorf("删除训练作业版本失败:%s", result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func createInferenceJob(createJobParams models.CreateInferenceJobParams) (*models.CreateTrainJobResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.CreateTrainJobResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetBody(createJobParams).
- SetResult(&result).
- Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob)
-
- if err != nil {
- return nil, fmt.Errorf("resty create inference-job: %s", err)
- }
-
- req, _ := json.Marshal(createJobParams)
- log.Info("%s", req)
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- BootFileErrorMsg := "Invalid OBS path '" + createJobParams.InfConfig.BootFileUrl + "'."
- DataSetErrorMsg := "Invalid OBS path '" + createJobParams.InfConfig.DataUrl + "'."
- if temp.ErrorMsg == BootFileErrorMsg {
- log.Error("启动文件错误!createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("启动文件错误!")
- }
- if temp.ErrorMsg == DataSetErrorMsg {
- log.Error("数据集错误!createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("数据集错误!")
- }
- if res.StatusCode() == http.StatusBadGateway {
- return &result, fmt.Errorf(UnknownErrorPrefix+"createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- } else {
- return &result, fmt.Errorf("createInferenceJob failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
- }
-
- if !result.IsSuccess {
- log.Error("createInferenceJob failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- return &result, fmt.Errorf("createInferenceJob failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func createInferenceJobUserImage(createJobParams models.CreateInfUserImageParams) (*models.CreateTrainJobResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.CreateTrainJobResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetBody(createJobParams).
- SetResult(&result).
- Post(HOST + "/v1/" + setting.ProjectID + urlTrainJob)
-
- if err != nil {
- return nil, fmt.Errorf("resty create train-job: %s", err)
- }
-
- req, _ := json.Marshal(createJobParams)
- log.Info("%s", req)
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("createInferenceJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- bootFileErrorMsg := "Invalid OBS path '" + createJobParams.Config.BootFileUrl + "'."
- dataSetErrorMsg := "Invalid OBS path '" + createJobParams.Config.DataUrl + "'."
- if temp.ErrorMsg == bootFileErrorMsg {
- log.Error("启动文件错误!createInferenceJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("启动文件错误!")
- }
- if temp.ErrorMsg == dataSetErrorMsg {
- log.Error("数据集错误!createInferenceJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("数据集错误!")
- }
- if res.StatusCode() == http.StatusBadGateway {
- return &result, fmt.Errorf(UnknownErrorPrefix+"createInferenceJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- } else {
- return &result, fmt.Errorf("createInferenceJobUserImage failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
- }
-
- if !result.IsSuccess {
- log.Error("createInferenceJobUserImage failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- return &result, fmt.Errorf("createInferenceJobUserImage failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func createNotebook2(createJobParams models.CreateNotebook2Params) (*models.CreateNotebookResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.CreateNotebookResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetHeader("Content-Type", "application/json").
- SetAuthToken(TOKEN).
- SetBody(createJobParams).
- SetResult(&result).
- Post(HOST + "/v1/" + setting.ProjectID + urlNotebook2)
-
- if err != nil {
- return nil, fmt.Errorf("resty create notebook2: %s", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- var response models.NotebookResult
- err = json.Unmarshal(res.Body(), &response)
- if err != nil {
- log.Error("json.Unmarshal failed: %s", err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed: %s", err.Error())
- }
-
- if res.StatusCode() == http.StatusBadGateway {
- return &result, fmt.Errorf(UnknownErrorPrefix+"createNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- }
-
- if len(response.ErrorCode) != 0 {
- log.Error("createNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- if response.ErrorCode == errorCodeExceedLimit {
- response.ErrorMsg = "所选规格使用数量已超过最大配额限制。"
- }
- if response.ErrorCode == modelartsIllegalToken && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
- return &result, fmt.Errorf("createNotebook2 failed(%s): %s", response.ErrorCode, response.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func GetTrainJobMetricStatistic(jobID, versionID, podName string) (*models.GetTrainJobMetricStatisticResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.GetTrainJobMetricStatisticResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetAuthToken(TOKEN).
- SetResult(&result).
- Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions/" + versionID + "/pod/" + podName + "/metric-statistic?statistic_type=each")
-
- if err != nil {
- return nil, fmt.Errorf("resty GetTrainJobMetricStatistic: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("GetTrainJobMetricStatistic failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf("GetTrainJobMetricStatistic failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- }
-
- if !result.IsSuccess {
- log.Error("GetTrainJobMetricStatistic(%s) failed", jobID)
- return &result, fmt.Errorf("获取任务资源占用情况失败:%s", result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func GetTrainJobList(perPage, page int, sortBy, order, searchContent string) (*models.GetTrainJobListResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.GetTrainJobListResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetQueryParams(map[string]string{
- "per_page": strconv.Itoa(perPage),
- "page": strconv.Itoa(page),
- "sortBy": sortBy,
- "order": order,
- "search_content": searchContent,
- }).
- SetAuthToken(TOKEN).
- SetResult(&result).
- Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob)
-
- if err != nil {
- return nil, fmt.Errorf("resty GetTrainJobList: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("GetTrainJobList failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf(temp.ErrorMsg)
- }
-
- if !result.IsSuccess {
- log.Error("GetTrainJobList failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- return &result, fmt.Errorf(result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func GetTrainJobVersionList(perPage, page int, jobID string) (*models.GetTrainJobVersionListResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.GetTrainJobVersionListResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetQueryParams(map[string]string{
- "per_page": strconv.Itoa(perPage),
- "page": strconv.Itoa(page),
- }).
- SetAuthToken(TOKEN).
- SetResult(&result).
- Get(HOST + "/v1/" + setting.ProjectID + urlTrainJob + "/" + jobID + "/versions")
-
- if err != nil {
- return nil, fmt.Errorf("resty GetTrainJobVersionList: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("GetTrainJobVersionList failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf(temp.ErrorMsg)
- }
-
- if !result.IsSuccess {
- log.Error("GetTrainJobVersionList failed(%s): %s", result.ErrorCode, result.ErrorMsg)
- return &result, fmt.Errorf(result.ErrorMsg)
- }
-
- return &result, nil
- }
-
- func GetNotebookList(limit, offset int, sortBy, order, searchContent string) (*models.GetNotebookListResult, error) {
- checkSetting()
- client := getRestyClient()
- var result models.GetNotebookListResult
-
- retry := 0
-
- sendjob:
- res, err := client.R().
- SetQueryParams(map[string]string{
- "limit": strconv.Itoa(limit),
- "offset": strconv.Itoa(offset),
- "name": searchContent,
- "sort_key": sortBy,
- "sort_dir": order,
- }).
- SetAuthToken(TOKEN).
- SetResult(&result).
- Get(HOST + "/v1/" + setting.ProjectID + urlNotebook2)
-
- if err != nil {
- return nil, fmt.Errorf("resty GetNotebookList: %v", err)
- }
-
- if res.StatusCode() == http.StatusUnauthorized && retry < 1 {
- retry++
- _ = getToken()
- goto sendjob
- }
-
- if res.StatusCode() != http.StatusOK {
- var temp models.ErrorResult
- if err = json.Unmarshal([]byte(res.String()), &temp); err != nil {
- log.Error("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- return &result, fmt.Errorf("json.Unmarshal failed(%s): %v", res.String(), err.Error())
- }
- log.Error("GetNotebookList failed(%d):%s(%s)", res.StatusCode(), temp.ErrorCode, temp.ErrorMsg)
- return &result, fmt.Errorf(temp.ErrorMsg)
- }
-
- return &result, nil
- }
|