|
12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487 |
- 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("%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
-
- 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
- }
|