From 4cb7633870439df347fd9e093edb26e72e932d49 Mon Sep 17 00:00:00 2001 From: tzwang Date: Thu, 19 Jun 2025 11:59:07 +0800 Subject: [PATCH] update octopusHttp service --- internal/scheduler/schedulers/aiScheduler.go | 11 ++++++++--- internal/scheduler/service/aiService.go | 2 +- internal/storeLink/octopusHttp/octopusHttp.go | 14 +++++++++----- 3 files changed, 18 insertions(+), 9 deletions(-) diff --git a/internal/scheduler/schedulers/aiScheduler.go b/internal/scheduler/schedulers/aiScheduler.go index 403298ee..11921477 100644 --- a/internal/scheduler/schedulers/aiScheduler.go +++ b/internal/scheduler/schedulers/aiScheduler.go @@ -471,13 +471,18 @@ func convertType(in interface{}) (*AiResult, error) { return &result, nil case *entity.OctCreateJobResp: - resp := (in).(entity.OctCreateJobResp) + resp := (in).(*entity.OctCreateJobResp) if resp.Code != 200 { result.Msg = resp.Msg } else { - job := (resp.Data).(*omodel.OctCreateJob) - result.JobId = job.JobId + var cj omodel.OctCreateJob + m, _ := json.Marshal(resp.Data) + err := json.Unmarshal([]byte(m), &cj) + if err != nil { + return nil, err + } + result.JobId = cj.JobId } return &result, nil diff --git a/internal/scheduler/service/aiService.go b/internal/scheduler/service/aiService.go index f38c0392..b0da7e5b 100644 --- a/internal/scheduler/service/aiService.go +++ b/internal/scheduler/service/aiService.go @@ -73,7 +73,7 @@ func InitAiClusterMap(conf *config.Config, clusters []types.ClusterInfo) (map[st collectorMap := make(map[string]collector.AiCollector) inferenceMap := make(map[string]inference.ICluster) for _, c := range clusters { - switch c.Name { + switch c.Driver { case OCTOPUS: id, _ := strconv.ParseInt(c.Id, 10, 64) octopus := octopusHttp.NewOctopusHttp(id, c.Nickname, c.Server, c.Address, c.Username, c.Password) diff --git a/internal/storeLink/octopusHttp/octopusHttp.go b/internal/storeLink/octopusHttp/octopusHttp.go index 74d8a464..228141ea 100644 --- a/internal/storeLink/octopusHttp/octopusHttp.go +++ b/internal/storeLink/octopusHttp/octopusHttp.go @@ -129,7 +129,6 @@ func (o *OctopusHttp) Stop(ctx context.Context, id string) error { } func (o *OctopusHttp) SubmitTask(ctx context.Context, imageId string, cmd string, envs []string, params []string, resourceId string, datasetsId string, algorithmId string, aiType string) (interface{}, error) { - // octopus提交任务 reqUrl := o.server + CreateTrainJobUrl @@ -188,6 +187,8 @@ func (o *OctopusHttp) SubmitTask(ctx context.Context, imageId string, cmd string req := common.GetRestyRequest(common.TIMEOUT) _, err = req. SetHeader("Authorization", "Bearer "+token). + SetQueryString("token=" + token). + SetQueryString("addr=" + o.host). SetBody(param). SetResult(resp). Post(reqUrl) @@ -531,13 +532,16 @@ func genClusterResources(cType string, cNum string, s *omodel.Spec) (*collector. cres := &collector.ClusterResource{} bres := make([]*collector.Usage, 0) + var cardNum int64 + var cpuCore int64 + cardNum, err := strconv.ParseInt(cNum, 10, 64) if err != nil { - return nil, err + cardNum = 0 } - cpuCore, err := strconv.ParseInt(s.ResourceQuantity.Cpu, 10, 64) + cpuCore, err = strconv.ParseInt(s.ResourceQuantity.Cpu, 10, 64) if err != nil { - return nil, err + cpuCore = 0 } gi := strings.Split(s.ResourceQuantity.Memory, Gi) if len(gi) != 2 { @@ -546,7 +550,7 @@ func genClusterResources(cType string, cNum string, s *omodel.Spec) (*collector. memGi, err := strconv.ParseInt(gi[0], 10, 64) if err != nil { - return nil, err + memGi = 0 } card := &collector.Usage{