From b2eb0a6156199173771691c81b45376438b8b780 Mon Sep 17 00:00:00 2001 From: qiwang <1364512070@qq.com> Date: Thu, 16 May 2024 15:49:53 +0800 Subject: [PATCH] fix: update ai Replicas Former-commit-id: f08e733c5be93758e45a19808851d12569a4f2d3 --- api/desc/schedule/pcm-schedule.api | 1 + api/internal/logic/core/commitvmtasklogic.go | 2 -- api/internal/logic/schedule/schedulesubmitlogic.go | 2 +- api/internal/scheduler/database/aiStorage.go | 2 +- api/internal/scheduler/schedulers/aiScheduler.go | 2 +- api/internal/scheduler/schedulers/option/aiOption.go | 2 +- api/internal/types/types.go | 1 + 7 files changed, 6 insertions(+), 6 deletions(-) diff --git a/api/desc/schedule/pcm-schedule.api b/api/desc/schedule/pcm-schedule.api index 2fd6e4f8..4a894e34 100644 --- a/api/desc/schedule/pcm-schedule.api +++ b/api/desc/schedule/pcm-schedule.api @@ -41,6 +41,7 @@ type ( Params []string `json:"params,optional"` Envs []string `json:"envs,optional"` Cmd string `json:"cmd,optional"` + Replica int32 `json:"replicas"` } AiResourceTypesResp { diff --git a/api/internal/logic/core/commitvmtasklogic.go b/api/internal/logic/core/commitvmtasklogic.go index ff378cc4..f339580a 100644 --- a/api/internal/logic/core/commitvmtasklogic.go +++ b/api/internal/logic/core/commitvmtasklogic.go @@ -110,7 +110,6 @@ func (l *CommitVmTaskLogic) CommitVmTask(req *types.CommitVmTaskReq) (resp *type AND vf.public_flavor_id = ? AND vi.public_image_id = ? AND vn.public_network_id = ?` - // err2 := l.svcCtx.DbEngin.Raw(sql, clusterId, req.FlavorRef, req.ImageRef, req.Uuid).Scan(&taskVm).Error txVm := l.svcCtx.DbEngin.Raw(sql, clusterId, req.FlavorRef, req.ImageRef, req.Uuid).Scan(&VmObject) if txVm.Error != nil { logx.Error(err) @@ -118,7 +117,6 @@ func (l *CommitVmTaskLogic) CommitVmTask(req *types.CommitVmTaskReq) (resp *type } if err2 != nil { logx.Errorf("CommitGeneralTask() => sql execution error: %v", err) - //return errors.Errorf("the cluster does not match the drive resources. Check the data"), nil } taskVm.Name = req.VmName taskVm.Status = "Saved" diff --git a/api/internal/logic/schedule/schedulesubmitlogic.go b/api/internal/logic/schedule/schedulesubmitlogic.go index b2aa78f5..6c23a99f 100644 --- a/api/internal/logic/schedule/schedulesubmitlogic.go +++ b/api/internal/logic/schedule/schedulesubmitlogic.go @@ -31,7 +31,7 @@ func (l *ScheduleSubmitLogic) ScheduleSubmit(req *types.ScheduleReq) (resp *type AdapterId: req.AiOption.AdapterId, TaskName: req.AiOption.TaskName, ResourceType: req.AiOption.ResourceType, - Replica: 1, + Replica: req.AiOption.Replica, Tops: req.AiOption.Tops, TaskType: req.AiOption.TaskType, DatasetsName: req.AiOption.Datasets, diff --git a/api/internal/scheduler/database/aiStorage.go b/api/internal/scheduler/database/aiStorage.go index f25cd5c2..24806cdd 100644 --- a/api/internal/scheduler/database/aiStorage.go +++ b/api/internal/scheduler/database/aiStorage.go @@ -105,7 +105,7 @@ func (s *AiStorage) SaveAiTask(taskId int64, option *option.AiOption, clusterId AdapterId: aId, ClusterId: cId, Name: option.TaskName, - Replica: option.Replica, + Replica: int64(option.Replica), JobId: jobId, TaskType: option.TaskType, Strategy: option.StrategyName, diff --git a/api/internal/scheduler/schedulers/aiScheduler.go b/api/internal/scheduler/schedulers/aiScheduler.go index 19f67b83..5e4cf8b9 100644 --- a/api/internal/scheduler/schedulers/aiScheduler.go +++ b/api/internal/scheduler/schedulers/aiScheduler.go @@ -103,7 +103,7 @@ func (as *AiScheduler) PickOptimalStrategy() (strategy.Strategy, error) { return strategy, nil case strategy.STATIC_WEIGHT: //todo resources should match cluster StaticWeightMap - strategy := strategy.NewStaticWeightStrategy(as.option.ClusterToStaticWeight, 1) + strategy := strategy.NewStaticWeightStrategy(as.option.ClusterToStaticWeight, as.option.Replica) return strategy, nil } diff --git a/api/internal/scheduler/schedulers/option/aiOption.go b/api/internal/scheduler/schedulers/option/aiOption.go index d2f8d3eb..b3cc4de0 100644 --- a/api/internal/scheduler/schedulers/option/aiOption.go +++ b/api/internal/scheduler/schedulers/option/aiOption.go @@ -4,7 +4,7 @@ type AiOption struct { AdapterId string ClusterIds []string TaskName string - Replica int64 + Replica int32 ResourceType string // cpu/gpu/compute card CpuCoreNum int64 TaskType string // pytorch/tensorflow/mindspore diff --git a/api/internal/types/types.go b/api/internal/types/types.go index e1620f47..853962ba 100644 --- a/api/internal/types/types.go +++ b/api/internal/types/types.go @@ -5642,6 +5642,7 @@ type AiOption struct { Params []string `json:"params,optional"` Envs []string `json:"envs,optional"` Cmd string `json:"cmd,optional"` + Replica int32 `json:"replicas"` } type AiResourceTypesResp struct {