floraachy
/
pcm-coordinator
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request 'scheduler module modified' (#12) from tzwang/pcm-coordinator:master into master 

Former-commit-id: 7cceae0438dce12298a59c312b1ffd05db791083
This commit is contained in:
tzwang 2024-01-24 18:10:59 +08:00
parent c89fddca14 8882790591
commit bf27a2d75c
23 changed files with 435 additions and 189 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
api/internal/mqs/ScheduleAi.go
View File

@ -17,7 +17,9 @@ package mqs
import ( import (
"context" "context"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/api/internal/svc" "gitlink.org.cn/jcce-pcm/pcm-coordinator/api/internal/svc"
scheduler2 "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler" "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/schedulers"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/service"
) )
/* /*
@ -25,36 +27,29 @@ import (
Listening to the payment flow status change notification message queue Listening to the payment flow status change notification message queue
*/ */
type AiQueue struct { type AiQueue struct {
ctx context.Context ctx context.Context
svcCtx *svc.ServiceContext svcCtx *svc.ServiceContext
scheduler *scheduler.Scheduler
} }
func NewAiMq(ctx context.Context, svcCtx *svc.ServiceContext) *AiQueue { func NewAiMq(ctx context.Context, svcCtx *svc.ServiceContext) *AiQueue {
aiExecutorMap, aiCollectorMap := service.InitAiClusterMap(svcCtx.ACRpc, svcCtx.ModelArtsRpc, svcCtx.ModelArtsImgRpc, svcCtx.OctopusRpc)
return &AiQueue{ return &AiQueue{
ctx: ctx, ctx: ctx,
svcCtx: svcCtx, svcCtx: svcCtx,
scheduler: scheduler.NewScheduler2(aiCollectorMap, nil, aiExecutorMap),
} }
} }
func (l *AiQueue) Consume(val string) error { func (l *AiQueue) Consume(val string) error {
// 接受消息, 根据标签筛选过滤 // 接受消息, 根据标签筛选过滤
aiSchdl := scheduler2.NewAiScheduler(val) aiSchdl, _ := schedulers.NewAiScheduler(val, l.scheduler)
schdl, err := scheduler2.NewScheduler(aiSchdl, val, l.svcCtx.DbEngin, nil)
if err != nil {
return err
}
schdl.MatchLabels()
// 调度算法 // 调度算法
err = schdl.AssignAndSchedule() err := l.scheduler.AssignAndSchedule(aiSchdl)
if err != nil { if err != nil {
return err return err
} }
// 存储数据
err = schdl.SaveToDb()
if err != nil {
return err
}
return nil return nil
} }

3
api/internal/mqs/ScheduleCloud.go
View File

@ -18,6 +18,7 @@ import (
"context" "context"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/api/internal/svc" "gitlink.org.cn/jcce-pcm/pcm-coordinator/api/internal/svc"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler" "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/schedulers"
) )
/* /*
@ -37,7 +38,7 @@ func NewCloudMq(ctx context.Context, svcCtx *svc.ServiceContext) *CloudMq {
func (l *CloudMq) Consume(val string) error { func (l *CloudMq) Consume(val string) error {
// 接受消息, 根据标签筛选过滤 // 接受消息, 根据标签筛选过滤
cloudScheduler := scheduler.NewCloudScheduler() cloudScheduler := schedulers.NewCloudScheduler()
schdl, err := scheduler.NewScheduler(cloudScheduler, val, l.svcCtx.DbEngin, l.svcCtx.ParticipantRpc) schdl, err := scheduler.NewScheduler(cloudScheduler, val, l.svcCtx.DbEngin, l.svcCtx.ParticipantRpc)
if err != nil { if err != nil {
return err return err

20
api/internal/mqs/ScheduleHpc.go
View File

@ -17,7 +17,6 @@ package mqs
import ( import (
"context" "context"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/api/internal/svc" "gitlink.org.cn/jcce-pcm/pcm-coordinator/api/internal/svc"
scheduler2 "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler"
) )
/* /*
@ -37,24 +36,5 @@ func NewHpcMq(ctx context.Context, svcCtx *svc.ServiceContext) *HpcMq {
} }
func (l *HpcMq) Consume(val string) error { func (l *HpcMq) Consume(val string) error {
// 接受消息, 根据标签筛选过滤
hpcSchdl := scheduler2.NewHpcScheduler(val)
schdl, err := scheduler2.NewScheduler(hpcSchdl, val, l.svcCtx.DbEngin, nil)
if err != nil {
return err
}
schdl.MatchLabels()
// 调度算法
err = schdl.AssignAndSchedule()
if err != nil {
return err
}
// 存储数据
err = schdl.SaveToDb()
if err != nil {
return err
}
return nil return nil
} }

58
pkg/scheduler/aiScheduler.go
View File

@ -1,58 +0,0 @@
/*
Copyright (c) [2023] [pcm]
[pcm-coordinator] is licensed under Mulan PSL v2.
You can use this software according to the terms and conditions of the Mulan PSL v2.
You may obtain a copy of Mulan PSL v2 at:
http://license.coscl.org.cn/MulanPSL2
THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
EITHER EXPaRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
See the Mulan PSL v2 for more details.
*/
package scheduler
import (
"gitlink.org.cn/jcce-pcm/pcm-coordinator/api/pkg/response"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/models"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/algorithm/providerPricing"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/collector"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/strategies"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/utils"
)
type AiScheduler struct {
yamlString string
collector collector.ResourceCollector
}
func NewAiScheduler(val string) *AiScheduler {
return &AiScheduler{yamlString: val}
}
func (as *AiScheduler) getNewStructForDb(task *response.TaskInfo, resource string, participantId int64) (interface{}, error) {
ai := models.Ai{
ParticipantId: participantId,
TaskId: task.TaskId,
Status: "Saved",
YamlString: as.yamlString,
}
utils.Convert(task.Metadata, &ai)
return ai, nil
}
func (as *AiScheduler) pickOptimalStrategy() (strategies.Strategy, error) {
//a, b := as.genTaskAndProviders()
return nil, nil
}
func (as *AiScheduler) genTaskAndProviders() (*providerPricing.Task, []*providerPricing.Provider) {
return nil, nil
}
func (as *AiScheduler) assignTask(clusters []*strategies.AssignedCluster) error {
return nil
}

13
pkg/scheduler/collector/acCollector.go
View File

@ -1,13 +0,0 @@
package collector
//单条作业费=作业运行秒数×(CPU核心数*CPU单价+GPU卡数×GPU单价+DCU卡数×DCU单价)/3600
//CPU单价=队列CPU费率×计算中心CPU单价
//GPU单价=队列GPU费率×计算中心GPU单价
//DCU单价=队列DCU费率×计算中心DCU单价
type ShuguangAiCollector struct {
}
func (a *ShuguangAiCollector) getResourceSpecs() {
}

19
pkg/scheduler/collector/collector.go
View File

@ -1,19 +0,0 @@
package collector
type ResourceCollector interface {
getResourceSpecs() ([]ResourceSpecs, error)
}
type ResourceSpecs struct {
CpuAvail float64
MemAvail float64
DiskAvail float64
GpuAvail float64
CardAvail []Card
}
type Card struct {
Type string
Name string
TOpsAtFp16 float64
}

16
pkg/scheduler/common.go → pkg/scheduler/common/common.go
View File

@ -12,23 +12,23 @@
*/ */
package scheduler package common
import ( import (
"gitlink.org.cn/jcce-pcm/pcm-coordinator/api/pkg/response" "gitlink.org.cn/jcce-pcm/pcm-coordinator/api/pkg/response"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/strategies" "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/strategy"
"math/rand" "math/rand"
"time" "time"
) )
type scheduleService interface { type SubSchedule interface {
getNewStructForDb(task *response.TaskInfo, resource string, participantId int64) (interface{}, error) GetNewStructForDb(task *response.TaskInfo, resource string, participantId int64) (interface{}, error)
pickOptimalStrategy() (strategies.Strategy, error) PickOptimalStrategy() (strategy.Strategy, error)
assignTask(clusters []*strategies.AssignedCluster) error AssignTask(clusters []*strategy.AssignedCluster) error
} }
// 求交集 // 求交集
func intersect(slice1, slice2 []int64) []int64 { func Intersect(slice1, slice2 []int64) []int64 {
m := make(map[int64]int) m := make(map[int64]int)
nn := make([]int64, 0) nn := make([]int64, 0)
for _, v := range slice1 { for _, v := range slice1 {
@ -44,7 +44,7 @@ func intersect(slice1, slice2 []int64) []int64 {
return nn return nn
} }
func micsSlice(origin []int64, count int) []int64 { func MicsSlice(origin []int64, count int) []int64 {
tmpOrigin := make([]int64, len(origin)) tmpOrigin := make([]int64, len(origin))
copy(tmpOrigin, origin) copy(tmpOrigin, origin)
//一定要seed //一定要seed

87
pkg/scheduler/scheduler.go
View File

@ -19,31 +19,42 @@ import (
"github.com/pkg/errors" "github.com/pkg/errors"
"github.com/zeromicro/go-zero/core/logx" "github.com/zeromicro/go-zero/core/logx"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/api/pkg/response" "gitlink.org.cn/jcce-pcm/pcm-coordinator/api/pkg/response"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/common"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/database"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/service/collector"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/service/executor"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/rpc/client/participantservice" "gitlink.org.cn/jcce-pcm/pcm-coordinator/rpc/client/participantservice"
"gorm.io/gorm" "gorm.io/gorm"
"sigs.k8s.io/yaml" "sigs.k8s.io/yaml"
"strings" "strings"
) )
type scheduler struct { type Scheduler struct {
task *response.TaskInfo task *response.TaskInfo
participantIds []int64 participantIds []int64
scheduleService scheduleService subSchedule common.SubSchedule
dbEngin *gorm.DB dbEngin *gorm.DB
result []string //pID:子任务yamlstring 键值对 result []string //pID:子任务yamlstring 键值对
participantRpc participantservice.ParticipantService participantRpc participantservice.ParticipantService
ResourceCollector *map[string]collector.ResourceCollector
Storages database.Storage
AiExecutor *map[string]executor.Executor
} }
func NewScheduler(scheduleService scheduleService, val string, dbEngin *gorm.DB, participantRpc participantservice.ParticipantService) (*scheduler, error) { func NewScheduler(subSchedule common.SubSchedule, val string, dbEngin *gorm.DB, participantRpc participantservice.ParticipantService) (*Scheduler, error) {
var task *response.TaskInfo var task *response.TaskInfo
err := json.Unmarshal([]byte(val), &task) err := json.Unmarshal([]byte(val), &task)
if err != nil { if err != nil {
return nil, errors.New("create scheduler failed : " + err.Error()) return nil, errors.New("create scheduler failed : " + err.Error())
} }
return &scheduler{task: task, scheduleService: scheduleService, dbEngin: dbEngin, participantRpc: participantRpc}, nil return &Scheduler{task: task, subSchedule: subSchedule, dbEngin: dbEngin, participantRpc: participantRpc}, nil
} }
func (s *scheduler) SpecifyClusters() { func NewScheduler2(resourceCollector *map[string]collector.ResourceCollector, storages database.Storage, aiExecutor *map[string]executor.Executor) *Scheduler {
return &Scheduler{ResourceCollector: resourceCollector, Storages: storages, AiExecutor: aiExecutor}
}
func (s *Scheduler) SpecifyClusters() {
// 如果已指定集群名，通过数据库查询后返回p端ip列表 // 如果已指定集群名，通过数据库查询后返回p端ip列表
if len(s.task.Clusters) != 0 { if len(s.task.Clusters) != 0 {
s.dbEngin.Raw("select id from sc_participant_phy_info where `name` in (?)", s.task.Clusters).Scan(&s.participantIds) s.dbEngin.Raw("select id from sc_participant_phy_info where `name` in (?)", s.task.Clusters).Scan(&s.participantIds)
@ -51,7 +62,7 @@ func (s *scheduler) SpecifyClusters() {
} }
} }
func (s *scheduler) SpecifyNsID() { func (s *Scheduler) SpecifyNsID() {
// 未指定集群名，只指定nsID // 未指定集群名，只指定nsID
if len(s.task.Clusters) == 0 { if len(s.task.Clusters) == 0 {
if len(s.task.NsID) != 0 { if len(s.task.NsID) != 0 {
@ -67,7 +78,7 @@ func (s *scheduler) SpecifyNsID() {
} }
} }
func (s *scheduler) MatchLabels() { func (s *Scheduler) MatchLabels() {
var ids []int64 var ids []int64
count := 0 count := 0
@ -80,7 +91,7 @@ func (s *scheduler) MatchLabels() {
if count == 0 { if count == 0 {
ids = participantIds ids = participantIds
} }
ids = intersect(ids, participantIds) ids = common.Intersect(ids, participantIds)
count++ count++
} }
s.participantIds = ids s.participantIds = ids
@ -90,7 +101,7 @@ func (s *scheduler) MatchLabels() {
} }
// TempAssign todo 屏蔽原调度算法 // TempAssign todo 屏蔽原调度算法
func (s *scheduler) TempAssign() error { func (s *Scheduler) TempAssign() error {
//需要判断task中的资源类型，针对metadata中的多个kind做不同处理 //需要判断task中的资源类型，针对metadata中的多个kind做不同处理
//输入副本数和集群列表，最终结果输出为pID对应副本数量列表，针对多个kind需要做拆分和重新拼接组合 //输入副本数和集群列表，最终结果输出为pID对应副本数量列表，针对多个kind需要做拆分和重新拼接组合
@ -110,28 +121,28 @@ func (s *scheduler) TempAssign() error {
return nil return nil
} }
func (s *scheduler) AssignAndSchedule() error { func (s *Scheduler) AssignAndSchedule(ss common.SubSchedule) error {
// 已指定 ParticipantId //// 已指定 ParticipantId
if s.task.ParticipantId != 0 { //if s.task.ParticipantId != 0 {
return nil // return nil
} //}
// 标签匹配以及后，未找到ParticipantIds //// 标签匹配以及后，未找到ParticipantIds
if len(s.participantIds) == 0 { //if len(s.participantIds) == 0 {
return errors.New("未找到匹配的ParticipantIds") // return errors.New("未找到匹配的ParticipantIds")
} //}
//
//// 指定或者标签匹配的结果只有一个集群，给任务信息指定
//if len(s.participantIds) == 1 {
// s.task.ParticipantId = s.participantIds[0]
// //replicas := s.task.Metadata.(map[string]interface{})["spec"].(map[string]interface{})["replicas"].(float64)
// //result := make(map[int64]string)
// //result[s.participantIds[0]] = strconv.FormatFloat(replicas, 'f', 2, 64)
// //s.result = result
//
// return nil
//}
// 指定或者标签匹配的结果只有一个集群，给任务信息指定 strategy, err := ss.PickOptimalStrategy()
if len(s.participantIds) == 1 {
s.task.ParticipantId = s.participantIds[0]
//replicas := s.task.Metadata.(map[string]interface{})["spec"].(map[string]interface{})["replicas"].(float64)
//result := make(map[int64]string)
//result[s.participantIds[0]] = strconv.FormatFloat(replicas, 'f', 2, 64)
//s.result = result
return nil
}
strategy, err := s.scheduleService.pickOptimalStrategy()
if err != nil { if err != nil {
return err return err
} }
@ -147,7 +158,7 @@ func (s *scheduler) AssignAndSchedule() error {
// return nil // return nil
//} //}
err = s.scheduleService.assignTask(clusters) err = ss.AssignTask(clusters)
if err != nil { if err != nil {
return err return err
} }
@ -155,12 +166,12 @@ func (s *scheduler) AssignAndSchedule() error {
return nil return nil
} }
func (s *scheduler) SaveToDb() error { func (s *Scheduler) SaveToDb() error {
for _, participantId := range s.participantIds { for _, participantId := range s.participantIds {
for _, resource := range s.task.Metadata { for _, resource := range s.task.Metadata {
structForDb, err := s.scheduleService.getNewStructForDb(s.task, resource, participantId) structForDb, err := s.subSchedule.GetNewStructForDb(s.task, resource, participantId)
if err != nil { if err != nil {
return err return err
} }

102
pkg/scheduler/schedulers/aiScheduler.go Normal file
View File

@ -0,0 +1,102 @@
/*
Copyright (c) [2023] [pcm]
[pcm-coordinator] is licensed under Mulan PSL v2.
You can use this software according to the terms and conditions of the Mulan PSL v2.
You may obtain a copy of Mulan PSL v2 at:
http://license.coscl.org.cn/MulanPSL2
THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
EITHER EXPaRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
See the Mulan PSL v2 for more details.
*/
package schedulers
import (
"errors"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/api/pkg/response"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/models"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/algorithm/providerPricing"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/entity"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/service/collector"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/strategy"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/utils"
)
type AiScheduler struct {
yamlString string
task *response.TaskInfo
*scheduler.Scheduler
}
func NewAiScheduler(val string, scheduler *scheduler.Scheduler) (*AiScheduler, error) {
return &AiScheduler{yamlString: val, Scheduler: scheduler}, nil
}
func (as *AiScheduler) GetNewStructForDb(task *response.TaskInfo, resource string, participantId int64) (interface{}, error) {
ai := models.Ai{
ParticipantId: participantId,
TaskId: task.TaskId,
Status: "Saved",
YamlString: as.yamlString,
}
utils.Convert(task.Metadata, &ai)
return ai, nil
}
func (as *AiScheduler) PickOptimalStrategy() (strategy.Strategy, error) {
resources, err := as.findProvidersWithResource()
if err != nil {
return nil, err
}
if len(resources) < 2 /*|| as.task */ {
var pros []entity.Participant
for _, resource := range resources {
pros = append(pros, entity.Participant{
Participant_id: resource.ParticipantId,
Name: resource.Name,
})
}
strategy := strategy.NewReplicationStrategy(nil, 0)
return strategy, nil
}
task, providerList := as.genTaskAndProviders()
if err != nil {
return nil, nil
}
strategy := strategy.NewPricingStrategy(task, providerList...)
return strategy, nil
}
func (as *AiScheduler) genTaskAndProviders() (*providerPricing.Task, []*providerPricing.Provider) {
return nil, nil
}
func (as *AiScheduler) AssignTask(clusters []*strategy.AssignedCluster) error {
if clusters == nil {
return errors.New("clusters is nil")
}
return nil
}
func (as *AiScheduler) findProvidersWithResource() ([]*collector.ResourceSpecs, error) {
var resourceSpecs []*collector.ResourceSpecs
for _, resourceCollector := range *as.ResourceCollector {
spec, err := resourceCollector.GetResourceSpecs()
if err != nil {
continue
}
resourceSpecs = append(resourceSpecs, spec)
}
if len(resourceSpecs) == 0 {
return nil, errors.New("no resource found")
}
return resourceSpecs, nil
}

12
pkg/scheduler/cloudScheduler.go → pkg/scheduler/schedulers/cloudScheduler.go
View File

@ -12,7 +12,7 @@
*/ */
package scheduler package schedulers
import ( import (
"bytes" "bytes"
@ -20,7 +20,7 @@ import (
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/models" "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/models"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/algorithm/providerPricing" "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/algorithm/providerPricing"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/database" "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/database"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/strategies" "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/strategy"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/utils" "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/utils"
"io" "io"
"k8s.io/apimachinery/pkg/apis/meta/v1/unstructured" "k8s.io/apimachinery/pkg/apis/meta/v1/unstructured"
@ -37,17 +37,17 @@ func NewCloudScheduler() *CloudScheduler {
return &CloudScheduler{} return &CloudScheduler{}
} }
func (cs *CloudScheduler) pickOptimalStrategy() (strategies.Strategy, error) { func (cs *CloudScheduler) PickOptimalStrategy() (strategy.Strategy, error) {
task, providerList, err := cs.genTaskAndProviders() task, providerList, err := cs.genTaskAndProviders()
if err != nil { if err != nil {
return nil, nil return nil, nil
} }
//调度算法 //调度算法
strategy := strategies.NewPricingStrategy(task, providerList...) strategy := strategy.NewPricingStrategy(task, providerList...)
return strategy, nil return strategy, nil
} }
func (cs *CloudScheduler) getNewStructForDb(task *response.TaskInfo, resource string, participantId int64) (interface{}, error) { func (cs *CloudScheduler) GetNewStructForDb(task *response.TaskInfo, resource string, participantId int64) (interface{}, error) {
cloud := cs.UnMarshalK8sStruct(resource, task.TaskId, task.NsID) cloud := cs.UnMarshalK8sStruct(resource, task.TaskId, task.NsID)
cloud.Id = utils.GenSnowflakeID() cloud.Id = utils.GenSnowflakeID()
cloud.NsID = task.NsID cloud.NsID = task.NsID
@ -117,6 +117,6 @@ func (cs *CloudScheduler) genTaskAndProviders() (*providerPricing.Task, []*provi
return nil, providerList, nil return nil, providerList, nil
} }
func (cs *CloudScheduler) assignTask(clusters []*strategies.AssignedCluster) error { func (cs *CloudScheduler) AssignTask(clusters []*strategy.AssignedCluster) error {
return nil return nil
} }

10
pkg/scheduler/hpcScheduler.go → pkg/scheduler/schedulers/hpcScheduler.go
View File

@ -12,14 +12,14 @@
*/ */
package scheduler package schedulers
import ( import (
"gitlink.org.cn/jcce-pcm/pcm-coordinator/api/pkg/response" "gitlink.org.cn/jcce-pcm/pcm-coordinator/api/pkg/response"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/constants" "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/constants"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/models" "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/models"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/algorithm/providerPricing" "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/algorithm/providerPricing"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/strategies" "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/strategy"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/utils" "gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/utils"
) )
@ -31,7 +31,7 @@ func NewHpcScheduler(val string) *HpcScheduler {
return &HpcScheduler{yamlString: val} return &HpcScheduler{yamlString: val}
} }
func (h *HpcScheduler) getNewStructForDb(task *response.TaskInfo, resource string, participantId int64) (interface{}, error) { func (h *HpcScheduler) GetNewStructForDb(task *response.TaskInfo, resource string, participantId int64) (interface{}, error) {
hpc := models.Hpc{} hpc := models.Hpc{}
utils.Convert(task.Metadata, &hpc) utils.Convert(task.Metadata, &hpc)
hpc.Id = utils.GenSnowflakeID() hpc.Id = utils.GenSnowflakeID()
@ -42,7 +42,7 @@ func (h *HpcScheduler) getNewStructForDb(task *response.TaskInfo, resource strin
return hpc, nil return hpc, nil
} }
func (h *HpcScheduler) pickOptimalStrategy() (strategies.Strategy, error) { func (h *HpcScheduler) PickOptimalStrategy() (strategy.Strategy, error) {
return nil, nil return nil, nil
} }
@ -50,6 +50,6 @@ func (h *HpcScheduler) genTaskAndProviders(task *response.TaskInfo) (*providerPr
return nil, nil return nil, nil
} }
func (h *HpcScheduler) assignTask(clusters []*strategies.AssignedCluster) error { func (h *HpcScheduler) AssignTask(clusters []*strategy.AssignedCluster) error {
return nil return nil
} }

24
pkg/scheduler/schedulers/vmScheduler.go Normal file
View File

@ -0,0 +1,24 @@
package schedulers
import (
"gitlink.org.cn/jcce-pcm/pcm-coordinator/api/pkg/response"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/strategy"
)
type VmScheduler struct {
}
func (v VmScheduler) GetNewStructForDb(task *response.TaskInfo, resource string, participantId int64) (interface{}, error) {
//TODO implement me
panic("implement me")
}
func (v VmScheduler) PickOptimalStrategy() (strategy.Strategy, error) {
//TODO implement me
panic("implement me")
}
func (v VmScheduler) AssignTask(clusters []*strategy.AssignedCluster) error {
//TODO implement me
panic("implement me")
}

49
pkg/scheduler/service/aiService.go Normal file
View File

@ -0,0 +1,49 @@
package service
import (
"gitlink.org.cn/jcce-pcm/pcm-ac/hpcacclient"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/service/collector"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/service/executor"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/service/impl"
"gitlink.org.cn/jcce-pcm/pcm-participant-modelarts/client/imagesservice"
"gitlink.org.cn/jcce-pcm/pcm-participant-modelarts/client/modelartsservice"
"gitlink.org.cn/jcce-pcm/pcm-participant-octopus/octopusclient"
)
const (
OCTOPUS = "Octopus"
MODELARTS = "Modelarts"
SHUGUANGAI = "ShuguangAi"
)
var (
AiTypeMap = map[string]string{
"Hanwuji": OCTOPUS,
"Suiyan": OCTOPUS,
"Sailingsi": OCTOPUS,
"modelarts-CloudBrain2": MODELARTS,
"ShuguangAi": SHUGUANGAI,
}
)
func InitAiClusterMap(ACRpc hpcacclient.HpcAC, ModelArtsRpc modelartsservice.ModelArtsService, ModelArtsImgRpc imagesservice.ImagesService, OctopusRpc octopusclient.Octopus) (*map[string]executor.Executor, *map[string]collector.ResourceCollector) {
executorMap := make(map[string]executor.Executor)
collectorMap := make(map[string]collector.ResourceCollector)
for k, v := range AiTypeMap {
switch v {
case OCTOPUS:
octopus := impl.NewOctopusExecutor(OctopusRpc, k)
collectorMap[k] = octopus
executorMap[k] = octopus
case MODELARTS:
modelarts := impl.NewModelartsExecutor(ModelArtsRpc, ModelArtsImgRpc, k)
collectorMap[k] = modelarts
executorMap[k] = modelarts
case SHUGUANGAI:
sgai := impl.NewShuguangAiExecutor(ACRpc, k)
collectorMap[k] = sgai
executorMap[k] = sgai
}
}
return &executorMap, &collectorMap
}

22
pkg/scheduler/service/collector/collector.go Normal file
View File

@ -0,0 +1,22 @@
package collector
type ResourceCollector interface {
GetResourceSpecs() (*ResourceSpecs, error)
}
type ResourceSpecs struct {
ParticipantId int64
Name string
CpuAvail float64
MemAvail float64
DiskAvail float64
GpuAvail float64
CardAvail []Card
Balance float64
}
type Card struct {
Type string
Name string
TOpsAtFp16 float64
}

17
pkg/scheduler/service/executor/executor.go Normal file
View File

@ -0,0 +1,17 @@
package executor
type Executor interface {
QueryImageList() ([]Image, error)
SubmitTask(imageId string, cmd string, envs []string, params []string, resourceId string) (Task, error)
QueryTask(taskId string) (Task, error)
QuerySpecs() (Spec, error)
}
type Image struct {
}
type Task struct {
}
type Spec struct {
}

44
pkg/scheduler/service/impl/modelarts.go Normal file
View File

@ -0,0 +1,44 @@
package impl
import (
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/service/collector"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/service/executor"
"gitlink.org.cn/jcce-pcm/pcm-participant-modelarts/client/imagesservice"
"gitlink.org.cn/jcce-pcm/pcm-participant-modelarts/client/modelartsservice"
)
type ModelArtsExecutor struct {
Name string
pageIndex int32
pageSize int32
ModelArtsRpc modelartsservice.ModelArtsService
ModelArtsImgRpc imagesservice.ImagesService
}
func NewModelartsExecutor(modelArtsRpc modelartsservice.ModelArtsService, modelArtsImgRpc imagesservice.ImagesService, name string) *ModelArtsExecutor {
return &ModelArtsExecutor{Name: name, ModelArtsRpc: modelArtsRpc, ModelArtsImgRpc: modelArtsImgRpc, pageIndex: 1, pageSize: 100}
}
func (m ModelArtsExecutor) QueryImageList() ([]executor.Image, error) {
//TODO implement me
panic("implement me")
}
func (m ModelArtsExecutor) SubmitTask(imageId string, cmd string, envs []string, params []string, resourceId string) (executor.Task, error) {
//TODO implement me
panic("implement me")
}
func (m ModelArtsExecutor) QueryTask(taskId string) (executor.Task, error) {
//TODO implement me
panic("implement me")
}
func (m ModelArtsExecutor) QuerySpecs() (executor.Spec, error) {
//TODO implement me
panic("implement me")
}
func (a *ModelArtsExecutor) GetResourceSpecs() (*collector.ResourceSpecs, error) {
return nil, nil
}

42
pkg/scheduler/service/impl/octopus.go Normal file
View File

@ -0,0 +1,42 @@
package impl
import (
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/service/collector"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/service/executor"
"gitlink.org.cn/jcce-pcm/pcm-participant-octopus/octopusclient"
)
type OctopusExecutor struct {
Name string
pageIndex int32
pageSize int32
OctopusRpc octopusclient.Octopus
}
func NewOctopusExecutor(OctopusRpc octopusclient.Octopus, name string) *OctopusExecutor {
return &OctopusExecutor{OctopusRpc: OctopusRpc, Name: name, pageIndex: 1, pageSize: 100}
}
func (o OctopusExecutor) QueryImageList() ([]executor.Image, error) {
//TODO implement me
panic("implement me")
}
func (o OctopusExecutor) SubmitTask(imageId string, cmd string, envs []string, params []string, resourceId string) (executor.Task, error) {
//TODO implement me
panic("implement me")
}
func (o OctopusExecutor) QueryTask(taskId string) (executor.Task, error) {
//TODO implement me
panic("implement me")
}
func (o OctopusExecutor) QuerySpecs() (executor.Spec, error) {
//TODO implement me
panic("implement me")
}
func (a *OctopusExecutor) GetResourceSpecs() (*collector.ResourceSpecs, error) {
return nil, nil
}

45
pkg/scheduler/service/impl/shuguangAi.go Normal file
View File

@ -0,0 +1,45 @@
package impl
import (
"gitlink.org.cn/jcce-pcm/pcm-ac/hpcacclient"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/service/collector"
"gitlink.org.cn/jcce-pcm/pcm-coordinator/pkg/scheduler/service/executor"
)
//单条作业费=作业运行秒数×(CPU核心数*CPU单价+GPU卡数×GPU单价+DCU卡数×DCU单价)/3600
//CPU单价=队列CPU费率×计算中心CPU单价
//GPU单价=队列GPU费率×计算中心GPU单价
//DCU单价=队列DCU费率×计算中心DCU单价
type ShuguangAiExecutor struct {
Name string
ACRpc hpcacclient.HpcAC
}
func NewShuguangAiExecutor(acRpc hpcacclient.HpcAC, name string) *ShuguangAiExecutor {
return &ShuguangAiExecutor{Name: name, ACRpc: acRpc}
}
func (s ShuguangAiExecutor) QueryImageList() ([]executor.Image, error) {
//TODO implement me
panic("implement me")
}
func (s ShuguangAiExecutor) SubmitTask(imageId string, cmd string, envs []string, params []string, resourceId string) (executor.Task, error) {
//TODO implement me
panic("implement me")
}
func (s ShuguangAiExecutor) QueryTask(taskId string) (executor.Task, error) {
//TODO implement me
panic("implement me")
}
func (s ShuguangAiExecutor) QuerySpecs() (executor.Spec, error) {
//TODO implement me
panic("implement me")
}
func (a *ShuguangAiExecutor) GetResourceSpecs() (*collector.ResourceSpecs, error) {
return nil, nil
}

4
pkg/scheduler/strategy/dynamicResources.go Normal file
View File

@ -0,0 +1,4 @@
package strategy
type DynamicResourcesStrategy struct {
}

2
pkg/scheduler/strategies/replication.go → pkg/scheduler/strategy/replication.go
View File

@ -1,4 +1,4 @@
package strategies package strategy
import ( import (
"github.com/pkg/errors" "github.com/pkg/errors"

2
pkg/scheduler/strategies/resourcePricing.go → pkg/scheduler/strategy/resourcePricing.go
View File

@ -12,7 +12,7 @@
*/ */
package strategies package strategy
import ( import (
"errors" "errors"

2
pkg/scheduler/strategies/staticWeight.go → pkg/scheduler/strategy/staticWeight.go
View File

@ -1,4 +1,4 @@
package strategies package strategy
type StaticWeightStrategy struct { type StaticWeightStrategy struct {
// TODO: add fields // TODO: add fields

2
pkg/scheduler/strategies/strategy.go → pkg/scheduler/strategy/strategy.go
View File

@ -1,4 +1,4 @@
package strategies package strategy
type Strategy interface { type Strategy interface {
Schedule() ([]*AssignedCluster, error) Schedule() ([]*AssignedCluster, error)