added GetAlgorithms for ai scheduler

Former-commit-id: 02fabcc30a84ad7b9afbde1a693a34ffacf47d4b
This commit is contained in:
tzwang 2024-03-26 16:34:46 +08:00
parent 6641c7452f
commit 08138fe02e
7 changed files with 110 additions and 0 deletions

View File

@ -876,6 +876,9 @@ service pcm {
@handler ScheduleGetStrategyHandler
get /schedule/ai/getStrategies returns (AiStrategyResp)
@handler ScheduleGetAlgorithmsHandler
get /schedule/ai/getAlgorithms (AiAlgorithmsReq) returns (AiAlgorithmsResp)
@handler ScheduleSubmitHandler
post /schedule/submit (ScheduleReq) returns (ScheduleResp)
}

View File

@ -41,4 +41,14 @@ type (
AiStrategyResp {
Strategies []string `json:"strategies"`
}
AiAlgorithmsReq {
ResourceType string `json:"resourceType"`
TaskType string `json:"taskType"`
Dataset string `json:"dataset"`
}
AiAlgorithmsResp {
Algorithms []string `json:"algorithms"`
}
)

View File

@ -3,6 +3,7 @@ package collector
type AiCollector interface {
GetResourceStats() (*ResourceStats, error)
GetDatasetsSpecs() ([]*DatasetsSpecs, error)
GetAlgorithms() ([]*Algorithm, error)
}
type ResourceStats struct {
@ -33,3 +34,9 @@ type DatasetsSpecs struct {
Name string
Size string
}
type Algorithm struct {
Name string
Platform string
TaskType string
}

View File

@ -157,6 +157,10 @@ func (m *ModelArtsLink) GetDatasetsSpecs() ([]*collector.DatasetsSpecs, error) {
return nil, nil
}
func (m *ModelArtsLink) GetAlgorithms() ([]*collector.Algorithm, error) {
return nil, nil
}
func (m *ModelArtsLink) Execute(option *option.AiOption) (interface{}, error) {
err := m.GenerateSubmitParams(option)
if err != nil {

View File

@ -315,6 +315,29 @@ func (o *OctopusLink) GetDatasetsSpecs() ([]*collector.DatasetsSpecs, error) {
return specs, nil
}
func (o *OctopusLink) GetAlgorithms() ([]*collector.Algorithm, error) {
var algorithms []*collector.Algorithm
req := &octopus.GetMyAlgorithmListReq{
Platform: o.platform,
PageIndex: o.pageIndex,
PageSize: o.pageSize,
}
resp, err := o.svcCtx.OctopusRpc.GetMyAlgorithmList(o.ctx, req)
if err != nil {
return nil, err
}
if !resp.Success {
return nil, errors.New("failed to get algorithms")
}
for _, a := range resp.Payload.Algorithms {
algorithm := &collector.Algorithm{Name: a.AlgorithmName, Platform: OCTOPUS, TaskType: strings.ToLower(a.FrameworkName)}
algorithms = append(algorithms, algorithm)
}
return algorithms, nil
}
func (o *OctopusLink) Execute(option *option.AiOption) (interface{}, error) {
err := o.GenerateSubmitParams(option)
if err != nil {

View File

@ -341,6 +341,26 @@ func (s *ShuguangAi) GetDatasetsSpecs() ([]*collector.DatasetsSpecs, error) {
return specs, nil
}
func (s *ShuguangAi) GetAlgorithms() ([]*collector.Algorithm, error) {
var algorithms []*collector.Algorithm
for _, t := range GetTaskTypes() {
taskType := t
req := &hpcAC.GetFileListReq{Limit: 100, Path: ALGORITHM_DIR + FORWARD_SLASH + taskType, Start: 0}
list, err := s.svcCtx.ACRpc.GetFileList(s.ctx, req)
if err != nil {
return nil, err
}
if list.Code != "0" {
return nil, errors.New(list.Msg)
}
for _, file := range list.Data.FileList {
algorithm := &collector.Algorithm{Name: file.Name, Platform: SHUGUANGAI, TaskType: taskType}
algorithms = append(algorithms, algorithm)
}
}
return algorithms, nil
}
func (s *ShuguangAi) Execute(option *option.AiOption) (interface{}, error) {
err := s.GenerateSubmitParams(option)
if err != nil {

View File

@ -27,6 +27,7 @@ import (
"gitlink.org.cn/jcce-pcm/pcm-participant-modelarts/modelarts"
"gitlink.org.cn/jcce-pcm/pcm-participant-octopus/octopus"
"gorm.io/gorm"
"strings"
)
type Linkage interface {
@ -152,6 +153,48 @@ func GetDatasetsNames(collectorMap *map[string]collector.AiCollector) ([]string,
return names, nil
}
func GetAlgorithms(collectorMap *map[string]collector.AiCollector, resourceType string, taskType string, dataset string) ([]string, error) {
var names []string
colMap := *collectorMap
for _, col := range colMap {
var ns []string
algorithms, err := col.GetAlgorithms()
if err != nil {
return nil, err
}
for _, algorithm := range algorithms {
if algorithm.TaskType != taskType {
continue
}
switch algorithm.Platform {
case OCTOPUS:
splitns := strings.Split(algorithm.Name, UNDERSCORE)
if dataset != splitns[0] || len(splitns) == 1 {
continue
}
ns = append(ns, splitns[1])
case SHUGUANGAI:
splitns := strings.Split(algorithm.Name, DASH)
if dataset != splitns[0] || len(splitns) == 1 {
continue
}
ns = append(ns, splitns[1])
}
}
if len(ns) == 0 {
continue
}
if len(names) == 0 {
names = ns
continue
}
names = common.IntersectString(names, ns)
}
names = common.RemoveDuplicates(names)
return names, nil
}
func GetTaskTypes() []string {
return taskTypes
}