forked from JointCloud/pcm-coordinator
Merge branch 'master' of https://gitlink.org.cn/JointCloud/pcm-coordinator
Former-commit-id: a47bea1ee2
This commit is contained in:
commit
b61d9b3a52
|
@ -166,6 +166,18 @@ type (
|
||||||
ReqBody []string `json:"reqBody"`
|
ReqBody []string `json:"reqBody"`
|
||||||
Replicas int64 `json:"replicas,string"`
|
Replicas int64 `json:"replicas,string"`
|
||||||
}
|
}
|
||||||
|
|
||||||
|
PodLogsReq {
|
||||||
|
TaskId string `json:"taskId"`
|
||||||
|
TaskName string `json:"taskName"`
|
||||||
|
ClusterId string `json:"clusterId"`
|
||||||
|
ClusterName string `json:"clusterName"`
|
||||||
|
AdapterId string `json:"adapterId"`
|
||||||
|
AdapterName string `json:"adapterName"`
|
||||||
|
PodName string `json:"podName"`
|
||||||
|
stream bool `json:"stream"`
|
||||||
|
|
||||||
|
}
|
||||||
)
|
)
|
||||||
|
|
||||||
type deleteTaskReq {
|
type deleteTaskReq {
|
||||||
|
|
|
@ -231,6 +231,9 @@ service pcm {
|
||||||
@doc "Create cloud computing common tasks"
|
@doc "Create cloud computing common tasks"
|
||||||
@handler commitGeneralTask
|
@handler commitGeneralTask
|
||||||
post /cloud/task/create (GeneralTaskReq) returns ()
|
post /cloud/task/create (GeneralTaskReq) returns ()
|
||||||
|
|
||||||
|
@handler podLogs
|
||||||
|
post /cloud/pod/logs (PodLogsReq) returns (string)
|
||||||
}
|
}
|
||||||
|
|
||||||
//智算二级接口
|
//智算二级接口
|
||||||
|
@ -968,10 +971,10 @@ service pcm {
|
||||||
post /schedule/getOverview returns (ScheduleOverviewResp)
|
post /schedule/getOverview returns (ScheduleOverviewResp)
|
||||||
|
|
||||||
@handler DownloadAlgothmCodeHandler
|
@handler DownloadAlgothmCodeHandler
|
||||||
get /schedule/getDownloadAlgothmCode (DownloadAlgorithmCodeReq) returns (DownloadAlgorithmCodeResp)
|
get /schedule/downloadAlgorithmCode (DownloadAlgorithmCodeReq) returns (DownloadAlgorithmCodeResp)
|
||||||
|
|
||||||
@handler UploadAlgothmCodeHandler
|
@handler UploadAlgothmCodeHandler
|
||||||
post /schedule/getDownloadAlgothmCode (UploadAlgorithmCodeReq) returns (UploadAlgorithmCodeResp)
|
post /schedule/uploadAlgorithmCode (UploadAlgorithmCodeReq) returns (UploadAlgorithmCodeResp)
|
||||||
|
|
||||||
@handler GetComputeCardsByClusterHandler
|
@handler GetComputeCardsByClusterHandler
|
||||||
get /schedule/getComputeCardsByCluster/:adapterId/:clusterId (GetComputeCardsByClusterReq) returns (GetComputeCardsByClusterResp)
|
get /schedule/getComputeCardsByCluster/:adapterId/:clusterId (GetComputeCardsByClusterReq) returns (GetComputeCardsByClusterResp)
|
||||||
|
|
|
@ -109,11 +109,10 @@ type (
|
||||||
TaskType string `form:"taskType"`
|
TaskType string `form:"taskType"`
|
||||||
Dataset string `form:"dataset"`
|
Dataset string `form:"dataset"`
|
||||||
Algorithm string `form:"algorithm"`
|
Algorithm string `form:"algorithm"`
|
||||||
Code string `form:"code"`
|
|
||||||
}
|
}
|
||||||
|
|
||||||
DownloadAlgorithmCodeResp {
|
DownloadAlgorithmCodeResp {
|
||||||
Code string `json:"algorithms"`
|
Code string `json:"code"`
|
||||||
}
|
}
|
||||||
|
|
||||||
UploadAlgorithmCodeReq {
|
UploadAlgorithmCodeReq {
|
||||||
|
|
|
@ -0,0 +1,24 @@
|
||||||
|
package cloud
|
||||||
|
|
||||||
|
import (
|
||||||
|
"github.com/zeromicro/go-zero/rest/httpx"
|
||||||
|
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/logic/cloud"
|
||||||
|
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/svc"
|
||||||
|
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/types"
|
||||||
|
"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/repository/result"
|
||||||
|
"net/http"
|
||||||
|
)
|
||||||
|
|
||||||
|
func PodLogsHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
|
||||||
|
return func(w http.ResponseWriter, r *http.Request) {
|
||||||
|
var req types.PodLogsReq
|
||||||
|
if err := httpx.Parse(r, &req); err != nil {
|
||||||
|
result.ParamErrorResult(r, w, err)
|
||||||
|
return
|
||||||
|
}
|
||||||
|
|
||||||
|
l := cloud.NewPodLogsLogic(r.Context(), svcCtx, w)
|
||||||
|
resp, err := l.PodLogs(&req, w)
|
||||||
|
result.HttpResult(r, w, resp, err)
|
||||||
|
}
|
||||||
|
}
|
|
@ -277,6 +277,11 @@ func RegisterHandlers(server *rest.Server, serverCtx *svc.ServiceContext) {
|
||||||
Path: "/cloud/task/create",
|
Path: "/cloud/task/create",
|
||||||
Handler: cloud.CommitGeneralTaskHandler(serverCtx),
|
Handler: cloud.CommitGeneralTaskHandler(serverCtx),
|
||||||
},
|
},
|
||||||
|
{
|
||||||
|
Method: http.MethodPost,
|
||||||
|
Path: "/cloud/pod/logs",
|
||||||
|
Handler: cloud.PodLogsHandler(serverCtx),
|
||||||
|
},
|
||||||
},
|
},
|
||||||
rest.WithPrefix("/pcm/v1"),
|
rest.WithPrefix("/pcm/v1"),
|
||||||
)
|
)
|
||||||
|
@ -1207,12 +1212,12 @@ func RegisterHandlers(server *rest.Server, serverCtx *svc.ServiceContext) {
|
||||||
},
|
},
|
||||||
{
|
{
|
||||||
Method: http.MethodGet,
|
Method: http.MethodGet,
|
||||||
Path: "/schedule/getDownloadAlgothmCode",
|
Path: "/schedule/downloadAlgorithmCode",
|
||||||
Handler: schedule.DownloadAlgothmCodeHandler(serverCtx),
|
Handler: schedule.DownloadAlgothmCodeHandler(serverCtx),
|
||||||
},
|
},
|
||||||
{
|
{
|
||||||
Method: http.MethodPost,
|
Method: http.MethodPost,
|
||||||
Path: "/schedule/getDownloadAlgothmCode",
|
Path: "/schedule/uploadAlgorithmCode",
|
||||||
Handler: schedule.UploadAlgothmCodeHandler(serverCtx),
|
Handler: schedule.UploadAlgothmCodeHandler(serverCtx),
|
||||||
},
|
},
|
||||||
{
|
{
|
||||||
|
|
|
@ -7,22 +7,19 @@ import (
|
||||||
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/logic/schedule"
|
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/logic/schedule"
|
||||||
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/svc"
|
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/svc"
|
||||||
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/types"
|
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/types"
|
||||||
|
"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/repository/result"
|
||||||
)
|
)
|
||||||
|
|
||||||
func DownloadAlgothmCodeHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
|
func DownloadAlgothmCodeHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
|
||||||
return func(w http.ResponseWriter, r *http.Request) {
|
return func(w http.ResponseWriter, r *http.Request) {
|
||||||
var req types.DownloadAlgorithmCodeReq
|
var req types.DownloadAlgorithmCodeReq
|
||||||
if err := httpx.Parse(r, &req); err != nil {
|
if err := httpx.Parse(r, &req); err != nil {
|
||||||
httpx.ErrorCtx(r.Context(), w, err)
|
result.ParamErrorResult(r, w, err)
|
||||||
return
|
return
|
||||||
}
|
}
|
||||||
|
|
||||||
l := schedule.NewDownloadAlgothmCodeLogic(r.Context(), svcCtx)
|
l := schedule.NewDownloadAlgothmCodeLogic(r.Context(), svcCtx)
|
||||||
resp, err := l.DownloadAlgothmCode(&req)
|
resp, err := l.DownloadAlgorithmCode(&req)
|
||||||
if err != nil {
|
result.HttpResult(r, w, resp, err)
|
||||||
httpx.ErrorCtx(r.Context(), w, err)
|
|
||||||
} else {
|
|
||||||
httpx.OkJsonCtx(r.Context(), w, resp)
|
|
||||||
}
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
|
@ -7,22 +7,19 @@ import (
|
||||||
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/logic/schedule"
|
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/logic/schedule"
|
||||||
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/svc"
|
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/svc"
|
||||||
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/types"
|
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/types"
|
||||||
|
"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/repository/result"
|
||||||
)
|
)
|
||||||
|
|
||||||
func UploadAlgothmCodeHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
|
func UploadAlgothmCodeHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
|
||||||
return func(w http.ResponseWriter, r *http.Request) {
|
return func(w http.ResponseWriter, r *http.Request) {
|
||||||
var req types.UploadAlgorithmCodeReq
|
var req types.UploadAlgorithmCodeReq
|
||||||
if err := httpx.Parse(r, &req); err != nil {
|
if err := httpx.Parse(r, &req); err != nil {
|
||||||
httpx.ErrorCtx(r.Context(), w, err)
|
result.ParamErrorResult(r, w, err)
|
||||||
return
|
return
|
||||||
}
|
}
|
||||||
|
|
||||||
l := schedule.NewUploadAlgothmCodeLogic(r.Context(), svcCtx)
|
l := schedule.NewUploadAlgothmCodeLogic(r.Context(), svcCtx)
|
||||||
resp, err := l.UploadAlgothmCode(&req)
|
resp, err := l.UploadAlgorithmCode(&req)
|
||||||
if err != nil {
|
result.HttpResult(r, w, resp, err)
|
||||||
httpx.ErrorCtx(r.Context(), w, err)
|
|
||||||
} else {
|
|
||||||
httpx.OkJsonCtx(r.Context(), w, resp)
|
|
||||||
}
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
|
@ -99,6 +99,8 @@ func (l *CommitGeneralTaskLogic) CommitGeneralTask(req *types.GeneralTaskReq) er
|
||||||
Strategy: strategy,
|
Strategy: strategy,
|
||||||
}
|
}
|
||||||
var taskClouds []cloud.TaskCloudModel
|
var taskClouds []cloud.TaskCloudModel
|
||||||
|
adapterName := ""
|
||||||
|
tx.Table("t_adapter").Select("name").Where("id=?", adapterId).Find(&adapterName)
|
||||||
for _, r := range rs {
|
for _, r := range rs {
|
||||||
for _, s := range req.ReqBody {
|
for _, s := range req.ReqBody {
|
||||||
sStruct := UnMarshalK8sStruct(s, int64(r.Replica))
|
sStruct := UnMarshalK8sStruct(s, int64(r.Replica))
|
||||||
|
@ -107,6 +109,7 @@ func (l *CommitGeneralTaskLogic) CommitGeneralTask(req *types.GeneralTaskReq) er
|
||||||
taskCloud.TaskId = uint(taskModel.Id)
|
taskCloud.TaskId = uint(taskModel.Id)
|
||||||
clusterId, _ := strconv.ParseUint(r.ClusterId, 10, 64)
|
clusterId, _ := strconv.ParseUint(r.ClusterId, 10, 64)
|
||||||
taskCloud.AdapterId = uint(adapterId)
|
taskCloud.AdapterId = uint(adapterId)
|
||||||
|
taskCloud.AdapterName = adapterName
|
||||||
taskCloud.ClusterId = uint(clusterId)
|
taskCloud.ClusterId = uint(clusterId)
|
||||||
taskCloud.ClusterName = r.ClusterName
|
taskCloud.ClusterName = r.ClusterName
|
||||||
taskCloud.Status = constants.Saved
|
taskCloud.Status = constants.Saved
|
||||||
|
@ -116,8 +119,6 @@ func (l *CommitGeneralTaskLogic) CommitGeneralTask(req *types.GeneralTaskReq) er
|
||||||
taskClouds = append(taskClouds, taskCloud)
|
taskClouds = append(taskClouds, taskCloud)
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
adapterName := ""
|
|
||||||
tx.Table("t_adapter").Select("name").Where("id=?", adapterId).Find(&adapterName)
|
|
||||||
noticeInfo := clientCore.NoticeInfo{
|
noticeInfo := clientCore.NoticeInfo{
|
||||||
AdapterId: int64(adapterId),
|
AdapterId: int64(adapterId),
|
||||||
AdapterName: adapterName,
|
AdapterName: adapterName,
|
||||||
|
|
|
@ -0,0 +1,30 @@
|
||||||
|
package cloud
|
||||||
|
|
||||||
|
import (
|
||||||
|
"context"
|
||||||
|
"github.com/zeromicro/go-zero/core/logx"
|
||||||
|
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/svc"
|
||||||
|
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/types"
|
||||||
|
"net/http"
|
||||||
|
)
|
||||||
|
|
||||||
|
type PodLogsLogic struct {
|
||||||
|
logx.Logger
|
||||||
|
ctx context.Context
|
||||||
|
svcCtx *svc.ServiceContext
|
||||||
|
w http.ResponseWriter
|
||||||
|
}
|
||||||
|
|
||||||
|
func NewPodLogsLogic(ctx context.Context, svcCtx *svc.ServiceContext, w http.ResponseWriter) *PodLogsLogic {
|
||||||
|
return &PodLogsLogic{
|
||||||
|
Logger: logx.WithContext(ctx),
|
||||||
|
ctx: ctx,
|
||||||
|
svcCtx: svcCtx,
|
||||||
|
w: w,
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
func (l *PodLogsLogic) PodLogs(req *types.PodLogsReq, w http.ResponseWriter) (resp string, err error) {
|
||||||
|
// todo: add your logic here and delete this line
|
||||||
|
return
|
||||||
|
}
|
|
@ -2,13 +2,12 @@ package core
|
||||||
|
|
||||||
import (
|
import (
|
||||||
"context"
|
"context"
|
||||||
|
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/svc"
|
||||||
|
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/types"
|
||||||
"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/repository/result"
|
"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/repository/result"
|
||||||
"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/utils/timeutils"
|
"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/utils/timeutils"
|
||||||
"time"
|
"time"
|
||||||
|
|
||||||
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/svc"
|
|
||||||
"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/types"
|
|
||||||
|
|
||||||
"github.com/zeromicro/go-zero/core/logx"
|
"github.com/zeromicro/go-zero/core/logx"
|
||||||
)
|
)
|
||||||
|
|
||||||
|
@ -51,15 +50,15 @@ func (l *PageListTaskLogic) PageListTask(req *types.PageTaskReq) (resp *types.Pa
|
||||||
return nil, result.NewDefaultError(err.Error())
|
return nil, result.NewDefaultError(err.Error())
|
||||||
}
|
}
|
||||||
for _, model := range list {
|
for _, model := range list {
|
||||||
if model.EndTime != "" && model.StartTime != "" {
|
if model.StartTime != "" && model.EndTime == "" {
|
||||||
|
startTime := timeutils.TimeStringToGoTime(model.StartTime)
|
||||||
|
model.RunningTime = int64(time.Now().Sub(startTime).Seconds())
|
||||||
|
}
|
||||||
|
if model.StartTime != "" && model.EndTime != "" {
|
||||||
startTime := timeutils.TimeStringToGoTime(model.StartTime)
|
startTime := timeutils.TimeStringToGoTime(model.StartTime)
|
||||||
endTime := timeutils.TimeStringToGoTime(model.EndTime)
|
endTime := timeutils.TimeStringToGoTime(model.EndTime)
|
||||||
model.RunningTime = int64(endTime.Sub(startTime).Seconds())
|
model.RunningTime = int64(endTime.Sub(startTime).Seconds())
|
||||||
}
|
}
|
||||||
if model.StartTime != "" {
|
|
||||||
startTime := timeutils.TimeStringToGoTime(model.StartTime)
|
|
||||||
model.RunningTime = int64(time.Now().Sub(startTime).Seconds())
|
|
||||||
}
|
|
||||||
}
|
}
|
||||||
resp.List = &list
|
resp.List = &list
|
||||||
resp.PageSize = req.PageSize
|
resp.PageSize = req.PageSize
|
||||||
|
|
|
@ -17,6 +17,7 @@ package core
|
||||||
import (
|
import (
|
||||||
"context"
|
"context"
|
||||||
"fmt"
|
"fmt"
|
||||||
|
"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/constants"
|
||||||
"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/models"
|
"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/models"
|
||||||
"strconv"
|
"strconv"
|
||||||
"time"
|
"time"
|
||||||
|
@ -55,6 +56,11 @@ func (l *TaskListLogic) TaskList(req *types.TaskListReq) (resp *types.TaskListRe
|
||||||
if len(tasks) == 0 {
|
if len(tasks) == 0 {
|
||||||
return nil, nil
|
return nil, nil
|
||||||
}
|
}
|
||||||
|
|
||||||
|
// 更新智算任务状态
|
||||||
|
var ch = make(chan struct{})
|
||||||
|
go l.updateAitaskStatus(tasks, ch)
|
||||||
|
|
||||||
// 查询任务总数
|
// 查询任务总数
|
||||||
l.svcCtx.DbEngin.Model(&models.Task{}).Count(&resp.TotalCount)
|
l.svcCtx.DbEngin.Model(&models.Task{}).Count(&resp.TotalCount)
|
||||||
|
|
||||||
|
@ -106,5 +112,64 @@ func (l *TaskListLogic) TaskList(req *types.TaskListReq) (resp *types.TaskListRe
|
||||||
|
|
||||||
}
|
}
|
||||||
|
|
||||||
return
|
select {
|
||||||
|
case _ = <-ch:
|
||||||
|
return resp, nil
|
||||||
|
case <-time.After(1 * time.Second):
|
||||||
|
return resp, nil
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
func (l *TaskListLogic) updateAitaskStatus(tasks []models.Task, ch chan<- struct{}) {
|
||||||
|
for _, task := range tasks {
|
||||||
|
if task.AdapterTypeDict != 1 {
|
||||||
|
continue
|
||||||
|
}
|
||||||
|
if task.Status == constants.Succeeded {
|
||||||
|
continue
|
||||||
|
}
|
||||||
|
|
||||||
|
var aiTask []*models.TaskAi
|
||||||
|
tx := l.svcCtx.DbEngin.Raw("select * from task_ai where `task_id` = ? ", task.Id).Scan(&aiTask)
|
||||||
|
if tx.Error != nil {
|
||||||
|
logx.Errorf(tx.Error.Error())
|
||||||
|
return
|
||||||
|
}
|
||||||
|
|
||||||
|
start, _ := time.ParseInLocation(constants.Layout, aiTask[0].StartTime, time.Local)
|
||||||
|
end, _ := time.ParseInLocation(constants.Layout, aiTask[0].EndTime, time.Local)
|
||||||
|
var status = constants.Succeeded
|
||||||
|
for _, a := range aiTask {
|
||||||
|
s, _ := time.ParseInLocation(constants.Layout, a.StartTime, time.Local)
|
||||||
|
e, _ := time.ParseInLocation(constants.Layout, a.EndTime, time.Local)
|
||||||
|
|
||||||
|
if s.Before(start) {
|
||||||
|
start = s
|
||||||
|
}
|
||||||
|
|
||||||
|
if e.After(end) {
|
||||||
|
end = e
|
||||||
|
}
|
||||||
|
|
||||||
|
if a.Status == constants.Failed {
|
||||||
|
status = a.Status
|
||||||
|
break
|
||||||
|
}
|
||||||
|
|
||||||
|
if a.Status == constants.Running {
|
||||||
|
status = a.Status
|
||||||
|
continue
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
task.Status = status
|
||||||
|
task.StartTime = &start
|
||||||
|
task.EndTime = &end
|
||||||
|
|
||||||
|
tx = l.svcCtx.DbEngin.Updates(task)
|
||||||
|
if tx.Error != nil {
|
||||||
|
return
|
||||||
|
}
|
||||||
|
}
|
||||||
|
ch <- struct{}{}
|
||||||
}
|
}
|
||||||
|
|
|
@ -23,8 +23,14 @@ func NewDownloadAlgothmCodeLogic(ctx context.Context, svcCtx *svc.ServiceContext
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
func (l *DownloadAlgothmCodeLogic) DownloadAlgothmCode(req *types.DownloadAlgorithmCodeReq) (resp *types.DownloadAlgorithmCodeResp, err error) {
|
func (l *DownloadAlgothmCodeLogic) DownloadAlgorithmCode(req *types.DownloadAlgorithmCodeReq) (resp *types.DownloadAlgorithmCodeResp, err error) {
|
||||||
// todo: add your logic here and delete this line
|
resp = &types.DownloadAlgorithmCodeResp{}
|
||||||
|
code, err := l.svcCtx.Scheduler.AiService.AiCollectorAdapterMap[req.AdapterId][req.ClusterId].DownloadAlgorithmCode(l.ctx,
|
||||||
|
req.ResourceType, req.Card, req.TaskType, req.Dataset, req.Algorithm)
|
||||||
|
if err != nil {
|
||||||
|
return nil, err
|
||||||
|
}
|
||||||
|
resp.Code = code
|
||||||
|
|
||||||
return
|
return resp, nil
|
||||||
}
|
}
|
||||||
|
|
|
@ -23,8 +23,13 @@ func NewUploadAlgothmCodeLogic(ctx context.Context, svcCtx *svc.ServiceContext)
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
func (l *UploadAlgothmCodeLogic) UploadAlgothmCode(req *types.UploadAlgorithmCodeReq) (resp *types.UploadAlgorithmCodeResp, err error) {
|
func (l *UploadAlgothmCodeLogic) UploadAlgorithmCode(req *types.UploadAlgorithmCodeReq) (resp *types.UploadAlgorithmCodeResp, err error) {
|
||||||
// todo: add your logic here and delete this line
|
resp = &types.UploadAlgorithmCodeResp{}
|
||||||
|
err = l.svcCtx.Scheduler.AiService.AiCollectorAdapterMap[req.AdapterId][req.ClusterId].UploadAlgorithmCode(l.ctx,
|
||||||
|
req.ResourceType, req.Card, req.TaskType, req.Dataset, req.Algorithm, req.Code)
|
||||||
|
if err != nil {
|
||||||
|
return nil, err
|
||||||
|
}
|
||||||
|
|
||||||
return
|
return resp, nil
|
||||||
}
|
}
|
||||||
|
|
|
@ -54,6 +54,7 @@ const (
|
||||||
CAMBRICON = "cambricon"
|
CAMBRICON = "cambricon"
|
||||||
TRAIN_CMD = "cd /code; python train.py"
|
TRAIN_CMD = "cd /code; python train.py"
|
||||||
VERSION = "V1"
|
VERSION = "V1"
|
||||||
|
DOMAIN = "http://192.168.242.41:8001/"
|
||||||
)
|
)
|
||||||
|
|
||||||
var (
|
var (
|
||||||
|
@ -340,7 +341,44 @@ func (o *OctopusLink) GetAlgorithms(ctx context.Context) ([]*collector.Algorithm
|
||||||
}
|
}
|
||||||
|
|
||||||
func (o *OctopusLink) DownloadAlgorithmCode(ctx context.Context, resourceType string, card string, taskType string, dataset string, algorithm string) (string, error) {
|
func (o *OctopusLink) DownloadAlgorithmCode(ctx context.Context, resourceType string, card string, taskType string, dataset string, algorithm string) (string, error) {
|
||||||
return "", nil
|
dcReq := &octopus.DownloadCompressReq{
|
||||||
|
Platform: o.platform,
|
||||||
|
Version: VERSION,
|
||||||
|
AlgorithmId: "",
|
||||||
|
}
|
||||||
|
dcResp, err := o.octopusRpc.DownloadCompress(ctx, dcReq)
|
||||||
|
if err != nil {
|
||||||
|
return "", err
|
||||||
|
}
|
||||||
|
|
||||||
|
if !dcResp.Success {
|
||||||
|
return "", errors.New(dcResp.Error.Message)
|
||||||
|
}
|
||||||
|
|
||||||
|
daReq := &octopus.DownloadAlgorithmReq{
|
||||||
|
Platform: o.platform,
|
||||||
|
Version: VERSION,
|
||||||
|
AlgorithmId: "",
|
||||||
|
CompressAt: dcResp.Payload.CompressAt,
|
||||||
|
Domain: DOMAIN,
|
||||||
|
}
|
||||||
|
daResp, err := o.octopusRpc.DownloadAlgorithm(ctx, daReq)
|
||||||
|
if err != nil {
|
||||||
|
return "", err
|
||||||
|
}
|
||||||
|
if !daResp.Success {
|
||||||
|
return "", errors.New(dcResp.Error.Message)
|
||||||
|
}
|
||||||
|
urlReq := &octopus.AlgorithmUrlReq{
|
||||||
|
Platform: o.platform,
|
||||||
|
Url: daResp.Payload.DownloadUrl,
|
||||||
|
}
|
||||||
|
urlResp, err := o.octopusRpc.DownloadAlgorithmUrl(ctx, urlReq)
|
||||||
|
if err != nil {
|
||||||
|
return "", err
|
||||||
|
}
|
||||||
|
|
||||||
|
return urlResp.Algorithm, nil
|
||||||
}
|
}
|
||||||
|
|
||||||
func (o *OctopusLink) UploadAlgorithmCode(ctx context.Context, resourceType string, card string, taskType string, dataset string, algorithm string, code string) error {
|
func (o *OctopusLink) UploadAlgorithmCode(ctx context.Context, resourceType string, card string, taskType string, dataset string, algorithm string, code string) error {
|
||||||
|
|
|
@ -150,6 +150,17 @@ type GeneralTaskReq struct {
|
||||||
Replicas int64 `json:"replicas,string"`
|
Replicas int64 `json:"replicas,string"`
|
||||||
}
|
}
|
||||||
|
|
||||||
|
type PodLogsReq struct {
|
||||||
|
TaskId string `json:"taskId"`
|
||||||
|
TaskName string `json:"taskName"`
|
||||||
|
ClusterId string `json:"clusterId"`
|
||||||
|
ClusterName string `json:"clusterName"`
|
||||||
|
AdapterId string `json:"adapterId"`
|
||||||
|
AdapterName string `json:"adapterName"`
|
||||||
|
PodName string `json:"podName"`
|
||||||
|
Stream bool `json:"stream"`
|
||||||
|
}
|
||||||
|
|
||||||
type DeleteTaskReq struct {
|
type DeleteTaskReq struct {
|
||||||
Id int64 `path:"id"`
|
Id int64 `path:"id"`
|
||||||
}
|
}
|
||||||
|
@ -938,6 +949,184 @@ type ListResult struct {
|
||||||
List interface{} `json:"list,omitempty"`
|
List interface{} `json:"list,omitempty"`
|
||||||
}
|
}
|
||||||
|
|
||||||
|
type HpcInfo struct {
|
||||||
|
Id int64 `json:"id"` // id
|
||||||
|
TaskId int64 `json:"task_id"` // 任务id
|
||||||
|
JobId string `json:"job_id"` // 作业id(在第三方系统中的作业id)
|
||||||
|
AdapterId int64 `json:"adapter_id"` // 执行任务的适配器id
|
||||||
|
ClusterId int64 `json:"cluster_id"` // 执行任务的集群id
|
||||||
|
ClusterType string `json:"cluster_type"` // 执行任务的集群类型
|
||||||
|
Name string `json:"name"` // 名称
|
||||||
|
Status string `json:"status"` // 状态
|
||||||
|
CmdScript string `json:"cmd_script"`
|
||||||
|
StartTime string `json:"start_time"` // 开始时间
|
||||||
|
RunningTime int64 `json:"running_time"` // 运行时间
|
||||||
|
DerivedEs string `json:"derived_es"`
|
||||||
|
Cluster string `json:"cluster"`
|
||||||
|
BlockId int64 `json:"block_id"`
|
||||||
|
AllocNodes int64 `json:"alloc_nodes"`
|
||||||
|
AllocCpu int64 `json:"alloc_cpu"`
|
||||||
|
CardCount int64 `json:"card_count"` // 卡数
|
||||||
|
Version string `json:"version"`
|
||||||
|
Account string `json:"account"`
|
||||||
|
WorkDir string `json:"work_dir"` // 工作路径
|
||||||
|
AssocId int64 `json:"assoc_id"`
|
||||||
|
ExitCode int64 `json:"exit_code"`
|
||||||
|
WallTime string `json:"wall_time"` // 最大运行时间
|
||||||
|
Result string `json:"result"` // 运行结果
|
||||||
|
DeletedAt string `json:"deleted_at"` // 删除时间
|
||||||
|
YamlString string `json:"yaml_string"`
|
||||||
|
AppType string `json:"app_type"` // 应用类型
|
||||||
|
AppName string `json:"app_name"` // 应用名称
|
||||||
|
Queue string `json:"queue"` // 队列名称
|
||||||
|
SubmitType string `json:"submit_type"` // cmd(命令行模式)
|
||||||
|
NNode string `json:"n_node"` // 节点个数(当指定该参数时,GAP_NODE_STRING必须为"")
|
||||||
|
StdOutFile string `json:"std_out_file"` // 工作路径/std.err.%j
|
||||||
|
StdErrFile string `json:"std_err_file"` // 工作路径/std.err.%j
|
||||||
|
StdInput string `json:"std_input"`
|
||||||
|
Environment string `json:"environment"`
|
||||||
|
DeletedFlag int64 `json:"deleted_flag"` // 是否删除(0-否,1-是)
|
||||||
|
CreatedBy int64 `json:"created_by"` // 创建人
|
||||||
|
CreateTime string `json:"created_time"` // 创建时间
|
||||||
|
UpdatedBy int64 `json:"updated_by"` // 更新人
|
||||||
|
UpdateTime string `json:"updated_time"` // 更新时间
|
||||||
|
}
|
||||||
|
|
||||||
|
type CloudInfo struct {
|
||||||
|
Participant int64 `json:"participant,omitempty"`
|
||||||
|
Id int64 `json:"id,omitempty"`
|
||||||
|
TaskId int64 `json:"taskId,omitempty"`
|
||||||
|
ApiVersion string `json:"apiVersion,omitempty"`
|
||||||
|
Kind string `json:"kind,omitempty"`
|
||||||
|
Namespace string `json:"namespace,omitempty"`
|
||||||
|
Name string `json:"name,omitempty"`
|
||||||
|
Status string `json:"status,omitempty"`
|
||||||
|
StartTime string `json:"startTime,omitempty"`
|
||||||
|
RunningTime int64 `json:"runningTime,omitempty"`
|
||||||
|
Result string `json:"result,omitempty"`
|
||||||
|
YamlString string `json:"yamlString,omitempty"`
|
||||||
|
}
|
||||||
|
|
||||||
|
type AiInfo struct {
|
||||||
|
ParticipantId int64 `json:"participantId,omitempty"`
|
||||||
|
TaskId int64 `json:"taskId,omitempty"`
|
||||||
|
ProjectId string `json:"project_id,omitempty"`
|
||||||
|
Name string `json:"name,omitempty"`
|
||||||
|
Status string `json:"status,omitempty"`
|
||||||
|
StartTime string `json:"startTime,omitempty"`
|
||||||
|
RunningTime int64 `json:"runningTime,omitempty"`
|
||||||
|
Result string `json:"result,omitempty"`
|
||||||
|
JobId string `json:"jobId,omitempty"`
|
||||||
|
CreateTime string `json:"createTime,omitempty"`
|
||||||
|
ImageUrl string `json:"imageUrl,omitempty"`
|
||||||
|
Command string `json:"command,omitempty"`
|
||||||
|
FlavorId string `json:"flavorId,omitempty"`
|
||||||
|
SubscriptionId string `json:"subscriptionId,omitempty"`
|
||||||
|
ItemVersionId string `json:"itemVersionId,omitempty"`
|
||||||
|
}
|
||||||
|
|
||||||
|
type VmInfo struct {
|
||||||
|
ParticipantId int64 `json:"participantId,omitempty"`
|
||||||
|
TaskId int64 `json:"taskId,omitempty"`
|
||||||
|
Name string `json:"name,omitempty"`
|
||||||
|
FlavorRef string `json:"flavor_ref,omitempty"`
|
||||||
|
ImageRef string `json:"image_ref,omitempty"`
|
||||||
|
NetworkUuid string `json:"network_uuid,omitempty"`
|
||||||
|
BlockUuid string `json:"block_uuid,omitempty"`
|
||||||
|
SourceType string `json:"source_type,omitempty"`
|
||||||
|
DeleteOnTermination bool `json:"delete_on_termination,omitempty"`
|
||||||
|
Status string `json:"status,omitempty"`
|
||||||
|
MinCount string `json:"min_count,omitempty"`
|
||||||
|
Platform string `json:"platform,omitempty"`
|
||||||
|
Uuid string `json:"uuid,omitempty"`
|
||||||
|
}
|
||||||
|
|
||||||
|
type PullTaskInfoReq struct {
|
||||||
|
AdapterId int64 `form:"adapterId"`
|
||||||
|
}
|
||||||
|
|
||||||
|
type PullTaskInfoResp struct {
|
||||||
|
HpcInfoList []*HpcInfo `json:"HpcInfoList,omitempty"`
|
||||||
|
CloudInfoList []*CloudInfo `json:"CloudInfoList,omitempty"`
|
||||||
|
AiInfoList []*AiInfo `json:"AiInfoList,omitempty"`
|
||||||
|
VmInfoList []*VmInfo `json:"VmInfoList,omitempty"`
|
||||||
|
}
|
||||||
|
|
||||||
|
type PushTaskInfoReq struct {
|
||||||
|
AdapterId int64 `json:"adapterId"`
|
||||||
|
HpcInfoList []*HpcInfo `json:"hpcInfoList"`
|
||||||
|
CloudInfoList []*CloudInfo `json:"cloudInfoList"`
|
||||||
|
AiInfoList []*AiInfo `json:"aiInfoList"`
|
||||||
|
VmInfoList []*VmInfo `json:"vmInfoList"`
|
||||||
|
}
|
||||||
|
|
||||||
|
type PushTaskInfoResp struct {
|
||||||
|
Code int64 `json:"code"`
|
||||||
|
Msg string `json:"msg"`
|
||||||
|
}
|
||||||
|
|
||||||
|
type PushResourceInfoReq struct {
|
||||||
|
AdapterId int64 `json:"adapterId"`
|
||||||
|
ResourceStats []ResourceStats `json:"resourceStats"`
|
||||||
|
}
|
||||||
|
|
||||||
|
type PushResourceInfoResp struct {
|
||||||
|
Code int64 `json:"code"`
|
||||||
|
Msg string `json:"msg"`
|
||||||
|
}
|
||||||
|
|
||||||
|
type NoticeInfo struct {
|
||||||
|
AdapterId int64 `json:"adapterId"`
|
||||||
|
AdapterName string `json:"adapterName"`
|
||||||
|
ClusterId int64 `json:"clusterId"`
|
||||||
|
ClusterName string `json:"clusterName"`
|
||||||
|
NoticeType string `json:"noticeType"`
|
||||||
|
TaskName string `json:"taskName"`
|
||||||
|
Incident string `json:"incident"`
|
||||||
|
}
|
||||||
|
|
||||||
|
type ListNoticeReq struct {
|
||||||
|
}
|
||||||
|
|
||||||
|
type ListNoticeResp struct {
|
||||||
|
Code int64 `json:"code"`
|
||||||
|
Msg string `json:"msg"`
|
||||||
|
Data []NoticeInfo `json:"data"`
|
||||||
|
}
|
||||||
|
|
||||||
|
type PushNoticeReq struct {
|
||||||
|
NoticeInfo NoticeInfo `json:"noticeInfo"`
|
||||||
|
}
|
||||||
|
|
||||||
|
type PushNoticeResp struct {
|
||||||
|
Code int64 `json:"code"`
|
||||||
|
Msg string `json:"msg"`
|
||||||
|
}
|
||||||
|
|
||||||
|
type ResourceStats struct {
|
||||||
|
ClusterId int64 `json:"clusterId"`
|
||||||
|
Name string `json:"name"`
|
||||||
|
CpuCoreAvail int64 `json:"cpuCoreAvail"`
|
||||||
|
CpuCoreTotal int64 `json:"cpuCoreTotal"`
|
||||||
|
MemAvail float64 `json:"memAvail"`
|
||||||
|
MemTotal float64 `json:"memTotal"`
|
||||||
|
DiskAvail float64 `json:"diskAvail"`
|
||||||
|
DiskTotal float64 `json:"diskTotal"`
|
||||||
|
GpuAvail int64 `json:"gpuAvail"`
|
||||||
|
CardsAvail []*Card `json:"cardsAvail"`
|
||||||
|
CpuCoreHours float64 `json:"cpuCoreHours"`
|
||||||
|
Balance float64 `json:"balance"`
|
||||||
|
}
|
||||||
|
|
||||||
|
type Card struct {
|
||||||
|
Platform string `json:"platform"`
|
||||||
|
Type string `json:"type"`
|
||||||
|
Name string `json:"name"`
|
||||||
|
TOpsAtFp16 float64 `json:"TOpsAtFp16"`
|
||||||
|
CardHours float64 `json:"cardHours"`
|
||||||
|
CardNum int32 `json:"cardNum"`
|
||||||
|
}
|
||||||
|
|
||||||
type TaskStatusResp struct {
|
type TaskStatusResp struct {
|
||||||
Succeeded int `json:"Succeeded"`
|
Succeeded int `json:"Succeeded"`
|
||||||
Failed int `json:"Failed"`
|
Failed int `json:"Failed"`
|
||||||
|
@ -5506,11 +5695,10 @@ type DownloadAlgorithmCodeReq struct {
|
||||||
TaskType string `form:"taskType"`
|
TaskType string `form:"taskType"`
|
||||||
Dataset string `form:"dataset"`
|
Dataset string `form:"dataset"`
|
||||||
Algorithm string `form:"algorithm"`
|
Algorithm string `form:"algorithm"`
|
||||||
Code string `form:"code"`
|
|
||||||
}
|
}
|
||||||
|
|
||||||
type DownloadAlgorithmCodeResp struct {
|
type DownloadAlgorithmCodeResp struct {
|
||||||
Code string `json:"algorithms"`
|
Code string `json:"code"`
|
||||||
}
|
}
|
||||||
|
|
||||||
type UploadAlgorithmCodeReq struct {
|
type UploadAlgorithmCodeReq struct {
|
||||||
|
|
2
go.mod
2
go.mod
|
@ -26,7 +26,7 @@ require (
|
||||||
github.com/zeromicro/go-zero v1.6.3
|
github.com/zeromicro/go-zero v1.6.3
|
||||||
gitlink.org.cn/JointCloud/pcm-ac v0.0.0-20240426095603-549fefd8bece
|
gitlink.org.cn/JointCloud/pcm-ac v0.0.0-20240426095603-549fefd8bece
|
||||||
gitlink.org.cn/JointCloud/pcm-kubernetes v0.0.0-20240301071143-347480abff2c
|
gitlink.org.cn/JointCloud/pcm-kubernetes v0.0.0-20240301071143-347480abff2c
|
||||||
gitlink.org.cn/JointCloud/pcm-octopus v0.0.0-20240424085753-6899615e9142
|
gitlink.org.cn/JointCloud/pcm-octopus v0.0.0-20240510133934-6a5526289b35
|
||||||
gitlink.org.cn/JointCloud/pcm-openstack v0.0.0-20240403033338-e7edabad4203
|
gitlink.org.cn/JointCloud/pcm-openstack v0.0.0-20240403033338-e7edabad4203
|
||||||
gitlink.org.cn/JointCloud/pcm-slurm v0.0.0-20240301080743-8b94bbaf57f5
|
gitlink.org.cn/JointCloud/pcm-slurm v0.0.0-20240301080743-8b94bbaf57f5
|
||||||
gitlink.org.cn/jcce-pcm/pcm-participant-ceph v0.0.0-20230904090036-24fc730ec87d
|
gitlink.org.cn/jcce-pcm/pcm-participant-ceph v0.0.0-20230904090036-24fc730ec87d
|
||||||
|
|
4
go.sum
4
go.sum
|
@ -1082,8 +1082,8 @@ gitlink.org.cn/JointCloud/pcm-ac v0.0.0-20240426095603-549fefd8bece h1:W3yBnvAVV
|
||||||
gitlink.org.cn/JointCloud/pcm-ac v0.0.0-20240426095603-549fefd8bece/go.mod h1:w3Nb5TNymCItQ7K3x4Q0JLuoq9OerwAzAWT2zsPE9Xo=
|
gitlink.org.cn/JointCloud/pcm-ac v0.0.0-20240426095603-549fefd8bece/go.mod h1:w3Nb5TNymCItQ7K3x4Q0JLuoq9OerwAzAWT2zsPE9Xo=
|
||||||
gitlink.org.cn/JointCloud/pcm-kubernetes v0.0.0-20240301071143-347480abff2c h1:2Wl/hvaSFjh6fmCSIQhjkr9llMRREQeqcXNLZ/HPY18=
|
gitlink.org.cn/JointCloud/pcm-kubernetes v0.0.0-20240301071143-347480abff2c h1:2Wl/hvaSFjh6fmCSIQhjkr9llMRREQeqcXNLZ/HPY18=
|
||||||
gitlink.org.cn/JointCloud/pcm-kubernetes v0.0.0-20240301071143-347480abff2c/go.mod h1:lSRfGs+PxFvw7CcndHWRd6UlLlGrZn0b0hp5cfaMNGw=
|
gitlink.org.cn/JointCloud/pcm-kubernetes v0.0.0-20240301071143-347480abff2c/go.mod h1:lSRfGs+PxFvw7CcndHWRd6UlLlGrZn0b0hp5cfaMNGw=
|
||||||
gitlink.org.cn/JointCloud/pcm-octopus v0.0.0-20240424085753-6899615e9142 h1:+po0nesBDSWsgCySBG7eEXk7i9Ytd58wqvjL1M9y6d8=
|
gitlink.org.cn/JointCloud/pcm-octopus v0.0.0-20240510133934-6a5526289b35 h1:E2QfpS3Y0FjR8Zyv5l2Ti/2NetQFqHG66c8+T/+J1u0=
|
||||||
gitlink.org.cn/JointCloud/pcm-octopus v0.0.0-20240424085753-6899615e9142/go.mod h1:QOD5+/l2D+AYBjF2h5T0mdJyfGAmF78QmeKdbBXbjLQ=
|
gitlink.org.cn/JointCloud/pcm-octopus v0.0.0-20240510133934-6a5526289b35/go.mod h1:QOD5+/l2D+AYBjF2h5T0mdJyfGAmF78QmeKdbBXbjLQ=
|
||||||
gitlink.org.cn/JointCloud/pcm-openstack v0.0.0-20240403033338-e7edabad4203 h1:s6PsZ1+bev294IWdZRlV7mnOwI1+UzFcldVW/BqhQzI=
|
gitlink.org.cn/JointCloud/pcm-openstack v0.0.0-20240403033338-e7edabad4203 h1:s6PsZ1+bev294IWdZRlV7mnOwI1+UzFcldVW/BqhQzI=
|
||||||
gitlink.org.cn/JointCloud/pcm-openstack v0.0.0-20240403033338-e7edabad4203/go.mod h1:i2rrbMQ+Fve345BY9Heh4MUqVTAimZQElQhzzRee5B8=
|
gitlink.org.cn/JointCloud/pcm-openstack v0.0.0-20240403033338-e7edabad4203/go.mod h1:i2rrbMQ+Fve345BY9Heh4MUqVTAimZQElQhzzRee5B8=
|
||||||
gitlink.org.cn/JointCloud/pcm-slurm v0.0.0-20240301080743-8b94bbaf57f5 h1:+/5vnzkJBfMRnya1NrhOzlroUtRa5ePiYbPKlHLoLV0=
|
gitlink.org.cn/JointCloud/pcm-slurm v0.0.0-20240301080743-8b94bbaf57f5 h1:+/5vnzkJBfMRnya1NrhOzlroUtRa5ePiYbPKlHLoLV0=
|
||||||
|
|
|
@ -8,6 +8,7 @@ import (
|
||||||
type TaskCloudModel struct {
|
type TaskCloudModel struct {
|
||||||
Id uint `json:"id" gorm:"primarykey;not null;comment:id"`
|
Id uint `json:"id" gorm:"primarykey;not null;comment:id"`
|
||||||
TaskId uint `json:"taskId" gorm:"not null;comment:task表id"`
|
TaskId uint `json:"taskId" gorm:"not null;comment:task表id"`
|
||||||
|
AdapterName string `json:"adapterName" gorm:"not null;comment:适配器名称"`
|
||||||
AdapterId uint `json:"adapterId" gorm:"not null;comment:适配器id"`
|
AdapterId uint `json:"adapterId" gorm:"not null;comment:适配器id"`
|
||||||
AdapterName string `json:"adapterName" gorm:"not null;comment:适配器名称"`
|
AdapterName string `json:"adapterName" gorm:"not null;comment:适配器名称"`
|
||||||
ClusterId uint `json:"clusterId" gorm:"not null;comment:集群id"`
|
ClusterId uint `json:"clusterId" gorm:"not null;comment:集群id"`
|
||||||
|
|
Loading…
Reference in New Issue