Merge pull request 'updated deployinstance logics' (#278) from tzwang/pcm-coordinator:master into master

desc/inference/inference.api

@@ -120,4 +120,28 @@ type (
           Running int32 `json:"running"`
           Total int32 `json:"total"`
      }
+
+     StartAllByDeployTaskIdReq {
+          Id      string `form:"deployTaskid"`
+     }
+
+     StartAllByDeployTaskIdResp {
+
+     }
+
+     StopAllByDeployTaskIdReq {
+          Id      string `form:"deployTaskid"`
+     }
+
+     StopAllByDeployTaskIdResp {
+
+     }
+
+     GetDeployTasksReq {
+          PageInfo
+     }
+
+     GetDeployTasksResp {
+          PageResult
+     }
 )

desc/pcm.api

@@ -959,6 +959,15 @@ service pcm {
 
 	@handler InferenceTaskStatHandler
 	get /inference/taskStat (InferenceTaskStatReq) returns (InferenceTaskStatResp)
+
+	@handler StartAllByDeployTaskId
+	post /inference/startAll (StartAllByDeployTaskIdReq) returns (StartAllByDeployTaskIdResp)
+
+	@handler StopAllByDeployTaskId
+	post /inference/stopAll (StopAllByDeployTaskIdReq) returns (StopAllByDeployTaskIdResp)
+
+	@handler GetDeployTasks
+	get /inference/getDeployTasks (GetDeployTasksReq) returns (GetDeployTasksResp)
 }
 
 @server(

internal/handler/inference/getdeploytaskshandler.go

@@ -0,0 +1,28 @@
+package inference
+
+import (
+	"net/http"
+
+	"github.com/zeromicro/go-zero/rest/httpx"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/logic/inference"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
+)
+
+func GetDeployTasksHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		var req types.GetDeployTasksReq
+		if err := httpx.Parse(r, &req); err != nil {
+			httpx.ErrorCtx(r.Context(), w, err)
+			return
+		}
+
+		l := inference.NewGetDeployTasksLogic(r.Context(), svcCtx)
+		resp, err := l.GetDeployTasks(&req)
+		if err != nil {
+			httpx.ErrorCtx(r.Context(), w, err)
+		} else {
+			httpx.OkJsonCtx(r.Context(), w, resp)
+		}
+	}
+}

internal/handler/inference/startallbydeploytaskidhandler.go

@@ -0,0 +1,28 @@
+package inference
+
+import (
+	"net/http"
+
+	"github.com/zeromicro/go-zero/rest/httpx"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/logic/inference"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
+)
+
+func StartAllByDeployTaskIdHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		var req types.StartAllByDeployTaskIdReq
+		if err := httpx.Parse(r, &req); err != nil {
+			httpx.ErrorCtx(r.Context(), w, err)
+			return
+		}
+
+		l := inference.NewStartAllByDeployTaskIdLogic(r.Context(), svcCtx)
+		resp, err := l.StartAllByDeployTaskId(&req)
+		if err != nil {
+			httpx.ErrorCtx(r.Context(), w, err)
+		} else {
+			httpx.OkJsonCtx(r.Context(), w, resp)
+		}
+	}
+}

internal/handler/inference/stopallbydeploytaskidhandler.go

@@ -0,0 +1,28 @@
+package inference
+
+import (
+	"net/http"
+
+	"github.com/zeromicro/go-zero/rest/httpx"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/logic/inference"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
+)
+
+func StopAllByDeployTaskIdHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		var req types.StopAllByDeployTaskIdReq
+		if err := httpx.Parse(r, &req); err != nil {
+			httpx.ErrorCtx(r.Context(), w, err)
+			return
+		}
+
+		l := inference.NewStopAllByDeployTaskIdLogic(r.Context(), svcCtx)
+		resp, err := l.StopAllByDeployTaskId(&req)
+		if err != nil {
+			httpx.ErrorCtx(r.Context(), w, err)
+		} else {
+			httpx.OkJsonCtx(r.Context(), w, resp)
+		}
+	}
+}

internal/handler/routes.go

@@ -1213,6 +1213,21 @@ func RegisterHandlers(server *rest.Server, serverCtx *svc.ServiceContext) {
 				Path:    "/inference/taskStat",
 				Handler: inference.InferenceTaskStatHandler(serverCtx),
 			},
+			{
+				Method:  http.MethodPost,
+				Path:    "/inference/startAll",
+				Handler: inference.StartAllByDeployTaskIdHandler(serverCtx),
+			},
+			{
+				Method:  http.MethodPost,
+				Path:    "/inference/stopAll",
+				Handler: inference.StopAllByDeployTaskIdHandler(serverCtx),
+			},
+			{
+				Method:  http.MethodGet,
+				Path:    "/inference/getDeployTasks",
+				Handler: inference.GetDeployTasksHandler(serverCtx),
+			},
 		},
 		rest.WithPrefix("/pcm/v1"),
 	)

internal/logic/inference/getdeploytaskslogic.go

@@ -0,0 +1,30 @@
+package inference
+
+import (
+	"context"
+
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
+
+	"github.com/zeromicro/go-zero/core/logx"
+)
+
+type GetDeployTasksLogic struct {
+	logx.Logger
+	ctx    context.Context
+	svcCtx *svc.ServiceContext
+}
+
+func NewGetDeployTasksLogic(ctx context.Context, svcCtx *svc.ServiceContext) *GetDeployTasksLogic {
+	return &GetDeployTasksLogic{
+		Logger: logx.WithContext(ctx),
+		ctx:    ctx,
+		svcCtx: svcCtx,
+	}
+}
+
+func (l *GetDeployTasksLogic) GetDeployTasks(req *types.GetDeployTasksReq) (resp *types.GetDeployTasksResp, err error) {
+	// todo: add your logic here and delete this line
+
+	return
+}

internal/logic/inference/startallbydeploytaskidlogic.go

@@ -0,0 +1,47 @@
+package inference
+
+import (
+	"context"
+	"errors"
+	"github.com/zeromicro/go-zero/core/logx"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/updater"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
+	"strconv"
+)
+
+type StartAllByDeployTaskIdLogic struct {
+	logx.Logger
+	ctx    context.Context
+	svcCtx *svc.ServiceContext
+}
+
+func NewStartAllByDeployTaskIdLogic(ctx context.Context, svcCtx *svc.ServiceContext) *StartAllByDeployTaskIdLogic {
+	return &StartAllByDeployTaskIdLogic{
+		Logger: logx.WithContext(ctx),
+		ctx:    ctx,
+		svcCtx: svcCtx,
+	}
+}
+
+func (l *StartAllByDeployTaskIdLogic) StartAllByDeployTaskId(req *types.StartAllByDeployTaskIdReq) (resp *types.StartAllByDeployTaskIdResp, err error) {
+	resp = &types.StartAllByDeployTaskIdResp{}
+
+	id, err := strconv.ParseInt(req.Id, 10, 64)
+
+	list, err := l.svcCtx.Scheduler.AiStorages.GetInstanceListByDeployTaskId(id)
+	if err != nil {
+		return nil, err
+	}
+
+	for _, ins := range list {
+		success := l.svcCtx.Scheduler.AiService.InferenceAdapterMap[strconv.FormatInt(ins.AdapterId, 10)][strconv.FormatInt(ins.ClusterId, 10)].StartInferDeployInstance(l.ctx, ins.InstanceId)
+		if !success {
+			return nil, errors.New(ins.InstanceName + " start failed")
+		}
+	}
+
+	go updater.UpdateDeployInstanceStatusBatch(l.svcCtx, list)
+
+	return resp, nil
+}

internal/logic/inference/stopallbydeploytaskidlogic.go

@@ -0,0 +1,48 @@
+package inference
+
+import (
+	"context"
+	"errors"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/updater"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
+	"strconv"
+
+	"github.com/zeromicro/go-zero/core/logx"
+)
+
+type StopAllByDeployTaskIdLogic struct {
+	logx.Logger
+	ctx    context.Context
+	svcCtx *svc.ServiceContext
+}
+
+func NewStopAllByDeployTaskIdLogic(ctx context.Context, svcCtx *svc.ServiceContext) *StopAllByDeployTaskIdLogic {
+	return &StopAllByDeployTaskIdLogic{
+		Logger: logx.WithContext(ctx),
+		ctx:    ctx,
+		svcCtx: svcCtx,
+	}
+}
+
+func (l *StopAllByDeployTaskIdLogic) StopAllByDeployTaskId(req *types.StopAllByDeployTaskIdReq) (resp *types.StopAllByDeployTaskIdResp, err error) {
+	resp = &types.StopAllByDeployTaskIdResp{}
+
+	id, err := strconv.ParseInt(req.Id, 10, 64)
+
+	list, err := l.svcCtx.Scheduler.AiStorages.GetInstanceListByDeployTaskId(id)
+	if err != nil {
+		return nil, err
+	}
+
+	for _, ins := range list {
+		success := l.svcCtx.Scheduler.AiService.InferenceAdapterMap[strconv.FormatInt(ins.AdapterId, 10)][strconv.FormatInt(ins.ClusterId, 10)].StopInferDeployInstance(l.ctx, ins.InstanceId)
+		if !success {
+			return nil, errors.New(ins.InstanceName + " stop failed")
+		}
+	}
+
+	go updater.UpdateDeployInstanceStatusBatch(l.svcCtx, list)
+
+	return resp, nil
+}

internal/scheduler/database/aiStorage.go

@@ -421,6 +421,16 @@ func (s *AiStorage) GetInferDeployInstanceById(id int64) (*models.AiInferDeployI
 	return &deployIns, nil
 }
 
+func (s *AiStorage) GetInstanceListByDeployTaskId(id int64) ([]*models.AiInferDeployInstance, error) {
+	var list []*models.AiInferDeployInstance
+	tx := s.DbEngin.Raw("select * from ai_infer_deploy_instance where `deploy_instance_task_id` = ?", id).Scan(&list)
+	if tx.Error != nil {
+		logx.Errorf(tx.Error.Error())
+		return nil, tx.Error
+	}
+	return list, nil
+}
+
 func (s *AiStorage) GetInferDeployInstanceList() ([]*models.AiInferDeployInstance, error) {
 	var list []*models.AiInferDeployInstance
 	tx := s.DbEngin.Raw("select * from ai_infer_deploy_instance").Scan(&list)

internal/scheduler/service/inference/imageInference/imageToText.go

@@ -3,7 +3,7 @@ package imageInference
 import "gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/service/inference"
 
 const (
-	IMAGETOTEXT        = "image-to-text"
+	IMAGETOTEXT        = "image"
 	IMAGETOTEXT_AiTYPE = "13"
 )
 

internal/storeLink/storeLink.go

@@ -81,6 +81,7 @@ var (
 	ModelTypeMap = map[string][]string{
 		"image_classification": {"imagenet_resnet50"},
 		"text_to_text":         {"chatGLM_6B"},
+		"image_to_text":        {"blip-image-captioning-base"},
 	}
 	AITYPE = map[string]string{
 		"1": OCTOPUS,

internal/types/types.go

@@ -6012,3 +6012,25 @@ type InferenceTaskStatResp struct {
 	Running int32 `json:"running"`
 	Total   int32 `json:"total"`
 }
+
+type StartAllByDeployTaskIdReq struct {
+	Id string `form:"deployTaskid"`
+}
+
+type StartAllByDeployTaskIdResp struct {
+}
+
+type StopAllByDeployTaskIdReq struct {
+	Id string `form:"deployTaskid"`
+}
+
+type StopAllByDeployTaskIdResp struct {
+}
+
+type GetDeployTasksReq struct {
+	PageInfo
+}
+
+type GetDeployTasksResp struct {
+	PageResult
+}

pkg/models/aiinferdeployinstancemodel_gen.go

@@ -35,20 +35,21 @@ type (
 	}
 
 	AiInferDeployInstance struct {
-		Id           int64  `db:"id" json:"id,string"`
-		InstanceId   string `db:"instance_id" json:"instanceId"`
-		InstanceName string `db:"instance_name" json:"instanceName"`
-		AdapterId    int64  `db:"adapter_id" json:"adapterId,string"`
-		AdapterName  string `db:"adapter_name" json:"adapterName"	`
-		ClusterId    int64  `db:"cluster_id" json:"clusterId,string"`
-		ClusterName  string `db:"cluster_name" json:"clusterName"`
-		ModelName    string `db:"model_name" json:"modelName"`
-		ModelType    string `db:"model_type" json:"modelType"`
-		InferCard    string `db:"infer_card" json:"inferCard"`
-		Status       string `db:"status" json:"status"`
-		CreateTime   string `db:"create_time" json:"createTime"`
-		UpdateTime   string `db:"update_time" json:"updateTime"`
-		ClusterType  string `db:"cluster_type" json:"clusterType"`
+		Id                   int64  `db:"id" json:"id,string"`
+		DeployInstanceTaskId int64  `db:"deploy_instance_task_id" json:"deployTaskId,string"`
+		InstanceId           string `db:"instance_id" json:"instanceId"`
+		InstanceName         string `db:"instance_name" json:"instanceName"`
+		AdapterId            int64  `db:"adapter_id" json:"adapterId,string"`
+		AdapterName          string `db:"adapter_name" json:"adapterName"	`
+		ClusterId            int64  `db:"cluster_id" json:"clusterId,string"`
+		ClusterName          string `db:"cluster_name" json:"clusterName"`
+		ModelName            string `db:"model_name" json:"modelName"`
+		ModelType            string `db:"model_type" json:"modelType"`
+		InferCard            string `db:"infer_card" json:"inferCard"`
+		Status               string `db:"status" json:"status"`
+		CreateTime           string `db:"create_time" json:"createTime"`
+		UpdateTime           string `db:"update_time" json:"updateTime"`
+		ClusterType          string `db:"cluster_type" json:"clusterType"`
 	}
 )