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

updated inference apis 

Former-commit-id: d907a498a213be552236b306f48903942d63078d
This commit is contained in:
tzwang 2024-08-27 15:09:58 +08:00
parent 46e3bc064f
commit 5ec517690a
9 changed files with 101 additions and 88 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
desc/inference/inference.api
View File

@ -92,7 +92,7 @@ type (
AiClusterIds []string `form:"aiClusterIds"`
}
TextToImageInferenceResp{
Result []byte
Result []byte `json:"result"`
}
/******************Deploy instance*************************/
@ -156,20 +156,12 @@ type (
}
GetDeployTasksReq {
PageInfo
}
GetDeployTasksResp {
PageResult
}
GetRunningInstanceReq {
AdapterIds []string `form:"adapterIds"`
ModelType string `form:"modelType"`
}
GetRunningInstanceResp {
List interface{} `json:"list,omitempty"`
List interface{} `json:"list"`
}
GetDeployTasksByTypeReq {
@ -177,6 +169,6 @@ type (
}
GetDeployTasksByTypeResp {
List interface{} `json:"list,omitempty"`
List interface{} `json:"list"`
}
)

8
desc/pcm.api
View File

@ -966,11 +966,11 @@ service pcm {
@handler StopAllByDeployTaskId
post /inference/stopAll (StopAllByDeployTaskIdReq) returns (StopAllByDeployTaskIdResp)
@handler GetDeployTasks
get /inference/getDeployTasks (GetDeployTasksReq) returns (GetDeployTasksResp)
@handler GetRunningInstanceByType
get /inference/getInstanceByType (GetRunningInstanceReq) returns (GetRunningInstanceResp)
@handler GetRunningInstanceByModel
get /inference/getInstanceByModel (GetRunningInstanceReq) returns (GetRunningInstanceResp)
@handler GetDeployTasksByType
get /inference/getDeployTasksByType (GetDeployTasksByTypeReq) returns (GetDeployTasksByTypeResp)
}
@server(

8
internal/handler/inference/getdeploytaskshandler.go → internal/handler/inference/getdeploytasksbytypehandler.go
View File

@ -9,16 +9,16 @@ import (
"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
)
func GetDeployTasksHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
func GetDeployTasksByTypeHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
return func(w http.ResponseWriter, r *http.Request) {
var req types.GetDeployTasksReq
var req types.GetDeployTasksByTypeReq
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)
l := inference.NewGetDeployTasksByTypeLogic(r.Context(), svcCtx)
resp, err := l.GetDeployTasksByType(&req)
if err != nil {
httpx.ErrorCtx(r.Context(), w, err)
} else {

15
internal/handler/inference/getrunninginstancebymodelhandler.go → internal/handler/inference/getrunninginstancebytypehandler.go
View File

@ -1,7 +1,6 @@
package inference
import (
"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/repository/result"
"net/http"
"github.com/zeromicro/go-zero/rest/httpx"
@ -10,16 +9,20 @@ import (
"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
)
func GetRunningInstanceByModelHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
func GetRunningInstanceByTypeHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
return func(w http.ResponseWriter, r *http.Request) {
var req types.GetRunningInstanceReq
if err := httpx.Parse(r, &req); err != nil {
result.ParamErrorResult(r, w, err)
httpx.ErrorCtx(r.Context(), w, err)
return
}
l := inference.NewGetRunningInstanceByModelLogic(r.Context(), svcCtx)
resp, err := l.GetRunningInstanceByModel(&req)
result.HttpResult(r, w, resp, err)
l := inference.NewGetRunningInstanceByTypeLogic(r.Context(), svcCtx)
resp, err := l.GetRunningInstanceByType(&req)
if err != nil {
httpx.ErrorCtx(r.Context(), w, err)
} else {
httpx.OkJsonCtx(r.Context(), w, resp)
}
}
}

8
internal/handler/routes.go
View File

@ -1225,13 +1225,13 @@ func RegisterHandlers(server *rest.Server, serverCtx *svc.ServiceContext) {
},
{
Method: http.MethodGet,
Path: "/inference/getDeployTasks",
Handler: inference.GetDeployTasksHandler(serverCtx),
Path: "/inference/getInstanceByType",
Handler: inference.GetRunningInstanceByTypeHandler(serverCtx),
},
{
Method: http.MethodGet,
Path: "/inference/getInstanceByModel",
Handler: inference.GetRunningInstanceByModelHandler(serverCtx),
Path: "/inference/getDeployTasksByType",
Handler: inference.GetDeployTasksByTypeHandler(serverCtx),
},
},
rest.WithPrefix("/pcm/v1"),

8
internal/logic/inference/getdeploytaskslogic.go → internal/logic/inference/getdeploytasksbytypelogic.go
View File

@ -9,21 +9,21 @@ import (
"github.com/zeromicro/go-zero/core/logx"
)
type GetDeployTasksLogic struct {
type GetDeployTasksByTypeLogic struct {
logx.Logger
ctx context.Context
svcCtx *svc.ServiceContext
}
func NewGetDeployTasksLogic(ctx context.Context, svcCtx *svc.ServiceContext) *GetDeployTasksLogic {
return &GetDeployTasksLogic{
func NewGetDeployTasksByTypeLogic(ctx context.Context, svcCtx *svc.ServiceContext) *GetDeployTasksByTypeLogic {
return &GetDeployTasksByTypeLogic{
Logger: logx.WithContext(ctx),
ctx: ctx,
svcCtx: svcCtx,
}
}
func (l *GetDeployTasksLogic) GetDeployTasks(req *types.GetDeployTasksReq) (resp *types.GetDeployTasksResp, err error) {
func (l *GetDeployTasksByTypeLogic) GetDeployTasksByType(req *types.GetDeployTasksByTypeReq) (resp *types.GetDeployTasksByTypeResp, err error) {
// todo: add your logic here and delete this line
return

30
internal/logic/inference/getrunninginstancebymodellogic.go
View File

@ -1,30 +0,0 @@
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 GetRunningInstanceByModelLogic struct {
logx.Logger
ctx context.Context
svcCtx *svc.ServiceContext
}
func NewGetRunningInstanceByModelLogic(ctx context.Context, svcCtx *svc.ServiceContext) *GetRunningInstanceByModelLogic {
return &GetRunningInstanceByModelLogic{
Logger: logx.WithContext(ctx),
ctx: ctx,
svcCtx: svcCtx,
}
}
func (l *GetRunningInstanceByModelLogic) GetRunningInstanceByModel(req *types.GetRunningInstanceReq) (resp *types.GetRunningInstanceResp, err error) {
resp = &types.GetRunningInstanceResp{}
return
}

30
internal/logic/inference/getrunninginstancebytypelogic.go Normal file
View File

@ -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 GetRunningInstanceByTypeLogic struct {
logx.Logger
ctx context.Context
svcCtx *svc.ServiceContext
}
func NewGetRunningInstanceByTypeLogic(ctx context.Context, svcCtx *svc.ServiceContext) *GetRunningInstanceByTypeLogic {
return &GetRunningInstanceByTypeLogic{
Logger: logx.WithContext(ctx),
ctx: ctx,
svcCtx: svcCtx,
}
}
func (l *GetRunningInstanceByTypeLogic) GetRunningInstanceByType(req *types.GetRunningInstanceReq) (resp *types.GetRunningInstanceResp, err error) {
// todo: add your logic here and delete this line
return
}

68
internal/types/types.go
View File

@ -5904,6 +5904,19 @@ type Category struct {
Name string `json:"name"`
}
type DeployInstance struct {
InstanceId string `json:"instanceId"`
InstanceName string `json:"instanceName"`
AdapterId string `json:"adapterId"`
AdapterName string `json:"adapterName"`
ClusterId string `json:"clusterId"`
ClusterName string `json:"clusterName"`
ModelName string `json:"modelName"`
ModelType string `json:"modelType"`
InferCard string `json:"inferCard"`
Status string `json:"status"`
}
type ModelTypesResp struct {
ModelTypes []string `json:"types"`
}
@ -5917,20 +5930,13 @@ type ModelNamesResp struct {
}
type ImageInferenceReq struct {
TaskName string `form:"taskName"`
TaskDesc string `form:"taskDesc"`
ModelName string `form:"modelName"`
ModelType string `form:"modelType"`
AdapterId string `form:"adapterId"`
AiClusterIds []string `form:"aiClusterIds,optional"`
ResourceType string `form:"resourceType,optional"`
ComputeCard string `form:"card,optional"`
Strategy string `form:"strategy"`
StaticWeightMap map[string]int32 `form:"staticWeightMap,optional"`
Params []string `form:"params,optional"`
Envs []string `form:"envs,optional"`
Cmd string `form:"cmd,optional"`
Replica int32 `form:"replicas,optional"`
TaskName string `json:"taskName"`
TaskDesc string `json:"taskDesc"`
ModelType string `json:"modelType"`
Instances []DeployInstance `json:"instances"`
Strategy string `json:"strategy,,optional"`
StaticWeightMap map[string]int32 `json:"staticWeightMap,optional"`
Replica int32 `json:"replicas,optional"`
}
type ImageInferenceResp struct {
@ -5976,6 +5982,18 @@ type TextToTextInferenceReq struct {
type TextToTextInferenceResp struct {
}
type TextToImageInferenceReq struct {
TaskName string `form:"taskName"`
TaskDesc string `form:"taskDesc"`
ModelName string `form:"modelName"`
ModelType string `form:"modelType"`
AiClusterIds []string `form:"aiClusterIds"`
}
type TextToImageInferenceResp struct {
Result []byte `json:"result"`
}
type DeployInstanceListReq struct {
PageInfo
}
@ -6034,19 +6052,19 @@ type StopAllByDeployTaskIdReq struct {
type StopAllByDeployTaskIdResp struct {
}
type GetDeployTasksReq struct {
PageInfo
}
type GetDeployTasksResp struct {
PageResult
}
type GetRunningInstanceReq struct {
ModelType string `path:"modelType"`
ModelName string `path:"modelName"`
AdapterIds []string `form:"adapterIds"`
ModelType string `form:"modelType"`
}
type GetRunningInstanceResp struct {
List interface{} `json:"list,omitempty"`
List interface{} `json:"list"`
}
type GetDeployTasksByTypeReq struct {
ModelType string `form:"modelType"`
}
type GetDeployTasksByTypeResp struct {
List interface{} `json:"list"`
}