Files
model-gateway/service/task/task_service.go

277 lines
7.6 KiB
Go
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
package task
import (
"context"
"errors"
"fmt"
"model-gateway/common/util"
"model-gateway/consts/public"
"time"
"model-gateway/dao"
"model-gateway/model/dto"
"model-gateway/model/entity"
"gitea.redpowerfuture.com/red-future/common/beans"
"gitea.redpowerfuture.com/red-future/common/utils"
"github.com/gogf/gf/v2/database/gdb"
"github.com/gogf/gf/v2/frame/g"
"github.com/gogf/gf/v2/util/gconv"
"github.com/google/uuid"
)
var ModelGatewayTask = &taskService{}
type taskService struct{}
// Create 创建任务
func (s *taskService) Create(ctx context.Context, req *dto.CreateTaskReq) (res *dto.CreateTaskRes, err error) {
taskID := uuid.NewString()
startAt := time.Now()
// 1) 获取用户信息
userInfo, err := utils.GetUserInfo(ctx)
if err != nil {
return nil, err
}
// 2) 检查模型配置
model, err := dao.ModelGatewayModels.Get(ctx, &entity.ModelGatewayModel{
SQLBaseDO: beans.SQLBaseDO{
TenantId: userInfo.TenantId,
Creator: userInfo.UserName,
},
ModelName: req.ModelName,
})
if err != nil {
return nil, err
}
if model == nil || (model.Enabled != nil && *model.Enabled != 1) {
return nil, errors.New("模型不存在或未启用")
}
// TODO: 排队控制暂时关闭,后续需要时取消注释
// limit := queue.GetRuntimeQueueLimit(ctx, req.ModelName, model.MaxConcurrency*2)
// if limit > 0 {
// ok, err := queue.AcquireQueueSlot(ctx, req.ModelName, taskID, limit, model.TimeoutSeconds)
// if err != nil {
// return nil, err
// }
// if !ok {
// return nil, errors.New("任务排队已满,请稍后再试")
// }
// }
// 3) 构建任务实体
task := &entity.ModelGatewayTask{
ModelName: model.ModelName,
TaskID: taskID,
State: public.TaskStatusRunning,
BizName: req.BizName,
CallbackURL: req.CallbackUrl,
RequestPayload: &entity.RequestPayload{
Body: req.RequestPayload,
Headers: util.ParseHeadMsgHeaders(model.HeadMsg),
},
EpicycleId: req.EpicycleId,
}
// 4) 插入任务记录
id, err := dao.ModelGatewayTask.Insert(ctx, task)
if err != nil {
// TODO: 恢复排队逻辑后,此处需要回滚排队占位
// queue.ReleaseQueueSlot(ctx, req.ModelName, taskID)
return nil, err
}
task.Id = id
// 5) 记录操作日志(非关键路径,失败不影响主流程)
ip, ua := "", ""
if r := g.RequestFromCtx(ctx); r != nil {
ip = utils.GetLocalIP()
ua = r.UserAgent()
}
_, _ = dao.ModelGatewayLogsOp.Insert(ctx, &entity.ModelGatewayLogsOp{
IP: ip,
UserAgent: ua,
APIPath: "/task/createTask",
HttpMethod: "POST",
BizName: req.BizName,
ModelName: req.ModelName,
TaskID: taskID,
OpType: "createTask",
Success: 1,
CostMs: time.Since(startAt).Milliseconds(),
RequestPayload: task.RequestPayload,
ResponsePayload: gdb.Map{"taskId": taskID},
})
// 6) 异步执行任务
go AsyncWorker.handleOne(util.AsyncCtx(ctx), task, model, req)
return &dto.CreateTaskRes{TaskID: taskID}, nil
}
// GetResult 获取任务结果
func (s *taskService) GetResult(ctx context.Context, taskID string) (res *dto.GetTaskResultRes, err error) {
t, err := dao.ModelGatewayTask.Get(ctx, &entity.ModelGatewayTask{
TaskID: taskID,
})
if err != nil {
return nil, err
}
if t == nil {
return nil, errors.New("任务不存在")
}
return &dto.GetTaskResultRes{
OssFile: t.ResultFile.OssFile,
State: t.State,
}, nil
}
// GetBatch 批量查询任务;将成功(state=2)的任务更新为已下载(state=4),并写入过期时间
func (s *taskService) GetBatch(ctx context.Context, req *dto.GetTaskBatchReq) (res *dto.GetTaskBatchRes, err error) {
if req == nil || len(req.TaskIDs) == 0 {
return &dto.GetTaskBatchRes{List: []dto.GetTaskBatchItem{}}, nil
}
// 1) 先查当前租户下的任务列表
list, err := dao.ModelGatewayTask.ListByTaskIDs(ctx, req.TaskIDs)
if err != nil {
return nil, err
}
// 2) 对成功(state=2)的任务:标记为已下载(state=4)
for _, t := range list {
if t == nil {
continue
}
if t.State != public.BuildTypeNode {
continue
}
_ = dao.ModelGatewayTask.MarkDownloadedByID(ctx, t.Id)
// 为了本次返回一致性,内存里也更新
t.State = public.TaskStatusDownloaded
}
// 3) 组装返回
items := make([]dto.GetTaskBatchItem, 0, len(list))
for _, t := range list {
if t == nil {
continue
}
items = append(items, dto.GetTaskBatchItem{
TaskID: t.TaskID,
State: t.State,
OssFile: t.ResultFile.OssFile,
TextResult: t.TextResult,
})
}
return &dto.GetTaskBatchRes{List: items}, nil
}
// List 获取任务列表
func (s *taskService) List(ctx context.Context, req *dto.ListTaskReq) (*dto.ListTaskRes, error) {
if req.PageNum <= 0 {
req.PageNum = 1
}
if req.PageSize <= 0 {
req.PageSize = 10
}
user, err := utils.GetUserInfo(ctx)
if err != nil {
return nil, err
}
list, total, err := dao.ModelGatewayTask.List(ctx, req.PageNum, req.PageSize, &entity.ModelGatewayTask{
SQLBaseDO: beans.SQLBaseDO{
Creator: user.UserName,
},
ModelName: req.ModelName,
BizName: req.BizName,
State: req.State,
TaskID: req.TaskID,
})
if err != nil {
return nil, err
}
return &dto.ListTaskRes{List: list, Total: total}, nil
}
// ModelTaskCallback 模型异步任务的回调通知
func (s *taskService) ModelTaskCallback(ctx context.Context, req *dto.ModelTaskCallbackReq) (*dto.ModelTaskCallbackRes, error) {
g.Log().Infof(ctx, "[模型回调] 收到通知 taskID=%s status=%s", req.TaskID, req.Status)
// 1. 查本地任务
task, err := dao.ModelGatewayTask.Get(ctx, &entity.ModelGatewayTask{
TaskID: req.TaskID,
})
if err != nil || task == nil {
return nil, fmt.Errorf("任务不存在: %s", req.TaskID)
}
// 2. 成功:取 video_url 和 usage
if req.Status == "succeeded" {
result := map[string]any{
"video_url": req.Content["video_url"],
"usage": req.Usage,
}
NotifyAsyncResult(req.TaskID, result, nil)
return &dto.ModelTaskCallbackRes{Success: true}, nil
}
// 3. 失败/过期
if req.Status == "failed" || req.Status == "expired" {
NotifyAsyncResult(req.TaskID, nil, fmt.Errorf(req.Status))
return &dto.ModelTaskCallbackRes{Success: true}, nil
}
return &dto.ModelTaskCallbackRes{Success: true}, nil
}
// QueryPendingTasks 批量轮询进行中的异步任务
func (s *taskService) QueryPendingTasks(ctx context.Context, req *dto.QueryPendingTasksReq) (*dto.QueryPendingTasksRes, error) {
limit := req.Limit
if limit <= 0 {
limit = g.Cfg().MustGet(ctx, "asynch.queryPending.limit", 10).Int()
}
// 1. 查 state=1执行中的异步任务
tasks, err := dao.ModelGatewayTask.GetPendingAsyncTasks(ctx, limit)
if err != nil {
return nil, err
}
// 2. 逐个查询
var results []dto.QueryTaskItem
for _, t := range tasks {
// 拿到模型配置
model, err := dao.ModelGatewayModels.GetByModelNameForTenant(ctx, t.TenantId, t.ModelName)
if err != nil || model == nil || model.QueryConfig == nil {
continue
}
result, err := util.PullTaskResult(ctx, nil, model.QueryConfig, model.HeadMsg)
if err != nil {
g.Log().Warningf(ctx, "[轮询] 查询失败 taskID=%s err=%v", t.TaskID, err)
continue
}
status := gconv.String(result["status"])
item := dto.QueryTaskItem{
TaskID: t.TaskID,
Status: status,
Content: result["content"].(map[string]any),
Usage: result["usage"].(map[string]any),
}
results = append(results, item)
// 如果任务完成,通知等待通道
if status == "succeeded" || status == "failed" || status == "expired" {
NotifyAsyncResult(t.TaskID, result["content"].(map[string]any), nil)
}
}
return &dto.QueryPendingTasksRes{
Total: len(results),
Results: results,
}, nil
}