refactor(model): 重构模型实体和数据访问层

This commit is contained in:
2026-05-21 10:41:37 +08:00
parent a080a5536d
commit 170568e03e
35 changed files with 903 additions and 1072 deletions

View File

@@ -2,8 +2,10 @@ package service
import (
"context"
"errors"
"fmt"
"math"
"model-gateway/model/dto"
"model-gateway/consts/public"
"model-gateway/model/entity"
@@ -34,9 +36,12 @@ type AutoTuneResult struct {
// - 基于吞吐与 P90 执行耗时估算 max_concurrency 的运行时值(不超过 cap
// - queue_limit 与 expected_seconds 绑定(允许排队时间 = expected_seconds * 2生成运行时值不超过 cap
// - 单次调整幅度限制 ±50%,写入 Redis带 TTL
func AutoTune(ctx context.Context, windowSeconds int) ([]AutoTuneResult, error) {
if windowSeconds <= 0 {
windowSeconds = 3600
func AutoTune(ctx context.Context, req *dto.AutoTuneReq) (res *dto.AutoTuneRes, err error) {
if req == nil {
return nil, errors.New("request cannot be nil")
}
if req.WindowSeconds <= 0 {
req.WindowSeconds = 3600 // 默认1小时
}
// 1) 读取模型配置cap按 model_name 聚合去重(如果表里有多租户重复数据,取较大上限)
var modelRows []*entity.AsynchModel
@@ -68,7 +73,7 @@ func AutoTune(ctx context.Context, windowSeconds int) ([]AutoTuneResult, error)
}
}
if len(modelMap) == 0 {
return []AutoTuneResult{}, nil
return nil, errors.New("no models found")
}
// 2) 统计指定窗口:按 model_name 计算 cnt 和 P90 执行耗时
@@ -89,7 +94,7 @@ SELECT model_name,
AND finished_at IS NOT NULL
AND finished_at >= (NOW() - (? || ' seconds')::interval)
GROUP BY model_name`, public.TableNameTask)
r, err := gfdb.DB(ctx).GetAll(ctx, sql, windowSeconds)
r, err := gfdb.DB(ctx).GetAll(ctx, sql, req.WindowSeconds)
if err != nil {
return nil, err
}
@@ -189,6 +194,8 @@ SELECT model_name,
})
}
g.Log().Infof(ctx, "[auto_tune] done models=%d windowSeconds=%d", len(out), windowSeconds)
return out, nil
g.Log().Infof(ctx, "[auto_tune] done models=%d windowSeconds=%d", len(out), req.WindowSeconds)
return &dto.AutoTuneRes{
List: out,
}, nil
}