四版本接口WRK压测QPS汇总

四版本接口WRK压测QPS汇总表

from config import PG_USER, PG_PASSWORD, PG_HOST, PG_PORT, PG_DATABASE
from config import DEBUG
db_url = "postgresql+asyncpg://{username}:{password}@{host}:{port}/{dbname}".format(username=PG_USER,
                                                                                    password=PG_PASSWORD,
                                                                                    host=PG_HOST,
                                                                                    port=PG_PORT,
                                                                                    dbname=PG_DATABASE)
# sessionmanager = DatabaseSessionManager(db_url, {"echo": True})
# print(db_url)
sessionmanager = DatabaseSessionManager(db_url, echo=DEBUG, pool_pre_ping=True)
# sessionmanager = DatabaseSessionManager(settings.database_url, {"echo": settings.echo_sql})

async def get_db():
    async with sessionmanager.session() as session:
        yield session


async def get_conn():
    async with sessionmanager.connect() as conn:
        yield conn

import uuid6


@router.post("/test/v1/run")
async def rule_run_v1(user_id: str, proj_id: str, request: Request):
    res = {
        'status': 0,
        'data': {"trace_id":uuid6.uuid6().hex},
        'message': 'version 1.0'
    }
    return res


@router.post("/test/v2/run")
async def rule_run_v2(user_id: str, proj_id: str, request: Request):
    # 引入了数据库
    async with sessionmanager.session() as db:
        cache_data = await disk_cache_get(db, user_id, cache_type='user')
    db_user_key = cache_data.get("user_key")
    db_username = cache_data.get("username")
    res = {
        'status': 0,
        'data': {"trace_id":uuid6.uuid6().hex},
        'message': 'version 1.0'
    }
    return res


@router.post("/test/v3/run")
async def rule_run_v3(user_id: str, proj_id: str, request: Request, db: AsyncSession = Depends(get_db)):
    user_key = request.query_params.get('user_key')
    cache_data = await disk_cache_get(db, user_id, cache_type='user')
    db_user_key = cache_data.get("user_key")
    db_username = cache_data.get("username")
    res = {
        'status': 0,
        'data': {"trace_id":uuid6.uuid6().hex},
        'message': 'version 1.0'
    }
    return res

版本说明

• V1：纯空接口（性能基准）
• V2：手动创建DB会话执行业务查询
• V3：FastAPI依赖注入DB会话执行业务查询
• 计算规则：降幅 = (V1-QPS - 当前QPS) ÷ V1-QPS × 100%

并发连接	V1基准QPS	V2手动DB QPS	V3注入DB QPS	V2相对V1降幅	V3相对V1降幅
10	4429.57	2905.49	2691.92	34.41%	39.23%
30	3579.51	2745.21	2598.52	23.31%	27.41%
50	3288.42	2605.41	2460.50	20.77%	25.18%
70	3201.90	2541.41	2388.81	20.63%	25.39%
90	3178.76	2528.87	2162.61	20.44%	31.97%
110	2934.36	2463.22	2019.72	16.05%	31.17%
130	3046.97	2345.72	2248.57	22.99%	26.20%
150	2812.34	2255.18	2229.34	19.81%	20.73%
170	2820.46	2300.91	2196.14	18.42%	22.13%
190	2742.53	2276.42	2042.78	16.99%	25.51%

---

核心结论

1. 空接口V1性能天花板极高，10并发可达4429 QPS，无任何业务与数据库开销
2. 接入数据库查询后性能大幅下跌，手动DB会话(V2)全程优于依赖注入(V3)
3. 同业务逻辑下，依赖注入DB相比手动自建会话，整体再多跌3%~8%性能
4. 并发越高，三者性能差距逐步收窄，瓶颈由代码开销转向数据库连接池与IO