Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
31 KiB
接口 QPS + 耗时可观测(OpenObserve)实现计划
For agentic workers: REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (
- [ ]) syntax for tracking.
Goal: 给 app-server 每个接口采集 QPS + 耗时 + 错误率,经轻量 ASGI 中间件 + 后台 worker 批量直采到本地 Docker 的 OpenObserve。
Architecture: 纯 ASGI 中间件测每请求耗时/抓路由模板+状态码 → 非阻塞入有界队列(满则丢、绝不阻塞)→ 后台 asyncio worker 批量 POST 到 OpenObserve _json ingest 端点。请求路径零 I/O;未配置观测则整套 no-op;上报失败丢批不重试。
Tech Stack: FastAPI / Starlette ASGI 中间件、asyncio.Queue、httpx.AsyncClient(已有依赖)、pydantic-settings、OpenObserve(Docker)。
参考 spec:docs/superpowers/specs/2026-07-06-openobserve-api-metrics-design.md
文件结构
| 文件 | 职责 |
|---|---|
app/core/config.py(改) |
新增 OBSERVE_* 配置 + observe_configured 门槛属性 |
app/core/observe.py(新) |
有界事件队列 + record_event + 路由模板解析 + RequestMetricsMiddleware |
app/core/observe_worker.py(新) |
后台批量上报 worker:_collect_batch / _post_batch / start_* / stop_* |
app/main.py(改) |
挂中间件(最外层)+ lifespan 启停 worker |
.env.example(改) |
新增 OBSERVE_* 注释段 |
deploy/openobserve/docker-compose.yml(新) |
本地 OpenObserve 容器 |
deploy/openobserve/README.md(新) |
部署步骤 + 查询/仪表盘 SQL |
tests/test_observe.py(新) |
配置门槛 / 队列 / 中间件 / worker 单测 |
关键接口契约(跨任务一致,勿改名):
app.core.observe.get_queue() -> asyncio.Queue[dict]app.core.observe.record_event(event: dict) -> Noneapp.core.observe.take_dropped() -> intapp.core.observe.RequestMetricsMiddleware(ASGI class,__init__(self, app))- 事件字段:
_timestamp(µs int) /service/env/method/route/status/duration_ms(float) app.core.observe_worker.start_observe_worker() -> asyncio.Task | Noneapp.core.observe_worker.stop_observe_worker(task) -> Nonesettings.observe_configured -> bool
Task 1: 配置项 OBSERVE_* + observe_configured
Files:
-
Modify:
app/core/config.py(在cors_origins_listproperty 之后、is_prodproperty 之前插入) -
Test:
tests/test_observe.py(新建) -
Step 1: 写失败测试
新建 tests/test_observe.py:
"""接口指标可观测(observe)单测:配置门槛 / 队列 / 中间件 / worker。
沿用仓库约定:TestClient + monkeypatch,绝不打真网络。observe 默认关(conftest 未设
OBSERVE_*),需要开启的用例用 monkeypatch 改 settings 单例属性。
"""
from __future__ import annotations
from app.core.config import settings
def test_observe_configured_requires_switch_and_creds(monkeypatch):
# 开关开 + endpoint(默认 localhost)+ user + password 齐全 → True
monkeypatch.setattr(settings, "OBSERVE_ENABLED", True)
monkeypatch.setattr(settings, "OBSERVE_USER", "u")
monkeypatch.setattr(settings, "OBSERVE_PASSWORD", "p")
assert settings.observe_configured is True
# 缺密码 → False
monkeypatch.setattr(settings, "OBSERVE_PASSWORD", "")
assert settings.observe_configured is False
# 开关关 → False(即便凭证齐全)
monkeypatch.setattr(settings, "OBSERVE_PASSWORD", "p")
monkeypatch.setattr(settings, "OBSERVE_ENABLED", False)
assert settings.observe_configured is False
- Step 2: 运行,确认失败
Run: pytest tests/test_observe.py::test_observe_configured_requires_switch_and_creds -q
Expected: FAIL —— AttributeError(settings 无 OBSERVE_ENABLED / 无 observe_configured)
- Step 3: 实现配置
在 app/core/config.py 的 cors_origins_list property 之后、is_prod property 之前插入:
# ===== 可观测(OpenObserve 接口指标)=====
# 采集每个接口的 QPS + 耗时 + 错误率,批量直采到 OpenObserve(本地 Docker)。
# 默认关(prod 安全):未开启 → 中间件透传、worker 不启动,整套 no-op。
# 开启需 ENABLED=true 且 ENDPOINT/USER/PASSWORD 齐全(见 observe_configured)。
OBSERVE_ENABLED: bool = False
OBSERVE_ENDPOINT: str = "http://localhost:5080" # OpenObserve base URL
OBSERVE_ORG: str = "default" # 组织名
OBSERVE_STREAM: str = "app_requests" # stream 名(首次上报自动建)
OBSERVE_USER: str = "" # Basic auth 邮箱
OBSERVE_PASSWORD: str = "" # Basic auth 密码/token
OBSERVE_FLUSH_INTERVAL_SEC: float = 5.0 # worker 最长攒批间隔
OBSERVE_BATCH_MAX: int = 200 # 单批最大事件数
OBSERVE_QUEUE_MAX: int = 10000 # 有界队列上限,满则丢
OBSERVE_TIMEOUT_SEC: float = 5.0 # 上报 HTTP 超时
@property
def observe_configured(self) -> bool:
"""观测上报可用 = 总开关开 且 endpoint/账号/密码齐全(缺则整套 no-op)。"""
return bool(
self.OBSERVE_ENABLED
and self.OBSERVE_ENDPOINT
and self.OBSERVE_USER
and self.OBSERVE_PASSWORD
)
- Step 4: 运行,确认通过
Run: pytest tests/test_observe.py::test_observe_configured_requires_switch_and_creds -q
Expected: PASS
- Step 5: 提交
git add app/core/config.py tests/test_observe.py
git commit -m "feat(observe): 加 OBSERVE_* 配置与 observe_configured 门槛"
Task 2: 事件队列 + record_event + take_dropped
Files:
-
Create:
app/core/observe.py -
Test:
tests/test_observe.py(追加) -
Step 1: 写失败测试
在 tests/test_observe.py 顶部 import 区补 import asyncio 和 from app.core import observe,并追加:
def test_record_event_enqueues(monkeypatch):
q = asyncio.Queue(maxsize=10)
monkeypatch.setattr(observe, "_queue", q)
observe.record_event({"route": "/x"})
assert q.get_nowait() == {"route": "/x"}
def test_record_event_drops_when_full(monkeypatch):
q = asyncio.Queue(maxsize=1)
monkeypatch.setattr(observe, "_queue", q)
monkeypatch.setattr(observe, "_dropped", 0)
observe.record_event({"n": 1}) # 占满
observe.record_event({"n": 2}) # 满 → 丢弃当前,不抛异常
assert observe.take_dropped() == 1
assert observe.take_dropped() == 0 # 取出后清零
assert q.get_nowait() == {"n": 1} # 保留的是先到的
- Step 2: 运行,确认失败
Run: pytest tests/test_observe.py -q -k record_event
Expected: FAIL —— ModuleNotFoundError: app.core.observe 或无 record_event
- Step 3: 实现
app/core/observe.py(先只放队列部分)
注意:本步只放队列相关代码。中间件用到的
os/time/Match及_SKIP_PATHS/_UNMATCHED/_SERVICE常量放到 Task 3 一并加入——否则本步提交时 ruff 会报 F401 未用导入。
新建 app/core/observe.py:
"""接口指标埋点:有界事件队列 + 纯 ASGI 中间件。
每个 HTTP 请求测总耗时、抓路由模板 + 状态码,非阻塞塞进有界队列;由 observe_worker
后台批量上报到 OpenObserve。请求路径上无任何 I/O。未配置观测时中间件直接透传。
"""
from __future__ import annotations
import asyncio
from app.core.config import settings
# 有界事件队列(懒创建,见 get_queue):首次取用时在运行中的 loop 里建,避免 import 期
# 无 loop 的边角问题;put_nowait/get_nowait 不需运行中的 loop → 可在无 loop 下测试。
_queue: asyncio.Queue[dict] | None = None
# 队列满时的丢弃计数,worker 定期取出打日志。
_dropped = 0
def get_queue() -> asyncio.Queue[dict]:
"""返回全局有界事件队列(懒创建)。测试可 monkeypatch 模块级 _queue 换成小队列。"""
global _queue
if _queue is None:
_queue = asyncio.Queue(maxsize=settings.OBSERVE_QUEUE_MAX)
return _queue
def take_dropped() -> int:
"""取出并清零累计丢弃数(供 worker 打点)。"""
global _dropped
n, _dropped = _dropped, 0
return n
def record_event(event: dict) -> None:
"""非阻塞入队;队列满则丢弃当前事件并计数。永不抛异常、永不阻塞请求。"""
global _dropped
try:
get_queue().put_nowait(event)
except asyncio.QueueFull:
_dropped += 1
- Step 4: 运行,确认通过
Run: pytest tests/test_observe.py -q -k record_event
Expected: PASS
- Step 5: 提交
git add app/core/observe.py tests/test_observe.py
git commit -m "feat(observe): 加有界事件队列与 record_event(满则丢)"
Task 3: RequestMetricsMiddleware(路由模板 + 状态码 + 耗时)
Files:
-
Modify:
app/core/observe.py(追加_resolve_route和中间件 class) -
Test:
tests/test_observe.py(追加) -
Step 1: 写失败测试
在 tests/test_observe.py 顶部 import 区补:
import pytest
from fastapi import FastAPI
from fastapi.testclient import TestClient
并追加:
def _make_probe_app() -> FastAPI:
"""独立最小 app:只挂中间件 + 两个无鉴权路由,不碰真业务 DB/auth。"""
app = FastAPI()
app.add_middleware(observe.RequestMetricsMiddleware)
@app.get("/things/{tid}")
def get_thing(tid: str):
return {"tid": tid}
@app.get("/health")
def health():
return {"ok": True}
return app
@pytest.fixture
def observe_on(monkeypatch):
"""开启观测 + 换一个干净小队列,返回该队列供断言。"""
monkeypatch.setattr(settings, "OBSERVE_ENABLED", True)
monkeypatch.setattr(settings, "OBSERVE_USER", "u")
monkeypatch.setattr(settings, "OBSERVE_PASSWORD", "p")
q = asyncio.Queue(maxsize=100)
monkeypatch.setattr(observe, "_queue", q)
return q
def test_middleware_records_route_template(observe_on):
client = TestClient(_make_probe_app())
r = client.get("/things/42")
assert r.status_code == 200
evt = observe_on.get_nowait()
assert evt["route"] == "/things/{tid}" # 模板,不是 /things/42
assert evt["method"] == "GET"
assert evt["status"] == 200
assert evt["duration_ms"] >= 0
assert evt["service"] and "env" in evt and isinstance(evt["_timestamp"], int)
def test_middleware_skips_health(observe_on):
client = TestClient(_make_probe_app())
client.get("/health")
assert observe_on.empty()
def test_middleware_unmatched_route_is_normalized(observe_on):
client = TestClient(_make_probe_app())
r = client.get("/definitely-not-a-route")
assert r.status_code == 404
evt = observe_on.get_nowait()
assert evt["route"] == "__unmatched__"
assert evt["status"] == 404
def test_middleware_noop_when_disabled(monkeypatch):
monkeypatch.setattr(settings, "OBSERVE_ENABLED", False)
q = asyncio.Queue(maxsize=100)
monkeypatch.setattr(observe, "_queue", q)
client = TestClient(_make_probe_app())
client.get("/things/1")
assert q.empty() # 未配置观测 → 零入队
- Step 2: 运行,确认失败
Run: pytest tests/test_observe.py -q -k middleware
Expected: FAIL —— AttributeError: module 'app.core.observe' has no attribute 'RequestMetricsMiddleware'
- Step 3a: 给
app/core/observe.py补中间件用的导入与常量
把顶部 import 段从
from __future__ import annotations
import asyncio
from app.core.config import settings
改成
from __future__ import annotations
import asyncio
import os
import time
from starlette.routing import Match
from app.core.config import settings
# 不采集的路径(纯噪音):健康检查。
_SKIP_PATHS = frozenset({"/health"})
# 未匹配路由(404/扫描器)归一到此,防维度爆炸。
_UNMATCHED = "__unmatched__"
# service 字段:与 logging.py 同源(LOG_SERVICE_NAME),默认 app-server。
_SERVICE = os.getenv("LOG_SERVICE_NAME", "app-server")
(_queue / _dropped / get_queue / take_dropped / record_event 保持不动。)
- Step 3b: 实现中间件(追加到
app/core/observe.py末尾)
def _resolve_route(scope) -> str:
"""从 scope 取路由模板(如 /things/{tid})。优先 scope['route'](现代 Starlette
路由后写入);取不到则手动匹配一次(老版本兜底);仍无 → __unmatched__(404/扫描器)。"""
route = scope.get("route")
path = getattr(route, "path", None)
if path:
return path
app_ = scope.get("app")
router = getattr(app_, "router", None)
for candidate in getattr(router, "routes", []):
try:
match, _ = candidate.matches(scope)
except Exception: # noqa: BLE001 - 匹配兜底,任一路由异常不影响整体
continue
if match == Match.FULL and getattr(candidate, "path", None):
return candidate.path
return _UNMATCHED
class RequestMetricsMiddleware:
"""纯 ASGI 中间件:测每个 http 请求耗时,记 method/route/status/duration。
放在最外层(main.py 里 CORS 之后 add),测到含 CORS 的完整耗时。未配置观测 → 透传。
"""
def __init__(self, app) -> None:
self.app = app
async def __call__(self, scope, receive, send) -> None:
if scope["type"] != "http" or not settings.observe_configured:
await self.app(scope, receive, send)
return
if scope.get("path") in _SKIP_PATHS:
await self.app(scope, receive, send)
return
start = time.perf_counter()
status_holder = {"status": 500} # 下游异常未产出 response 时兜底 500
async def send_wrapper(message) -> None:
if message["type"] == "http.response.start":
status_holder["status"] = message["status"]
await send(message)
try:
await self.app(scope, receive, send_wrapper)
finally:
duration_ms = (time.perf_counter() - start) * 1000.0
record_event({
"_timestamp": int(time.time() * 1_000_000), # µs,OpenObserve 时间列
"service": _SERVICE,
"env": settings.APP_ENV,
"method": scope.get("method", ""),
"route": _resolve_route(scope),
"status": status_holder["status"],
"duration_ms": round(duration_ms, 3),
})
- Step 4: 运行,确认通过
Run: pytest tests/test_observe.py -q -k middleware
Expected: PASS(4 个中间件用例全过)
若
test_middleware_records_route_template拿到的是/things/42而非模板,说明该 Starlette 版本未在scope["route"]写模板——此时_resolve_route的手动匹配兜底应已生效并返回模板;若仍不对,检查兜底分支是否被 import 顺序影响。
- Step 5: 提交
git add app/core/observe.py tests/test_observe.py
git commit -m "feat(observe): 加 RequestMetricsMiddleware(路由模板+状态码+耗时)"
Task 4: 后台上报 worker
Files:
-
Create:
app/core/observe_worker.py -
Test:
tests/test_observe.py(追加) -
Step 1: 写失败测试
在 tests/test_observe.py 顶部 import 区补:
import httpx
from app.core import observe_worker
并追加:
async def test_collect_batch_drains_up_to_batch_max(monkeypatch):
q = asyncio.Queue(maxsize=100)
monkeypatch.setattr(observe, "_queue", q)
monkeypatch.setattr(settings, "OBSERVE_FLUSH_INTERVAL_SEC", 0.1)
monkeypatch.setattr(settings, "OBSERVE_BATCH_MAX", 200)
for i in range(3):
q.put_nowait({"n": i})
batch = await observe_worker._collect_batch()
assert [e["n"] for e in batch] == [0, 1, 2]
async def test_collect_batch_timeout_returns_empty(monkeypatch):
q = asyncio.Queue(maxsize=100)
monkeypatch.setattr(observe, "_queue", q)
monkeypatch.setattr(settings, "OBSERVE_FLUSH_INTERVAL_SEC", 0.05)
batch = await observe_worker._collect_batch()
assert batch == []
async def test_post_batch_hits_json_ingest_url(monkeypatch):
monkeypatch.setattr(settings, "OBSERVE_ORG", "default")
monkeypatch.setattr(settings, "OBSERVE_STREAM", "app_requests")
captured = {}
def handler(request: httpx.Request) -> httpx.Response:
captured["url"] = str(request.url)
captured["json"] = request.content
return httpx.Response(200, json={"code": 200})
client = httpx.AsyncClient(
base_url="http://oo", transport=httpx.MockTransport(handler)
)
await observe_worker._post_batch(client, [{"route": "/x", "status": 200}])
await client.aclose()
assert captured["url"] == "http://oo/api/default/app_requests/_json"
assert b"/x" in captured["json"]
def test_start_observe_worker_noop_when_not_configured(monkeypatch):
monkeypatch.setattr(settings, "OBSERVE_ENABLED", False)
assert observe_worker.start_observe_worker() is None
- Step 2: 运行,确认失败
Run: pytest tests/test_observe.py -q -k "collect_batch or post_batch or start_observe"
Expected: FAIL —— ModuleNotFoundError: app.core.observe_worker
- Step 3: 实现
app/core/observe_worker.py
新建 app/core/observe_worker.py:
"""接口指标后台上报 worker:批量 drain 事件队列 → POST 到 OpenObserve。
对齐 heartbeat_monitor_worker 等的 start_*/stop_* 形态。best-effort 遥测:catch 全部
异常,上报失败直接丢批不重试。未配置观测 → start 返回 None(不启动),整套 no-op。
"""
from __future__ import annotations
import asyncio
import contextlib
import logging
import httpx
from app.core.config import settings
from app.core.observe import get_queue, take_dropped
logger = logging.getLogger("shagua.observe")
# 上报用的 httpx client,start 时建、stop 时关。
_client: httpx.AsyncClient | None = None
async def _collect_batch() -> list[dict]:
"""等到 ≥1 条(或到 flush 间隔)后,连抽到 BATCH_MAX 条或抽空。超时且空 → 返回 []。"""
queue = get_queue()
batch: list[dict] = []
try:
first = await asyncio.wait_for(
queue.get(), timeout=settings.OBSERVE_FLUSH_INTERVAL_SEC
)
except asyncio.TimeoutError:
return batch
batch.append(first)
while len(batch) < settings.OBSERVE_BATCH_MAX:
try:
batch.append(queue.get_nowait())
except asyncio.QueueEmpty:
break
return batch
async def _post_batch(client: httpx.AsyncClient, batch: list[dict]) -> None:
"""POST 一批事件到 OpenObserve 的 _json ingest 端点。非 2xx 仅告警。"""
url = f"/api/{settings.OBSERVE_ORG}/{settings.OBSERVE_STREAM}/_json"
resp = await client.post(url, json=batch)
if resp.status_code >= 300:
logger.warning(
"observe ingest failed status=%s body=%s",
resp.status_code,
resp.text[:200],
)
async def _run_loop(client: httpx.AsyncClient) -> None:
try:
while True:
batch = await _collect_batch()
dropped = take_dropped()
if dropped:
logger.warning("observe dropped %d events (queue full)", dropped)
if not batch:
continue
try:
await _post_batch(client, batch)
except Exception: # noqa: BLE001 - best-effort 遥测,失败丢批不重试、不退出
logger.warning(
"observe post batch failed, dropped %d events",
len(batch),
exc_info=True,
)
except asyncio.CancelledError:
logger.info("observe worker stopped")
raise
def start_observe_worker() -> asyncio.Task | None:
"""启动上报 worker。未配置观测 → 返回 None(no-op)。"""
global _client
if not settings.observe_configured:
return None
_client = httpx.AsyncClient(
base_url=settings.OBSERVE_ENDPOINT,
auth=(settings.OBSERVE_USER, settings.OBSERVE_PASSWORD),
timeout=settings.OBSERVE_TIMEOUT_SEC,
)
logger.info(
"observe worker started endpoint=%s org=%s stream=%s",
settings.OBSERVE_ENDPOINT,
settings.OBSERVE_ORG,
settings.OBSERVE_STREAM,
)
return asyncio.create_task(_run_loop(_client), name="observe-worker")
async def stop_observe_worker(task: asyncio.Task | None) -> None:
"""收尾:cancel worker → best-effort 发最后一批 → 关 client。"""
global _client
if task is None:
return
task.cancel()
with contextlib.suppress(asyncio.CancelledError):
await task
if _client is not None:
# worker 已停,安全 drain 剩余并 best-effort 发最后一批(短超时,不卡关停)。
try:
queue = get_queue()
final: list[dict] = []
while len(final) < settings.OBSERVE_BATCH_MAX:
try:
final.append(queue.get_nowait())
except asyncio.QueueEmpty:
break
if final:
await asyncio.wait_for(
_post_batch(_client, final), timeout=settings.OBSERVE_TIMEOUT_SEC
)
except Exception: # noqa: BLE001 - 关停期尽力而为,失败忽略
pass
await _client.aclose()
_client = None
- Step 4: 运行,确认通过
Run: pytest tests/test_observe.py -q -k "collect_batch or post_batch or start_observe"
Expected: PASS
- Step 5: 提交
git add app/core/observe_worker.py tests/test_observe.py
git commit -m "feat(observe): 加后台批量上报 worker(失败丢批不重试)"
Task 5: 接线到 app/main.py(挂中间件 + lifespan 启停)
Files:
-
Modify:
app/main.py(import 区、lifespan、CORS 之后) -
Test:
tests/test_observe.py(追加) -
Step 1: 写失败测试
在 tests/test_observe.py 追加:
def test_app_has_metrics_middleware():
from app.main import app
names = [m.cls.__name__ for m in app.user_middleware]
assert "RequestMetricsMiddleware" in names
- Step 2: 运行,确认失败
Run: pytest tests/test_observe.py::test_app_has_metrics_middleware -q
Expected: FAIL —— 断言失败(中间件尚未挂载)
- Step 3: 实现接线
3a. 在 app/main.py import 区(withdraw_reconcile_worker import 块之后)加:
from app.core.observe import RequestMetricsMiddleware
from app.core.observe_worker import (
start_observe_worker,
stop_observe_worker,
)
3b. lifespan 里加启停(现有 daily_exchange_task = start_daily_exchange_worker() 之后、try: 之前加一行;finally 里在 stop_daily_exchange_worker 之后加一行):
daily_exchange_task = start_daily_exchange_worker()
observe_task = start_observe_worker()
try:
yield
finally:
await stop_heartbeat_monitor(heartbeat_task)
await stop_withdraw_reconcile_worker(reconcile_task)
await stop_daily_exchange_worker(daily_exchange_task)
await stop_observe_worker(observe_task)
await aclose_pricebot_client()
logger.info("shutting down")
3c. 挂中间件——在 CORS 的 if settings.cors_origins_list: 整块之后加(使其成为最外层,测到含 CORS 的完整耗时):
# 接口指标埋点(最外层:测含 CORS 的完整耗时)。未配置观测时中间件自 no-op。
app.add_middleware(RequestMetricsMiddleware)
- Step 4: 运行,确认通过
Run: pytest tests/test_observe.py::test_app_has_metrics_middleware -q
Expected: PASS
- Step 5: 跑整套 observe 测试 + 全量回归,确认无破坏
Run: pytest tests/test_observe.py -q && pytest -q
Expected: 全 PASS(现有用例不受影响:conftest 未设 OBSERVE_* → 观测关 → worker no-op、中间件透传)
- Step 6: 提交
git add app/main.py tests/test_observe.py
git commit -m "feat(observe): main.py 挂中间件 + lifespan 启停上报 worker"
Task 6: OpenObserve 本地部署(compose + README + .env.example)
Files:
-
Create:
deploy/openobserve/docker-compose.yml -
Create:
deploy/openobserve/README.md -
Modify:
.env.example(追加OBSERVE_*段) -
Step 1: 写 docker-compose
新建 deploy/openobserve/docker-compose.yml:
# 本地开发用 OpenObserve(单容器 = local 模式)。用于接收 app-server 的接口指标。
# 启动: cd deploy/openobserve && docker compose up -d
# Web UI: http://localhost:5080 (账号见下方 env)
services:
openobserve:
image: public.ecr.aws/zinclabs/openobserve:latest
container_name: openobserve
ports:
- "5080:5080"
environment:
ZO_ROOT_USER_EMAIL: "admin@shaguabijia.local"
ZO_ROOT_USER_PASSWORD: "Complexpass#123"
ZO_DATA_DIR: "/data"
volumes:
- ./data:/data
restart: unless-stopped
- Step 2: 写 README
新建 deploy/openobserve/README.md:
# OpenObserve 本地部署(接口 QPS / 耗时可观测)
app-server 通过中间件采集每个接口的 QPS + 耗时 + 错误率,批量上报到这里。
## 启动
```bash
cd deploy/openobserve
docker compose up -d
```
- Web UI:http://localhost:5080
- 登录:`admin@shaguabijia.local` / `Complexpass#123`(见 `docker-compose.yml`)
- 数据落 `deploy/openobserve/data/`(已挂卷持久化;`data/` 建议 gitignore)
## 让 app-server 上报
在项目根的 `.env` 打开观测(`OBSERVE_*`,账号密码与 compose 里 root 一致):
```dotenv
OBSERVE_ENABLED=true
OBSERVE_ENDPOINT=http://localhost:5080
OBSERVE_ORG=default
OBSERVE_STREAM=app_requests
OBSERVE_USER=admin@shaguabijia.local
OBSERVE_PASSWORD=Complexpass#123
```
重启 app-server,随便打几个接口。stream `app_requests` **首次上报自动创建**,
在 UI 的 Logs → 选 `app_requests` 就能看到逐条请求事件。
## 查询(Logs 页 SQL,或建 Dashboard 面板)
各接口 QPS(1 分钟分桶,面板里再除 60 得每秒):
```sql
SELECT route, histogram(_timestamp, '1 minute') AS ts, count(*) AS cnt
FROM app_requests GROUP BY route, ts ORDER BY ts
```
各接口 P95 耗时(毫秒):
```sql
SELECT route, approx_percentile_cont(duration_ms, 0.95) AS p95_ms
FROM app_requests GROUP BY route ORDER BY p95_ms DESC
```
各接口错误率(5xx 占比):
```sql
SELECT route,
count(*) FILTER (WHERE status >= 500) * 100.0 / count(*) AS err_pct
FROM app_requests GROUP BY route ORDER BY err_pct DESC
```
## 停止 / 清数据
```bash
docker compose down # 停止(保留数据)
docker compose down -v && rm -rf data # 停止并清空数据
```
> 生产部署(持久化规格、独立 ingest 账号、鉴权收紧)见 spec 第 9 节,本期不做。
- Step 3: 追加
.env.example
在 .env.example 末尾追加:
# ===== 可观测(OpenObserve 接口指标)=====
# 采集每个接口 QPS + 耗时 + 错误率,批量直采到 OpenObserve(本地 Docker,见 deploy/openobserve/)。
# 默认关;开启需 ENABLED=true 且填 USER/PASSWORD(与 docker-compose 里 root 账号一致)。
OBSERVE_ENABLED=false
OBSERVE_ENDPOINT=http://localhost:5080
OBSERVE_ORG=default
OBSERVE_STREAM=app_requests
OBSERVE_USER=admin@shaguabijia.local
OBSERVE_PASSWORD=Complexpass#123
# 进阶(一般不用改):攒批间隔秒 / 单批最大条数 / 有界队列上限(满则丢) / 上报超时秒
OBSERVE_FLUSH_INTERVAL_SEC=5
OBSERVE_BATCH_MAX=200
OBSERVE_QUEUE_MAX=10000
OBSERVE_TIMEOUT_SEC=5
- Step 4: 校验 compose 语法(不需真拉镜像)
Run: docker compose -f deploy/openobserve/docker-compose.yml config
Expected: 打印规整后的配置、无报错(若本机无 docker,可跳过,标注为手动验证项)
- Step 5: 提交
git add deploy/openobserve/docker-compose.yml deploy/openobserve/README.md .env.example
git commit -m "feat(observe): 加 OpenObserve 本地 compose + README + .env.example"
Task 7: 端到端手动验证 + 全量 lint/test 收尾
Files: 无(验证 + 收尾)
- Step 1: 起 OpenObserve
Run: cd deploy/openobserve && docker compose up -d
Expected: 容器起来,浏览器打开 http://localhost:5080 能登录
- Step 2: 本地开观测起 app-server
在根 .env 设 OBSERVE_ENABLED=true + OBSERVE_USER/PASSWORD(同 compose),然后:
Run: ./run.sh(Windows 用 python -m uvicorn app.main:app --port 8770 --reload --reload-dir app)
Expected: 启动日志出现 observe worker started endpoint=http://localhost:5080 ...
- Step 3: 打几个接口产生数据
Run: curl http://localhost:8770/health && curl http://localhost:8770/things-does-not-exist -i(或正常业务接口若干)
Expected: 稍等 ≤5s(flush 间隔),OpenObserve UI 的 Logs → app_requests 出现事件;/health 不应出现;不存在的路径 route 为 __unmatched__
- Step 4: 验证三条查询
在 OpenObserve UI 分别粘贴 README 里的 QPS / P95 / 错误率 SQL,确认能出数。
- Step 5: lint(仅本改动涉及文件)+ 全量测试
说明:仓库基线有 ~558 个既有 ruff 错误、且未强制 ruff 通过。不要去清历史欠债(范围蔓延)。只要求本次新增/改动的文件零 ruff 错误。
Run: ruff check app/core/observe.py app/core/observe_worker.py tests/test_observe.py && python -m pytest -q
Expected: 上述三个新文件 ruff 无错;测试里 tests/test_observe.py 全 PASS,且全量失败数不超过基线的 4 个(test_compare_proxy ×2 / test_coupon_proxy ×1 / test_invite ×1,均与本功能无关)。
额外确认我对既有文件的改动没有引入新的 ruff 错误:ruff check app/core/config.py app/main.py(数量应与基线一致,不因本改动增加)。
- Step 6: 关观测复跑一次,确认降级
把 .env 的 OBSERVE_ENABLED 改回 false,ruff check . 不涉及,直接 pytest -q
Expected: 全 PASS(验证 observe 关闭时零副作用)
- Step 7: 收尾提交(如有 .env 之外的改动)
git add -A
git commit -m "chore(observe): 端到端验证与收尾" --allow-empty
.env不入 git(已 gitignore);本任务只验证,不提交.env。
Self-Review(写完计划后自查)
- Spec 覆盖:Docker 部署→Task 6/7;事件 schema→Task 3(
record_event事件字段);中间件→Task 3;worker→Task 4;配置→Task 1;main 接线→Task 5;查询/仪表盘→Task 6 README;测试→Task 1-5;决策(a)队列满丢→Task 2;(b)失败不重试→Task 4;(c)跳过 /health→Task 3。全覆盖。 - 占位符:无 TBD/TODO;每个代码步骤含完整代码。
- 类型/命名一致:
get_queue/record_event/take_dropped/RequestMetricsMiddleware/start_observe_worker/stop_observe_worker/observe_configured/ 事件字段名,跨 Task 1-5 与文件结构表一致。