Compare commits
18 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
| 00404c8b45 | |||
| 964032d16f | |||
| 1f97b19d21 | |||
| 8938b951fb | |||
| 0cf4cc706b | |||
| a5f63cb53c | |||
| 65e9b422fe | |||
| 0350a140fe | |||
| 95a4c5c3cc | |||
| 822c2ca2a6 | |||
| 9354588ba2 | |||
| 6c624c3cea | |||
| 70c5c4cf08 | |||
| 804de02188 | |||
| 76cb981d84 | |||
| 20fa32e884 | |||
| 29aa6fbff8 | |||
| a563c1ca4b |
@@ -137,3 +137,19 @@ PANGLE_REPORT_SECURITY_KEY=
|
||||
# GroMore AppId(报表 site_id 维度)→ 应用环境;默认取现网两个应用,按需覆盖。
|
||||
PANGLE_REPORT_SITE_ID_PROD=5830519
|
||||
PANGLE_REPORT_SITE_ID_TEST=5832303
|
||||
|
||||
# ===== 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)。
|
||||
# 未开/缺凭证 → 中间件透传、worker 不启动,整套 no-op,不影响业务。
|
||||
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
|
||||
|
||||
@@ -317,6 +317,31 @@ class Settings(BaseSettings):
|
||||
return []
|
||||
return [o.strip() for o in self.CORS_ALLOW_ORIGINS.split(",") if o.strip()]
|
||||
|
||||
# ===== 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
|
||||
)
|
||||
|
||||
@property
|
||||
def is_prod(self) -> bool:
|
||||
return self.APP_ENV == "prod"
|
||||
|
||||
@@ -0,0 +1,110 @@
|
||||
"""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
|
||||
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")
|
||||
|
||||
# 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
|
||||
|
||||
|
||||
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),
|
||||
})
|
||||
@@ -0,0 +1,128 @@
|
||||
"""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: # noqa: UP041 - 3.10 兼容:该版 wait_for 抛的 asyncio.TimeoutError ≠ 内置 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)。约定每进程只调一次(lifespan)。"""
|
||||
global _client
|
||||
if not settings.observe_configured:
|
||||
return None
|
||||
if _client is not None:
|
||||
# 约定 start 每进程只调一次;已启动则不重复建 client(避免泄漏旧连接池)。
|
||||
logger.warning("observe worker already started; ignoring duplicate start")
|
||||
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
|
||||
dropped = take_dropped() # 收口:补记最后一个 flush 窗口累计的丢弃数,不让账丢在关停期
|
||||
if dropped:
|
||||
logger.warning("observe dropped %d events (queue full) before shutdown", dropped)
|
||||
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);
|
||||
# 超过一批(BATCH_MAX)的剩余直接丢,不做多轮 flush(best-effort,关停从速)。
|
||||
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
|
||||
+10
@@ -50,6 +50,11 @@ from app.core.heartbeat_monitor_worker import (
|
||||
stop_heartbeat_monitor,
|
||||
)
|
||||
from app.core.logging import setup_logging
|
||||
from app.core.observe import RequestMetricsMiddleware
|
||||
from app.core.observe_worker import (
|
||||
start_observe_worker,
|
||||
stop_observe_worker,
|
||||
)
|
||||
from app.core.pricebot_client import aclose_pricebot_client, get_pricebot_client
|
||||
from app.core.withdraw_reconcile_worker import (
|
||||
start_withdraw_reconcile_worker,
|
||||
@@ -74,12 +79,14 @@ async def lifespan(_: FastAPI) -> AsyncIterator[None]:
|
||||
reconcile_task = start_withdraw_reconcile_worker()
|
||||
heartbeat_task = start_heartbeat_monitor()
|
||||
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")
|
||||
|
||||
@@ -101,6 +108,9 @@ if settings.cors_origins_list:
|
||||
allow_headers=["*"],
|
||||
)
|
||||
|
||||
# 接口指标埋点(放在 CORS 之后 = 最外层:测到含 CORS 的完整耗时)。未配置观测时中间件自 no-op。
|
||||
app.add_middleware(RequestMetricsMiddleware)
|
||||
|
||||
|
||||
@app.get("/health", tags=["meta"])
|
||||
def health() -> dict[str, str]:
|
||||
|
||||
@@ -0,0 +1,47 @@
|
||||
# OpenObserve 监控台反代(单独子域名)。前提:
|
||||
# - DNS: observe.shaguabijia.com A 记录 → 本机公网 IP
|
||||
# - 证书: /etc/nginx/ssl/observe.shaguabijia.com.pem|.key(同现有域名放法)
|
||||
# - OpenObserve 只绑 127.0.0.1:5080(见 docker-compose.prod.yml)
|
||||
# 三重防护:IP 白名单 + nginx TLS + OpenObserve 自身登录。
|
||||
# 只你/少数人看的话,更省事的是 SSH 隧道(README「UI 访问」方案 A),不用域名/证书。
|
||||
|
||||
# 80 → 301 → 443
|
||||
server {
|
||||
listen 80;
|
||||
listen [::]:80;
|
||||
server_name observe.shaguabijia.com;
|
||||
return 301 https://$host$request_uri;
|
||||
}
|
||||
|
||||
# 443 SSL 反代到本地 OpenObserve (127.0.0.1:5080)
|
||||
server {
|
||||
listen 443 ssl http2;
|
||||
listen [::]:443 ssl http2;
|
||||
server_name observe.shaguabijia.com;
|
||||
|
||||
ssl_certificate /etc/nginx/ssl/observe.shaguabijia.com.pem;
|
||||
ssl_certificate_key /etc/nginx/ssl/observe.shaguabijia.com.key;
|
||||
|
||||
ssl_protocols TLSv1.2 TLSv1.3;
|
||||
ssl_ciphers HIGH:!aNULL:!MD5;
|
||||
ssl_session_cache shared:SSL:10m;
|
||||
|
||||
# ★ 只放行你的固定出口 IP(办公/家宽)。监控台不必对全网开。
|
||||
allow 1.2.3.4; # ← 换成真实 IP,可多行
|
||||
deny all;
|
||||
|
||||
client_max_body_size 10m;
|
||||
|
||||
location / {
|
||||
proxy_pass http://127.0.0.1:5080;
|
||||
proxy_http_version 1.1;
|
||||
proxy_set_header Host $host;
|
||||
proxy_set_header X-Real-IP $remote_addr;
|
||||
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
|
||||
proxy_set_header X-Forwarded-Proto $scheme;
|
||||
# OpenObserve 有实时/流式,需透传 WebSocket
|
||||
proxy_set_header Upgrade $http_upgrade;
|
||||
proxy_set_header Connection "upgrade";
|
||||
proxy_read_timeout 300s;
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,4 @@
|
||||
# OpenObserve 落盘数据(parquet/索引/元数据),运行时产生,不入库。
|
||||
data/
|
||||
# 生产 compose 的密码文件(OO_ROOT_PASSWORD),含机密,不入库。
|
||||
.env
|
||||
@@ -0,0 +1,126 @@
|
||||
# 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。
|
||||
设计见 [../../docs/superpowers/specs/2026-07-06-openobserve-api-metrics-design.md](../../docs/superpowers/specs/2026-07-06-openobserve-api-metrics-design.md)。
|
||||
|
||||
## 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;该目录已 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(字段:`method` / `route` /
|
||||
`status` / `duration_ms` / `service` / `env`)。
|
||||
|
||||
> 未开 `OBSERVE_ENABLED` 或缺账号密码时,中间件透传、worker 不启动,整套 no-op,不影响业务。
|
||||
|
||||
## 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
|
||||
```
|
||||
|
||||
## 一键导入现成仪表盘(QPS / P95 / 分位 / 错误率)
|
||||
|
||||
备好了 [dashboard-api-metrics.json](dashboard-api-metrics.json),4 个面板:各接口每分钟请求数(QPS 源)、
|
||||
P95 耗时折线、P50/P95/P99 分位表、5xx 错误率表。
|
||||
|
||||
- **UI 导入**:Dashboards → 右上 **Import** → 选该 JSON 文件 → Import(每次导入新建,不覆盖)。
|
||||
- **或 API 导入**:
|
||||
```bash
|
||||
curl -u admin@shaguabijia.local:Complexpass#123 -H 'Content-Type: application/json' \
|
||||
-X POST 'http://localhost:5080/api/default/dashboards?folder=default' \
|
||||
--data-binary @deploy/openobserve/dashboard-api-metrics.json
|
||||
```
|
||||
|
||||
导入后进仪表盘,右上角时间调到「最近 15 分钟 / 1 小时」、开自动刷新即可。低流量下 QPS 面板看「每分钟请求数」比「每秒」直观。
|
||||
|
||||
## 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(单机)+ UI 访问
|
||||
|
||||
前提:app-server 与 OpenObserve **同机**,app→OO 走 localhost(`127.0.0.1:5080`)、不出网、无需 TLS。
|
||||
唯一要防的是**别把 :5080 裸暴露公网**。硬化版编排见 [docker-compose.prod.yml](docker-compose.prod.yml)。
|
||||
|
||||
### 部署步骤
|
||||
|
||||
```bash
|
||||
# 1) 密码文件(本目录,已 gitignore)
|
||||
echo "OO_ROOT_PASSWORD=$(python -c 'import secrets;print(secrets.token_urlsafe(24))')" > deploy/openobserve/.env
|
||||
|
||||
# 2) 起 OpenObserve(只绑 127.0.0.1、命名卷持久化、mem 1g)
|
||||
cd deploy/openobserve && docker compose -f docker-compose.prod.yml up -d
|
||||
sudo systemctl enable docker # 开机自起
|
||||
```
|
||||
|
||||
3) app-server 的 `.env` 打开观测并**重启**(用非 root 的专用 ingest 账号):
|
||||
```dotenv
|
||||
OBSERVE_ENABLED=true
|
||||
OBSERVE_ENDPOINT=http://127.0.0.1:5080
|
||||
OBSERVE_ORG=default
|
||||
OBSERVE_STREAM=app_requests
|
||||
OBSERVE_USER=ingest@shaguabijia.com # UI → Users 建的非 root 账号
|
||||
OBSERVE_PASSWORD=<该账号密码>
|
||||
```
|
||||
```bash
|
||||
sudo systemctl restart shaguabijia-app-server # 日志出现 "observe worker started" 即生效
|
||||
```
|
||||
|
||||
4) 两个必做收口(磁盘/安全):
|
||||
- **保留期**:UI → Streams → `app_requests` → Data Retention 设 14/30 天(一请求一行,不封顶迟早撑爆盘)。
|
||||
- **专用账号**:UI → Users 建非 root 账号给 app 上报,root 只留人工登 UI。
|
||||
|
||||
### UI 访问(二选一)
|
||||
|
||||
**A. SSH 隧道(推荐,零暴露、不用域名/证书):**
|
||||
```bash
|
||||
ssh -L 5080:127.0.0.1:5080 用户@服务器IP
|
||||
# 然后本机浏览器开 http://localhost:5080
|
||||
```
|
||||
|
||||
**B. nginx 子域名反代(要固定 URL / 团队常看):** 见 [../nginx/observe.shaguabijia.com.conf](../nginx/observe.shaguabijia.com.conf)。
|
||||
需 DNS `observe.shaguabijia.com` → 本机 + 证书放 `/etc/nginx/ssl/`;含 IP 白名单 + TLS + WebSocket 透传。
|
||||
|
||||
> ⚠️ prod compose 必须保持 `127.0.0.1:5080:5080`;写成 `5080:5080`(绑 0.0.0.0)= 裸暴露公网,这是唯一真正的坑。
|
||||
@@ -0,0 +1,301 @@
|
||||
{
|
||||
"version": 8,
|
||||
"title": "接口监控 (QPS / 耗时 / 错误率)",
|
||||
"description": "app-server 接口 QPS、P50/P95/P99 耗时、5xx 错误率。数据流 app_requests。",
|
||||
"role": "",
|
||||
"tabs": [
|
||||
{
|
||||
"tabId": "default",
|
||||
"name": "Default",
|
||||
"panels": [
|
||||
{
|
||||
"id": "panel_qps",
|
||||
"type": "line",
|
||||
"title": "各接口 每分钟请求数 (QPS 源)",
|
||||
"description": "",
|
||||
"config": {
|
||||
"show_legends": true,
|
||||
"legends_position": null,
|
||||
"decimals": 2.0,
|
||||
"axis_border_show": false,
|
||||
"base_map": null,
|
||||
"map_view": null
|
||||
},
|
||||
"queryType": "sql",
|
||||
"queries": [
|
||||
{
|
||||
"query": "SELECT histogram(_timestamp, '1 minute') as ts, route, count(*) as reqs FROM app_requests GROUP BY ts, route ORDER BY ts",
|
||||
"vrlFunctionQuery": "",
|
||||
"customQuery": true,
|
||||
"fields": {
|
||||
"stream": "app_requests",
|
||||
"stream_type": "logs",
|
||||
"x": [
|
||||
{
|
||||
"label": "ts",
|
||||
"alias": "ts",
|
||||
"column": "ts",
|
||||
"color": null,
|
||||
"sortBy": "ASC"
|
||||
}
|
||||
],
|
||||
"y": [
|
||||
{
|
||||
"label": "reqs",
|
||||
"alias": "reqs",
|
||||
"column": "reqs",
|
||||
"color": null
|
||||
}
|
||||
],
|
||||
"z": [],
|
||||
"breakdown": [
|
||||
{
|
||||
"label": "route",
|
||||
"alias": "route",
|
||||
"column": "route",
|
||||
"color": null
|
||||
}
|
||||
],
|
||||
"filter": {
|
||||
"filterType": "group",
|
||||
"logicalOperator": "AND",
|
||||
"conditions": []
|
||||
}
|
||||
},
|
||||
"config": {
|
||||
"promql_legend": "",
|
||||
"layer_type": "scatter",
|
||||
"weight_fixed": 1.0
|
||||
}
|
||||
}
|
||||
],
|
||||
"layout": {
|
||||
"x": 0,
|
||||
"y": 0,
|
||||
"w": 24,
|
||||
"h": 9,
|
||||
"i": 1
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "panel_p95",
|
||||
"type": "line",
|
||||
"title": "各接口 P95 耗时 (ms)",
|
||||
"description": "",
|
||||
"config": {
|
||||
"show_legends": true,
|
||||
"legends_position": null,
|
||||
"decimals": 2.0,
|
||||
"axis_border_show": false,
|
||||
"base_map": null,
|
||||
"map_view": null
|
||||
},
|
||||
"queryType": "sql",
|
||||
"queries": [
|
||||
{
|
||||
"query": "SELECT histogram(_timestamp, '1 minute') as ts, route, approx_percentile_cont(duration_ms, 0.95) as p95_ms FROM app_requests GROUP BY ts, route ORDER BY ts",
|
||||
"vrlFunctionQuery": "",
|
||||
"customQuery": true,
|
||||
"fields": {
|
||||
"stream": "app_requests",
|
||||
"stream_type": "logs",
|
||||
"x": [
|
||||
{
|
||||
"label": "ts",
|
||||
"alias": "ts",
|
||||
"column": "ts",
|
||||
"color": null,
|
||||
"sortBy": "ASC"
|
||||
}
|
||||
],
|
||||
"y": [
|
||||
{
|
||||
"label": "p95_ms",
|
||||
"alias": "p95_ms",
|
||||
"column": "p95_ms",
|
||||
"color": null
|
||||
}
|
||||
],
|
||||
"z": [],
|
||||
"breakdown": [
|
||||
{
|
||||
"label": "route",
|
||||
"alias": "route",
|
||||
"column": "route",
|
||||
"color": null
|
||||
}
|
||||
],
|
||||
"filter": {
|
||||
"filterType": "group",
|
||||
"logicalOperator": "AND",
|
||||
"conditions": []
|
||||
}
|
||||
},
|
||||
"config": {
|
||||
"promql_legend": "",
|
||||
"layer_type": "scatter",
|
||||
"weight_fixed": 1.0
|
||||
}
|
||||
}
|
||||
],
|
||||
"layout": {
|
||||
"x": 24,
|
||||
"y": 0,
|
||||
"w": 24,
|
||||
"h": 9,
|
||||
"i": 2
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "panel_pctl",
|
||||
"type": "table",
|
||||
"title": "各接口 耗时分位 P50/P95/P99 (ms)",
|
||||
"description": "",
|
||||
"config": {
|
||||
"show_legends": true,
|
||||
"legends_position": null,
|
||||
"decimals": 2.0,
|
||||
"axis_border_show": false,
|
||||
"base_map": null,
|
||||
"map_view": null
|
||||
},
|
||||
"queryType": "sql",
|
||||
"queries": [
|
||||
{
|
||||
"query": "SELECT route, approx_percentile_cont(duration_ms,0.5) as p50, approx_percentile_cont(duration_ms,0.95) as p95, approx_percentile_cont(duration_ms,0.99) as p99, count(*) as cnt FROM app_requests GROUP BY route ORDER BY p95 DESC",
|
||||
"vrlFunctionQuery": "",
|
||||
"customQuery": true,
|
||||
"fields": {
|
||||
"stream": "app_requests",
|
||||
"stream_type": "logs",
|
||||
"x": [
|
||||
{
|
||||
"label": "route",
|
||||
"alias": "route",
|
||||
"column": "route",
|
||||
"color": null
|
||||
}
|
||||
],
|
||||
"y": [
|
||||
{
|
||||
"label": "p50",
|
||||
"alias": "p50",
|
||||
"column": "p50",
|
||||
"color": null
|
||||
},
|
||||
{
|
||||
"label": "p95",
|
||||
"alias": "p95",
|
||||
"column": "p95",
|
||||
"color": null
|
||||
},
|
||||
{
|
||||
"label": "p99",
|
||||
"alias": "p99",
|
||||
"column": "p99",
|
||||
"color": null
|
||||
},
|
||||
{
|
||||
"label": "cnt",
|
||||
"alias": "cnt",
|
||||
"column": "cnt",
|
||||
"color": null
|
||||
}
|
||||
],
|
||||
"z": [],
|
||||
"breakdown": [],
|
||||
"filter": {
|
||||
"filterType": "group",
|
||||
"logicalOperator": "AND",
|
||||
"conditions": []
|
||||
}
|
||||
},
|
||||
"config": {
|
||||
"promql_legend": "",
|
||||
"layer_type": "scatter",
|
||||
"weight_fixed": 1.0
|
||||
}
|
||||
}
|
||||
],
|
||||
"layout": {
|
||||
"x": 0,
|
||||
"y": 9,
|
||||
"w": 24,
|
||||
"h": 9,
|
||||
"i": 3
|
||||
}
|
||||
},
|
||||
{
|
||||
"id": "panel_err",
|
||||
"type": "table",
|
||||
"title": "各接口 错误率 (5xx %)",
|
||||
"description": "",
|
||||
"config": {
|
||||
"show_legends": true,
|
||||
"legends_position": null,
|
||||
"decimals": 2.0,
|
||||
"axis_border_show": false,
|
||||
"base_map": null,
|
||||
"map_view": null
|
||||
},
|
||||
"queryType": "sql",
|
||||
"queries": [
|
||||
{
|
||||
"query": "SELECT route, count(*) FILTER (WHERE status >= 500) * 100.0 / count(*) as err_pct, count(*) as cnt FROM app_requests GROUP BY route ORDER BY err_pct DESC",
|
||||
"vrlFunctionQuery": "",
|
||||
"customQuery": true,
|
||||
"fields": {
|
||||
"stream": "app_requests",
|
||||
"stream_type": "logs",
|
||||
"x": [
|
||||
{
|
||||
"label": "route",
|
||||
"alias": "route",
|
||||
"column": "route",
|
||||
"color": null
|
||||
}
|
||||
],
|
||||
"y": [
|
||||
{
|
||||
"label": "err_pct",
|
||||
"alias": "err_pct",
|
||||
"column": "err_pct",
|
||||
"color": null
|
||||
},
|
||||
{
|
||||
"label": "cnt",
|
||||
"alias": "cnt",
|
||||
"column": "cnt",
|
||||
"color": null
|
||||
}
|
||||
],
|
||||
"z": [],
|
||||
"breakdown": [],
|
||||
"filter": {
|
||||
"filterType": "group",
|
||||
"logicalOperator": "AND",
|
||||
"conditions": []
|
||||
}
|
||||
},
|
||||
"config": {
|
||||
"promql_legend": "",
|
||||
"layer_type": "scatter",
|
||||
"weight_fixed": 1.0
|
||||
}
|
||||
}
|
||||
],
|
||||
"layout": {
|
||||
"x": 24,
|
||||
"y": 9,
|
||||
"w": 24,
|
||||
"h": 9,
|
||||
"i": 4
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
],
|
||||
"variables": {
|
||||
"list": []
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,27 @@
|
||||
# 生产用 OpenObserve(单机)。相对本地版 docker-compose.yml 的区别:
|
||||
# - 端口只绑 127.0.0.1 → 公网/外网都到不了(UI 访问走 SSH 隧道或 nginx 反代,见 README)
|
||||
# - root 密码走环境变量(放同目录 .env,已 gitignore,勿提交)
|
||||
# - 数据用 docker 命名卷(免 bind-mount 权限坑)+ 内存上限(与 app/PG 共存防抢内存)
|
||||
#
|
||||
# 用法:
|
||||
# 1) 本目录建 .env(已 gitignore):
|
||||
# OO_ROOT_PASSWORD=<强随机串> # 生成: python -c "import secrets;print(secrets.token_urlsafe(24))"
|
||||
# 2) docker compose -f docker-compose.prod.yml up -d
|
||||
# 3) 开机自起: sudo systemctl enable docker
|
||||
services:
|
||||
openobserve:
|
||||
image: public.ecr.aws/zinclabs/openobserve:latest
|
||||
container_name: openobserve
|
||||
ports:
|
||||
- "127.0.0.1:5080:5080" # ★只绑本机;写成 5080:5080 = 裸暴露公网,别这么干
|
||||
environment:
|
||||
ZO_ROOT_USER_EMAIL: "admin@shaguabijia.com"
|
||||
ZO_ROOT_USER_PASSWORD: "${OO_ROOT_PASSWORD:?请先在 deploy/openobserve/.env 里设 OO_ROOT_PASSWORD}"
|
||||
ZO_DATA_DIR: "/data"
|
||||
volumes:
|
||||
- openobserve_data:/data # docker 命名卷;想固定到某块盘改成 bind: /your/disk:/data
|
||||
restart: unless-stopped
|
||||
mem_limit: 1g # 按机器内存调;查询重可给 2g
|
||||
|
||||
volumes:
|
||||
openobserve_data:
|
||||
@@ -0,0 +1,16 @@
|
||||
# 本地开发用 OpenObserve(单容器 = local 模式)。用于接收 app-server 的接口指标(QPS/耗时/错误率)。
|
||||
# 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
|
||||
@@ -0,0 +1,897 @@
|
||||
# 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](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) -> None`
|
||||
- `app.core.observe.take_dropped() -> int`
|
||||
- `app.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 | None`
|
||||
- `app.core.observe_worker.stop_observe_worker(task) -> None`
|
||||
- `settings.observe_configured -> bool`
|
||||
|
||||
---
|
||||
|
||||
## Task 1: 配置项 `OBSERVE_*` + `observe_configured`
|
||||
|
||||
**Files:**
|
||||
- Modify: `app/core/config.py`(在 `cors_origins_list` property 之后、`is_prod` property 之前插入)
|
||||
- Test: `tests/test_observe.py`(新建)
|
||||
|
||||
- [ ] **Step 1: 写失败测试**
|
||||
|
||||
新建 `tests/test_observe.py`:
|
||||
|
||||
```python
|
||||
"""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:
|
||||
|
||||
```python
|
||||
# ===== 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**
|
||||
|
||||
```bash
|
||||
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:
|
||||
|
||||
```python
|
||||
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`:
|
||||
|
||||
```python
|
||||
"""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**
|
||||
|
||||
```bash
|
||||
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:
|
||||
|
||||
```python
|
||||
import pytest
|
||||
from fastapi import FastAPI
|
||||
from fastapi.testclient import TestClient
|
||||
```
|
||||
|
||||
并追加:
|
||||
|
||||
```python
|
||||
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
|
||||
|
||||
```python
|
||||
from __future__ import annotations
|
||||
|
||||
import asyncio
|
||||
|
||||
from app.core.config import settings
|
||||
```
|
||||
|
||||
改成
|
||||
|
||||
```python
|
||||
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)**
|
||||
|
||||
```python
|
||||
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**
|
||||
|
||||
```bash
|
||||
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:
|
||||
|
||||
```python
|
||||
import httpx
|
||||
from app.core import observe_worker
|
||||
```
|
||||
|
||||
并追加:
|
||||
|
||||
```python
|
||||
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`:
|
||||
|
||||
```python
|
||||
"""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**
|
||||
|
||||
```bash
|
||||
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:
|
||||
|
||||
```python
|
||||
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:
|
||||
|
||||
```python
|
||||
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):
|
||||
|
||||
```python
|
||||
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):
|
||||
|
||||
```python
|
||||
# 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**
|
||||
|
||||
```bash
|
||||
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`:
|
||||
|
||||
```yaml
|
||||
# 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`:
|
||||
|
||||
````markdown
|
||||
# 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:
|
||||
|
||||
```dotenv
|
||||
|
||||
# ===== 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**
|
||||
|
||||
```bash
|
||||
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)**
|
||||
|
||||
```bash
|
||||
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。
|
||||
@@ -0,0 +1,236 @@
|
||||
# 接口 QPS + 耗时可观测(OpenObserve)设计
|
||||
|
||||
- **日期**:2026-07-06
|
||||
- **状态**:已评审通过,待写实现计划
|
||||
- **范围**:仅 app-server(8770);admin(8771)暂不接入
|
||||
- **方案**:A —— 轻量自研 ASGI 中间件 + 后台 worker 批量直采到 OpenObserve
|
||||
|
||||
## 1. 背景与目标
|
||||
|
||||
app-server 目前除 CORS 外无任何中间件,也无接口级可观测。需要按**每个接口**采集:
|
||||
|
||||
- **QPS**(每秒请求数,可按接口/时间分桶)
|
||||
- **耗时**(P50/P95/P99 等分位)
|
||||
|
||||
顺带低成本拿到**错误率**(`status >= 500` 占比)。落地目标是:本地 Docker 跑一个 OpenObserve 实例接收数据,服务侧加埋点上报,在 OpenObserve 仪表盘上看各接口 QPS + 耗时。
|
||||
|
||||
### 非目标(YAGNI)
|
||||
|
||||
- 不做分布式 trace / span 关联(只要接口聚合指标)。
|
||||
- 不引入 OpenTelemetry / Prometheus 客户端等重依赖。
|
||||
- 不采集请求体 / query / 用户身份等,任何 PII 都不进上报。
|
||||
- admin(8771)本期不接(中间件写成可复用,未来一行挂载即可)。
|
||||
- 上报失败不做持久化重试 / 落盘补偿(best-effort)。
|
||||
|
||||
## 2. 方案选型
|
||||
|
||||
对比过三条路(详见评审记录):
|
||||
|
||||
- **A 轻量自研中间件 + JSON 直采**(选中):零新依赖(`httpx` 已在依赖里),完全贴合本仓库「后台 worker + JSON 事件 + `*_configured` 优雅降级」的既有习惯,恰好满足「每接口 QPS + 耗时 + 错误率」并保留原始事件下钻能力。
|
||||
- B OpenTelemetry 自动埋点 + OTLP:行业标准、顺带 trace,但多 5–6 个依赖、概念多、数据量/成本高于需求,与精简代码库风格相悖。
|
||||
- C Prometheus 进程内聚合 + remote_write/抓取:数据量最小,但 remote_write 编码复杂或需额外抓取进程,丢失单请求下钻,最不贴合 OpenObserve 的 log-first 强项。
|
||||
|
||||
**结论:A。**
|
||||
|
||||
## 3. 架构与数据流
|
||||
|
||||
```
|
||||
每个 HTTP 请求
|
||||
→ RequestMetricsMiddleware(最外层:测总耗时 / 抓路由模板 + 状态码)
|
||||
→ record_event() 非阻塞入队(有界队列,满则丢最旧,绝不阻塞、绝不 OOM)
|
||||
→ observe_worker(后台 asyncio.Task,随 lifespan 启停)批量 drain
|
||||
→ httpx POST {ENDPOINT}/api/{ORG}/{STREAM}/_json → OpenObserve
|
||||
→ 仪表盘 SQL 聚合出 QPS / 分位耗时 / 错误率
|
||||
```
|
||||
|
||||
**核心不变量**:
|
||||
|
||||
1. 请求路径上只做「测时 + 构建一个小 dict + `put_nowait`」,**无任何网络/磁盘 I/O**。
|
||||
2. 所有上报 I/O 在后台 worker;worker 捕获全部异常,绝不让埋点影响请求。
|
||||
3. 未配置观测(`observe_configured=False`)→ 中间件透传、worker 不启动,整套 no-op。
|
||||
4. OpenObserve 不可用 → 队列填满后丢弃事件 + 限流告警,业务零影响。
|
||||
|
||||
## 4. 组件设计
|
||||
|
||||
### 4.1 OpenObserve 本地部署 —— `deploy/openobserve/docker-compose.yml`(新增)
|
||||
|
||||
```yaml
|
||||
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
|
||||
```
|
||||
|
||||
- `docker compose up -d` 启动;Web UI `http://localhost:5080`,用上面邮箱/密码登录。
|
||||
- 单容器 = local 模式,数据落 `./data`(已挂卷持久化)。
|
||||
- **stream 首次上报自动创建**,无需预建 `app_requests`。
|
||||
- 上报鉴权:HTTP Basic auth(`email:password`),本地直接用 root 账号;生产应另建仅具 ingest 权限的用户/服务账号(本期不涉及)。
|
||||
|
||||
### 4.2 事件 schema(一请求一行 JSON)
|
||||
|
||||
```json
|
||||
{
|
||||
"_timestamp": 1720000000000000, // 微秒(µs)整数,请求完成时刻。OpenObserve 默认时间列 _timestamp 以微秒计
|
||||
"service": "app-server", // 取 LOG_SERVICE_NAME / 固定值
|
||||
"env": "dev", // settings.APP_ENV
|
||||
"method": "POST",
|
||||
"route": "/api/v1/coupon/step", // 路由模板(非实际 path)
|
||||
"status": 200,
|
||||
"duration_ms": 42.7 // float 毫秒
|
||||
}
|
||||
```
|
||||
|
||||
- **只存路由模板**(如 `/c/{code}`、`/media` 静态归一),避免 path 参数把维度打爆。
|
||||
- 未匹配路由(404 / 扫描器)归一到常量 `__unmatched__`。
|
||||
- 只采 method / route / status / duration —— 无 body、无 query、无 PII。
|
||||
|
||||
### 4.3 埋点中间件 —— `app/core/observe.py`(新增)
|
||||
|
||||
**纯 ASGI 中间件**(比 `BaseHTTPMiddleware` 开销低;能可靠读到路由与最终状态码;scope 按引用透传,内层 router 的 `scope["route"]` 外层可见)。
|
||||
|
||||
职责:
|
||||
|
||||
1. 非 `http` 请求、或 `not settings.observe_configured` → 直接透传,不测。
|
||||
2. `perf_counter()` 记起点;包一层 `send` 抓 `http.response.start` 的 `status`(默认兜底 500,覆盖下游抛异常未产出 response 的情况)。
|
||||
3. `finally` 里算 `duration_ms`,从 `scope` 取路由模板(见下),构建事件,调 `record_event()`。
|
||||
4. 跳过路径集合 `_SKIP_PATHS = {"/health"}`(纯噪音)。
|
||||
|
||||
**路由模板解析(跨 Starlette 版本稳健)**:
|
||||
|
||||
```python
|
||||
route = scope.get("route")
|
||||
template = getattr(route, "path", None)
|
||||
if template is None: # 未匹配 / 老版本未写 scope["route"]
|
||||
template = "__unmatched__"
|
||||
```
|
||||
|
||||
(若实测某 Starlette 版本不写 `scope["route"]`,回退用 `request.app.router.routes` 逐个 `route.matches(scope)==Match.FULL` 找模板;实现时以实际版本为准,优先 `scope["route"]`。)
|
||||
|
||||
**入队(`record_event`)**:模块级 `asyncio.Queue(maxsize=OBSERVE_QUEUE_MAX)`。用 `put_nowait`,`QueueFull` 则丢弃并累加一个 `_dropped` 计数(每累计 N 条限流打一条 WARNING)。**永不 `await put()`、永不阻塞请求**。
|
||||
|
||||
> 决策(a):队列满 → **丢弃**(不阻塞请求)。
|
||||
|
||||
### 4.4 上报 worker —— `app/core/observe_worker.py`(新增)
|
||||
|
||||
对齐现有 `heartbeat_monitor_worker.py` / `daily_exchange_worker.py` / `withdraw_reconcile_worker.py` 的 `start_*` / `stop_*` 形态。
|
||||
|
||||
- `start_observe_worker() -> asyncio.Task | None`
|
||||
- `not settings.observe_configured` → 返回 `None`(no-op)。
|
||||
- 否则建专用 `httpx.AsyncClient`(`base_url=ENDPOINT`,`auth=(USER, PASSWORD)`,`timeout=OBSERVE_TIMEOUT_SEC`),起 `_run_loop` task。
|
||||
- `_run_loop()`:循环
|
||||
1. `_collect_batch()`:`await asyncio.wait_for(queue.get(), timeout=FLUSH_INTERVAL)` 拿到首条(超时且空 → 返回空,continue);再 `get_nowait()` 连抽到 `BATCH_MAX` 条或抽空。
|
||||
2. `POST /api/{ORG}/{STREAM}/_json`,body 为事件数组。
|
||||
3. **catch 所有异常**:失败限流打 WARNING,**直接丢弃该批,不重试**。
|
||||
- `stop_observe_worker(task)`:best-effort 收尾 flush(短超时)→ `task.cancel()` → `await`(吞 `CancelledError`)→ 关 client。
|
||||
|
||||
> 决策(b):上报失败 → **直接丢弃,不重试**(best-effort 遥测)。
|
||||
|
||||
### 4.5 配置 —— `app/core/config.py`(改)
|
||||
|
||||
新增一段 `# ===== 可观测(OpenObserve 接口指标)=====`,默认全关(prod 安全):
|
||||
|
||||
| 配置 | 默认 | 说明 |
|
||||
|---|---|---|
|
||||
| `OBSERVE_ENABLED` | `False` | 总开关;默认关,opt-in |
|
||||
| `OBSERVE_ENDPOINT` | `http://localhost:5080` | OpenObserve base URL |
|
||||
| `OBSERVE_ORG` | `default` | 组织名 |
|
||||
| `OBSERVE_STREAM` | `app_requests` | stream 名 |
|
||||
| `OBSERVE_USER` | `""` | Basic auth 邮箱 |
|
||||
| `OBSERVE_PASSWORD` | `""` | Basic auth 密码/token |
|
||||
| `OBSERVE_FLUSH_INTERVAL_SEC` | `5.0` | worker 最长攒批间隔 |
|
||||
| `OBSERVE_BATCH_MAX` | `200` | 单批最大事件数 |
|
||||
| `OBSERVE_QUEUE_MAX` | `10000` | 有界队列上限,满则丢 |
|
||||
| `OBSERVE_TIMEOUT_SEC` | `5.0` | 上报 HTTP 超时 |
|
||||
|
||||
```python
|
||||
@property
|
||||
def observe_configured(self) -> bool:
|
||||
return bool(self.OBSERVE_ENABLED and self.OBSERVE_ENDPOINT
|
||||
and self.OBSERVE_USER and self.OBSERVE_PASSWORD)
|
||||
```
|
||||
|
||||
`.env.example` 同步补一段带注释的 `OBSERVE_*`(沿用该文件重注释风格),`OBSERVE_ENABLED=false`。
|
||||
|
||||
### 4.6 接线 —— `app/main.py`(改)
|
||||
|
||||
- import `RequestMetricsMiddleware`、`start_observe_worker` / `stop_observe_worker`。
|
||||
- `app.add_middleware(RequestMetricsMiddleware)`:放在 CORS `add_middleware` **之后** → 成为最外层,测到含 CORS 的完整耗时。无条件挂载(内部自 no-op)。
|
||||
- `lifespan`:启动 `observe_task = start_observe_worker()`;`finally` 里 `await stop_observe_worker(observe_task)`,与现有 worker 并列。
|
||||
|
||||
### 4.7 OpenObserve 查询 / 仪表盘 —— `deploy/openobserve/README.md`(新增)
|
||||
|
||||
含:compose 启停、登录、stream 自动创建说明、`.env` 接线,以及可直接粘的示例 SQL:
|
||||
|
||||
- **各接口 QPS**(1 分钟分桶):
|
||||
```sql
|
||||
SELECT route, histogram(_timestamp, '1 minute') AS ts, count(*) AS cnt
|
||||
FROM app_requests GROUP BY route, ts ORDER BY ts
|
||||
```
|
||||
(面板按 `cnt/60` 展示每秒;或用 OpenObserve 图表的 rate 能力。)
|
||||
- **各接口 P95 耗时**:
|
||||
```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**:
|
||||
```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
|
||||
```
|
||||
|
||||
## 5. 关键设计决策汇总
|
||||
|
||||
- **(a) 队列满 → 丢弃**(不阻塞请求):遥测让路于业务可用性。
|
||||
- **(b) 上报失败 → 不重试**:best-effort;避免 poison batch 堆积与队列无限增长。
|
||||
- **(c) 跳过 `/health`**:健康检查是纯噪音,硬编码在 `_SKIP_PATHS`。
|
||||
- **只存路由模板 + `__unmatched__`**:防维度爆炸。
|
||||
- **默认 OFF、opt-in**:prod 安全默认;开启后仍全异步 + 有界。
|
||||
- **纯 ASGI 中间件 + `perf_counter`**:请求路径开销微秒级,无 I/O。
|
||||
|
||||
## 6. 安全 / 性能保证
|
||||
|
||||
- 请求路径新增开销 ≈ 一次 `perf_counter` 差 + 一个小 dict + 一次 `put_nowait`(微秒级),无锁竞争的显著热点。
|
||||
- 失败隔离:入队丢弃 + worker 全异常捕获;OpenObserve 宕机不影响任何请求。
|
||||
- 有界内存:队列 `maxsize` 封顶,最坏丢事件不涨内存。
|
||||
- 无 PII:仅 method / route / status / duration。
|
||||
|
||||
## 7. 测试策略 —— `tests/test_observe.py`(新增)
|
||||
|
||||
沿用仓库约定(`TestClient` + `monkeypatch`,绝不打真网络;`conftest` 在 import 前设 env):
|
||||
|
||||
1. 埋点入队字段正确:模板路由、`status`、`duration_ms > 0`。
|
||||
2. 参数化路由 → 取到**模板**而非实际 path。
|
||||
3. 未匹配路径(404)→ `route == "__unmatched__"`。
|
||||
4. `OBSERVE_ENABLED=false` → 零入队、零 HTTP(现有测试不受影响)。
|
||||
5. 队列满 → `record_event` 不抛异常(走丢弃分支)。
|
||||
6. worker 批量 POST 的 URL / payload 正确(monkeypatch httpx client / `_post`,不打网络)。
|
||||
7. `/health` 被跳过 → 不入队。
|
||||
|
||||
> `settings` 是 `lru_cache` 单例;需要开启观测的用例通过 monkeypatch `settings` 属性或直接调 `record_event` / 中间件并 patch `observe_configured` 实现,避免全局 env 改动波及他用例。
|
||||
|
||||
## 8. 文件清单
|
||||
|
||||
| 文件 | 动作 |
|
||||
|---|---|
|
||||
| `deploy/openobserve/docker-compose.yml` | 新增(OpenObserve 容器)|
|
||||
| `deploy/openobserve/README.md` | 新增(部署步骤 + 查询/仪表盘)|
|
||||
| `app/core/observe.py` | 新增(中间件 + 有界队列 + `record_event` + 路由解析)|
|
||||
| `app/core/observe_worker.py` | 新增(后台批量上报 worker)|
|
||||
| `app/core/config.py` | 改(`OBSERVE_*` + `observe_configured`)|
|
||||
| `app/main.py` | 改(挂中间件 + lifespan 启停 worker)|
|
||||
| `.env.example` | 改(新增 `OBSERVE_*` 注释段)|
|
||||
| `tests/test_observe.py` | 新增 |
|
||||
|
||||
## 9. 未来工作(本期不做)
|
||||
|
||||
- admin(8771)接入同一套中间件(`service` 字段区分)。
|
||||
- 生产部署 OpenObserve(持久化、独立 ingest 账号、资源规格、鉴权收紧)。
|
||||
- 上报字段扩展(如按 user/设备维度、上游 pricebot 透传耗时拆分)。
|
||||
@@ -0,0 +1,234 @@
|
||||
"""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
|
||||
|
||||
import asyncio
|
||||
|
||||
import httpx
|
||||
import pytest
|
||||
from fastapi import FastAPI
|
||||
from fastapi.testclient import TestClient
|
||||
|
||||
from app.core import observe, observe_worker
|
||||
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
|
||||
|
||||
# 缺用户名 → False
|
||||
monkeypatch.setattr(settings, "OBSERVE_PASSWORD", "p")
|
||||
monkeypatch.setattr(settings, "OBSERVE_USER", "")
|
||||
assert settings.observe_configured is False
|
||||
|
||||
# 开关关 → False(即便凭证齐全)
|
||||
monkeypatch.setattr(settings, "OBSERVE_USER", "u")
|
||||
monkeypatch.setattr(settings, "OBSERVE_ENABLED", False)
|
||||
assert settings.observe_configured is False
|
||||
|
||||
|
||||
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
|
||||
|
||||
|
||||
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
|
||||
|
||||
|
||||
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
|
||||
|
||||
|
||||
async def test_run_loop_survives_post_failure(monkeypatch):
|
||||
"""_post_batch 抛异常时,loop 不崩溃、继续处理后续批次(best-effort 契约)。"""
|
||||
q = asyncio.Queue(maxsize=100)
|
||||
monkeypatch.setattr(observe, "_queue", q)
|
||||
monkeypatch.setattr(settings, "OBSERVE_FLUSH_INTERVAL_SEC", 0.02)
|
||||
monkeypatch.setattr(settings, "OBSERVE_BATCH_MAX", 200)
|
||||
seen: list[list[int]] = []
|
||||
|
||||
async def boom(client, batch):
|
||||
seen.append([e["n"] for e in batch])
|
||||
raise RuntimeError("boom")
|
||||
|
||||
monkeypatch.setattr(observe_worker, "_post_batch", boom)
|
||||
|
||||
q.put_nowait({"n": 1})
|
||||
task = asyncio.create_task(observe_worker._run_loop(None))
|
||||
try:
|
||||
for _ in range(50): # 轮询直到第 1 批被处理(失败),最多等 0.5s
|
||||
await asyncio.sleep(0.01)
|
||||
if seen:
|
||||
break
|
||||
q.put_nowait({"n": 2})
|
||||
for _ in range(50): # 第 2 批被处理 → 证明失败后 loop 仍存活
|
||||
await asyncio.sleep(0.01)
|
||||
if len(seen) >= 2:
|
||||
break
|
||||
finally:
|
||||
task.cancel()
|
||||
try:
|
||||
await task
|
||||
except asyncio.CancelledError:
|
||||
pass
|
||||
assert seen == [[1], [2]]
|
||||
|
||||
|
||||
async def test_stop_flushes_remaining_and_closes_client(monkeypatch):
|
||||
"""stop:cancel 后把剩余事件 best-effort 发出最后一批,并关闭 + 置空 client。"""
|
||||
q = asyncio.Queue(maxsize=100)
|
||||
monkeypatch.setattr(observe, "_queue", q)
|
||||
monkeypatch.setattr(settings, "OBSERVE_ORG", "default")
|
||||
monkeypatch.setattr(settings, "OBSERVE_STREAM", "app_requests")
|
||||
monkeypatch.setattr(settings, "OBSERVE_BATCH_MAX", 200)
|
||||
q.put_nowait({"n": 1})
|
||||
q.put_nowait({"n": 2})
|
||||
posted: dict = {}
|
||||
|
||||
def handler(request: httpx.Request) -> httpx.Response:
|
||||
posted["body"] = request.content
|
||||
return httpx.Response(200, json={"code": 200})
|
||||
|
||||
client = httpx.AsyncClient(
|
||||
base_url="http://oo", transport=httpx.MockTransport(handler)
|
||||
)
|
||||
monkeypatch.setattr(observe_worker, "_client", client)
|
||||
|
||||
async def _noop() -> None:
|
||||
return None
|
||||
|
||||
task = asyncio.create_task(_noop())
|
||||
await observe_worker.stop_observe_worker(task)
|
||||
|
||||
assert b'"n"' in posted["body"] # 关停时把剩余事件发了出去
|
||||
assert observe_worker._client is None # client 已关闭并置空
|
||||
|
||||
|
||||
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
|
||||
Reference in New Issue
Block a user