feat(calllog): 调用日志落地(SQLite / 90 天 / 管理员可查全部)

- app/calllog.py:SQLite 表 api_calls(21 列,含提示词与回复)+ 异步批量写 + 双向握手
  (流式请求的访问日志条目比响应晚到也能落库)+ 早失败补记 + 失败也记 Key 前 8 位
- 新增 4 个仅管理员端点:/logs/calls、/logs/calls/stats、/logs/calls/{id}、/logs/calls/export
- 前端:日志页新增「实时 / 调用日志」页签,新组件 CallLogs.vue(筛选/分页/详情/CSV 导出)
- 保留 90 天;队列满丢弃计数,不反压请求;零新依赖
- docs:新增 docs/call-log-design.md(方案对比与三个关键设计)
- 版本 2.7.3 → 2.8.0
This commit is contained in:
jidekin committed 2026-10-23 10:02:07 +08:00
1 parent 19224f1993
commit 2c53504ad4
10 files changed
+1098 -1

No files matched your search

+5
View File
@@ -58,3 +58,8 @@ frontend/dist/
dist/ dist/
*.egg-info/ *.egg-info/
.DS_Store .DS_Store
# 调用日志数据库(2026-10-23 落地方案):运行时生成,不入库
data/calllog.db
data/calllog.db-wal
data/calllog.db-shm
+42
View File
@@ -1,5 +1,47 @@
# CHANGELOG # CHANGELOG
## 2026-10-23 · 调用日志落地(SQLite / 90 天 / 仅管理员可查全部)(2.7.3 → 2.8.0)
老板:「ai网关项目,现在我想把调用日志落地,有什么方案」→ 方案确认后:「1,要记 / 2,90天 /
3,要前端查询界面,然后这个查询界面集成在日志菜单里,只有管理员有这个权限查看全部人的日志」。
### 后端:`app/calllog.py`(SQLite,零新依赖)
- 表 `api_calls` **21 列**:时间、rid、账号、Key 前 8 位、类型、方法、路径、模型、实际模型、
供应商、是否流式、成功、状态码、耗时、IP、UA、**提示词**、**回复**、网关版本。
- **异步批量写**:`AccessLog.add()` → 内存队列 → 后台 task 每 1 秒或攒满 200 条 `executemany` 提交;
队列满(默认 1 万条)即丢弃并计数,**绝不反压请求**;进程退出前 flush。
- **双向握手**(本轮最关键的设计):一条日志的两半来自不同位置 —— 访问日志条目(`AccessLog.add`)
和响应元数据(中间件收尾,含状态码/耗时/IP/UA)。两半先后顺序不确定:
非流式是「条目先、元数据后」,**流式是「元数据先、条目后」**(SSE 真正发完才 add)。
所以两边都先暂存,谁后到谁拼成一行落库。
- **早失败也记**:未知模型 400、鉴权 401/403、限流 429 这类请求走不到 `AccessLog.add`,
中间件用请求体补一条(含模型名与提示词),否则日志里只剩一行"400、无模型",排查时没有信息量。
- **失败也记 Key 前缀**:401 时 `request.state.credential` 是空的,但恰恰这时最需要知道
客户端拿的是哪把 Key —— 记前 8 位即可对账(2026-10-23 的真实工单就是靠这个抓到
「客户端在发 `wjd-5Da_…`,而库里他的有效 Key 是 `wjd-15zs…`」)。
- 保留 **90 天**:启动时清理 + 每 6 小时清理一次;`CALLLOG_RETENTION_DAYS` 可调。
- 隐私:**API Key 只存前 8 位**;提示词/回复按 `CALLLOG_CONTENT_MAX`(默认 8000 字)截断;
库文件权限 600。IP 存完整地址(仅管理员可见,便于与 WAF/防火墙日志对账)。
### 接口(新增 4 个,全部**仅管理员**)
| 端点 | 用途 |
|---|---|
| `GET /logs/calls` | 分页 + 筛选(时间段/账号/模型/成功失败/状态码/类型/关键词),`content=1` 时带内容 |
| `GET /logs/calls/stats` | 按天、按模型、按账号、按状态码聚合 + 库大小/写入/丢弃计数 |
| `GET /logs/calls/{id}` | 单条详情(含完整提示词与回复)|
| `GET /logs/calls/export` | CSV 导出(最多 5 万条,UTF-8 BOM,Excel 直接打开)|
成员访问以上端点一律 **403**(`_require_admin_logs`)。
### 前端:集成在「日志」菜单
- 日志页顶部新增两个页签:**实时**(原有内存环形缓冲)/ **调用日志**(仅管理员可见)。
- 新组件 `CallLogs.vue`:日期范围、账号、模型、仅成功/仅失败、关键词(内容/rid/IP/UA)筛选;
列表显示时间/账号/模型/耗时/状态/流式/UA;点击行弹出详情看**完整提示词与回复**;CSV 导出。
### 现状说明
- `/metrics` 仍是内存态(重启清零),本次没动它;**调用日志是持久化的**,
需要历史统计时可以基于 `api_calls` 做(二期可把 Dashboard 数据源切过来)。
## 2026-10-22 · 成员只看自己的用量 + AI 接口不再被 WAF 误杀(2.7.2 → 2.7.3) ## 2026-10-22 · 成员只看自己的用量 + AI 接口不再被 WAF 误杀(2.7.2 → 2.7.3)
老板:「改,替换重启,更新文档,推送 gitea」;随后补充「成员在首页是要看到自己调用的次数统计」。 老板:「改,替换重启,更新文档,推送 gitea」;随后补充「成员在首页是要看到自己调用的次数统计」。
+480
View File
@@ -0,0 +1,480 @@
"""调用日志落地(SQLite 持久化,默认保留 90 天)。
为什么需要它
------------
`AccessLog` 是**内存**环形缓冲(重启即丢),`/metrics` 只有聚合计数。出工单时最常被问的是
「谁、什么时候、用哪个模型、失败原因是什么、当时发了什么内容」—— 这些必须落盘可查。
设计取舍
--------
- **零新依赖**:`sqlite3` 是标准库,不引入 ORM / 数据库驱动(与本项目"依赖极简"的基调一致)。
- **不阻塞请求**:`stage()` 只往内存队列 append;后台 task 每 1 秒或攒满 `BATCH` 条批量写库。
队列满(默认 10000 条)即丢弃并计数,**绝不反压请求**。
- **隐私**:API Key 只存前 8 位(`key_id`);提示词/回复按 `CALLLOG_CONTENT_MAX` 截断(默认 8000 字)。
- **保留**:默认 90 天,启动时清理一次 + 每 6 小时清理一次(不依赖外部 cron)。
- **只增不改**:日志表不做 UPDATE;清理只按 `ts` 删。
环境变量
--------
- `CALLLOG_DB` 默认 `data/calllog.db`
- `CALLLOG_RETENTION_DAYS` 默认 90
- `CALLLOG_QUEUE_MAX` 默认 10000
- `CALLLOG_CONTENT_MAX` 默认 8000(提示词/回复各自截断长度)
- `CALLLOG_DISABLED=1` 整体关闭(不回退任何现有行为)
"""
from __future__ import annotations
import asyncio
import logging
import os
import sqlite3
import threading
import time
from collections import deque
from typing import Any
log = logging.getLogger("calllog")
_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
DB_PATH = os.environ.get("CALLLOG_DB") or os.path.join(_ROOT, "data", "calllog.db")
RETENTION_DAYS = int(os.environ.get("CALLLOG_RETENTION_DAYS") or 90)
QUEUE_MAX = int(os.environ.get("CALLLOG_QUEUE_MAX") or 10000)
CONTENT_MAX = int(os.environ.get("CALLLOG_CONTENT_MAX") or 8000)
DISABLED = os.environ.get("CALLLOG_DISABLED", "") not in ("", "0", "false", "False")
BATCH = 200 # 攒够多少条提交一次
FLUSH_INTERVAL = 1.0 # 最长多久提交一次(秒)
CLEAN_INTERVAL = 6 * 3600 # 清理间隔(秒)
SCHEMA = """
CREATE TABLE IF NOT EXISTS api_calls (
id INTEGER PRIMARY KEY AUTOINCREMENT,
ts REAL NOT NULL,
rid TEXT,
account TEXT,
key_id TEXT,
kind TEXT,
method TEXT,
path TEXT,
model TEXT,
served TEXT,
provider TEXT,
caller TEXT,
stream INTEGER DEFAULT 0,
ok INTEGER DEFAULT 1,
status INTEGER,
latency_ms INTEGER,
ip TEXT,
ua TEXT,
prompt TEXT,
completion TEXT,
ver TEXT
);
CREATE INDEX IF NOT EXISTS idx_calls_ts ON api_calls(ts);
CREATE INDEX IF NOT EXISTS idx_calls_acc_ts ON api_calls(account, ts);
CREATE INDEX IF NOT EXISTS idx_calls_model ON api_calls(model, ts);
CREATE INDEX IF NOT EXISTS idx_calls_ok ON api_calls(ok, ts);
CREATE INDEX IF NOT EXISTS idx_calls_status ON api_calls(status, ts);
CREATE INDEX IF NOT EXISTS idx_calls_rid ON api_calls(rid);
"""
_COLUMNS = ("ts", "rid", "account", "key_id", "kind", "method", "path", "model",
"served", "provider", "caller", "stream", "ok", "status", "latency_ms",
"ip", "ua", "prompt", "completion", "ver")
_INSERT = f"INSERT INTO api_calls ({', '.join(_COLUMNS)}) VALUES ({', '.join('?' * len(_COLUMNS))})"
def _clip(v: Any, n: int = CONTENT_MAX) -> str:
"""内容裁剪:非字符串转字符串,超长截断并标注。"""
if v is None:
return ""
s = v if isinstance(v, str) else str(v)
if len(s) > n:
return s[:n] + f"\n…(已截断,共 {len(s)} 字)"
return s
class CallLogStore:
"""SQLite 调用日志存储。线程安全的单写者 + 单读连接。"""
def __init__(self, path: str = DB_PATH, *, retention_days: int = RETENTION_DAYS,
queue_max: int = QUEUE_MAX, enabled: bool = True) -> None:
self.path = path
self.retention_days = retention_days
self.queue_max = queue_max
self.enabled = enabled and not DISABLED
self._q: deque[tuple] = deque()
self._lock = threading.Lock()
self._db: sqlite3.Connection | None = None
self._task: asyncio.Task | None = None
self._running = False
self.written = 0
self.dropped = 0
self.last_error = ""
# ── 生命周期 ────────────────────────────────────────────────
def open(self) -> None:
if not self.enabled:
log.info("调用日志已禁用(CALLLOG_DISABLED)")
return
os.makedirs(os.path.dirname(self.path) or ".", exist_ok=True)
db = sqlite3.connect(self.path, check_same_thread=False, timeout=10)
db.execute("PRAGMA journal_mode=WAL")
db.execute("PRAGMA synchronous=NORMAL")
db.executescript(SCHEMA)
db.commit()
try:
os.chmod(self.path, 0o600)
except OSError:
pass
self._db = db
log.info("调用日志已就绪: %s(保留 %d 天)", self.path, self.retention_days)
self.cleanup()
async def start(self) -> None:
self.open()
if not self.enabled or self._task is not None:
return
self._running = True
self._task = asyncio.create_task(self._loop(), name="calllog-writer")
log.info("调用日志后台写入已启动")
async def close(self) -> None:
self._running = False
if self._task:
self._task.cancel()
try:
await self._task
except (asyncio.CancelledError, Exception):
pass
self._task = None
self.flush()
if self._db:
self._db.close()
self._db = None
async def _loop(self) -> None:
last_clean = time.time()
while self._running:
await asyncio.sleep(FLUSH_INTERVAL)
try:
self.flush()
sweep()
if time.time() - last_clean > CLEAN_INTERVAL:
last_clean = time.time()
self.cleanup()
except Exception as e: # noqa: BLE001
self.last_error = f"{type(e).__name__}: {e}"
log.warning("调用日志写入失败: %s", e)
# ── 写入 ────────────────────────────────────────────────────
def stage(self, row: dict[str, Any]) -> None:
"""入队(非阻塞)。队列满则丢弃并计数。"""
if not self.enabled:
return
try:
tup = tuple(row.get(c) for c in _COLUMNS)
except Exception: # noqa: BLE001
return
with self._lock:
if len(self._q) >= self.queue_max:
self.dropped += 1
return
self._q.append(tup)
def flush(self) -> int:
"""把队列里的行一次批量写库。返回写入条数。"""
if not self.enabled or self._db is None:
return 0
with self._lock:
if not self._q:
return 0
rows = list(self._q)
self._q.clear()
try:
with self._lock:
self._db.executemany(_INSERT, rows)
self._db.commit()
self.written += len(rows)
return len(rows)
except Exception as e: # noqa: BLE001
self.last_error = f"{type(e).__name__}: {e}"
log.warning("调用日志批量写入失败(丢弃 %d 条): %s", len(rows), e)
return 0
# ── 清理 ────────────────────────────────────────────────────
def cleanup(self) -> int:
if not self.enabled or self._db is None:
return 0
cutoff = time.time() - self.retention_days * 86400
try:
with self._lock:
cur = self._db.execute("DELETE FROM api_calls WHERE ts < ?", (cutoff,))
self._db.commit()
n = cur.rowcount or 0
if n:
log.info("调用日志清理:删除 %d 条(早于 %d 天前)", n, self.retention_days)
return n
except Exception as e: # noqa: BLE001
log.warning("调用日志清理失败: %s", e)
return 0
# ── 查询 ────────────────────────────────────────────────────
def _where(self, *, start: float | None, end: float | None, account: str | None,
model: str | None, ok: int | None, status: int | None,
kind: str | None, q: str | None,
with_content: bool = False) -> tuple[str, list[Any]]:
w: list[str] = []
p: list[Any] = []
if start is not None:
w.append("ts >= ?")
p.append(start)
if end is not None:
w.append("ts <= ?")
p.append(end)
if account:
w.append("(account = ? OR caller = ?)")
p.extend([account, account])
if model:
w.append("(model = ? OR served = ?)")
p.extend([model, model])
if ok is not None:
w.append("ok = ?")
p.append(int(ok))
if status:
w.append("status = ?")
p.append(int(status))
if kind:
w.append("kind = ?")
p.append(kind)
if q:
like = f"%{q}%"
w.append("(rid LIKE ? OR account LIKE ? OR model LIKE ? OR prompt LIKE ? "
"OR completion LIKE ? OR ua LIKE ? OR ip LIKE ?)")
p.extend([like] * 7)
return (" WHERE " + " AND ".join(w)) if w else "", p
def query(self, *, start: float | None = None, end: float | None = None,
account: str | None = None, model: str | None = None,
ok: int | None = None, status: int | None = None,
kind: str | None = None, q: str | None = None,
limit: int = 50, offset: int = 0,
with_content: bool = False) -> dict[str, Any]:
"""分页查询。with_content=False 时不返回 prompt/completion(列表页省流量)。"""
if not self.enabled or self._db is None:
return {"total": 0, "items": [], "enabled": False}
where, params = self._where(start=start, end=end, account=account, model=model,
ok=ok, status=status, kind=kind, q=q)
cols = ("id, ts, rid, account, key_id, kind, method, path, model, served, provider, "
"caller, stream, ok, status, latency_ms, ip, ua, ver")
if with_content:
cols += ", prompt, completion"
limit = max(1, min(int(limit), 500))
offset = max(0, int(offset))
try:
with self._lock:
total = self._db.execute(f"SELECT COUNT(*) FROM api_calls{where}", params).fetchone()[0]
rows = self._db.execute(
f"SELECT {cols} FROM api_calls{where} ORDER BY ts DESC LIMIT ? OFFSET ?",
[*params, limit, offset],
).fetchall()
except Exception as e: # noqa: BLE001
log.warning("调用日志查询失败: %s", e)
return {"total": 0, "items": [], "error": str(e)}
names = [d[0] for d in self._db.execute(f"SELECT {cols} FROM api_calls LIMIT 0").description]
items = []
for r in rows:
d = dict(zip(names, r))
d["ts_iso"] = time.strftime("%Y-%m-%d %H:%M:%S", time.localtime(d["ts"]))
items.append(d)
return {"total": total, "items": items, "limit": limit, "offset": offset}
def get(self, call_id: int) -> dict[str, Any] | None:
if not self.enabled or self._db is None:
return None
try:
with self._lock:
cur = self._db.execute("SELECT * FROM api_calls WHERE id = ?", (int(call_id),))
row = cur.fetchone()
names = [d[0] for d in cur.description]
except Exception: # noqa: BLE001
return None
if not row:
return None
d = dict(zip(names, row))
d["ts_iso"] = time.strftime("%Y-%m-%d %H:%M:%S", time.localtime(d["ts"]))
return d
def stats(self, days: int = 7) -> dict[str, Any]:
"""按天 / 模型 / 账号 / 状态码聚合(前端图表与概览用)。"""
if not self.enabled or self._db is None:
return {"enabled": False}
since = time.time() - max(1, int(days)) * 86400
out: dict[str, Any] = {"enabled": True, "days": days}
try:
with self._lock:
out["daily"] = [
{"date": r[0], "total": r[1], "ok": r[2], "fail": r[3],
"avg_ms": round(r[4] or 0, 1)}
for r in self._db.execute(
"SELECT date(ts, 'unixepoch', 'localtime') d, COUNT(*), "
"SUM(ok), SUM(CASE WHEN ok=0 THEN 1 ELSE 0 END), AVG(latency_ms) "
"FROM api_calls WHERE ts >= ? GROUP BY d ORDER BY d", (since,))
]
out["models"] = [
{"model": r[0], "total": r[1], "fail": r[2], "avg_ms": round(r[3] or 0, 1)}
for r in self._db.execute(
"SELECT model, COUNT(*), SUM(CASE WHEN ok=0 THEN 1 ELSE 0 END), "
"AVG(latency_ms) FROM api_calls WHERE ts >= ? "
"GROUP BY model ORDER BY COUNT(*) DESC LIMIT 20", (since,))
]
out["accounts"] = [
{"account": r[0], "total": r[1], "fail": r[2]}
for r in self._db.execute(
"SELECT account, COUNT(*), SUM(CASE WHEN ok=0 THEN 1 ELSE 0 END) "
"FROM api_calls WHERE ts >= ? GROUP BY account "
"ORDER BY COUNT(*) DESC LIMIT 50", (since,))
]
out["statuses"] = [
{"status": r[0], "total": r[1]}
for r in self._db.execute(
"SELECT status, COUNT(*) FROM api_calls WHERE ts >= ? "
"GROUP BY status ORDER BY COUNT(*) DESC LIMIT 20", (since,))
]
out["total"] = self._db.execute(
"SELECT COUNT(*) FROM api_calls").fetchone()[0]
out["size_bytes"] = os.path.getsize(self.path) if os.path.exists(self.path) else 0
except Exception as e: # noqa: BLE001
out["error"] = str(e)
out["written"] = self.written
out["dropped"] = self.dropped
out["queued"] = len(self._q)
return out
def export_rows(self, **kw: Any) -> list[dict[str, Any]]:
"""导出用:不带分页,最多 50000 条。"""
kw.setdefault("limit", 50000)
kw.setdefault("offset", 0)
kw["with_content"] = True
return self.query(**kw).get("items", [])
store = CallLogStore()
# ── 与请求生命周期的双向握手 ──────────────────────────────────────
# 一条调用日志由两半拼成:
# ① 访问日志条目(AccessLog.add):kind/model/provider/caller/ok/stream/body/response
# ② 响应元数据(中间件收尾):status/耗时/IP/UA/Key 前 8 位
# 但两半的**先后顺序是不确定的**:
# · 非流式:① 在请求处理中途就 add 了 → ② 后到
# · 流式:② 在 StreamingResponse 建好就返回了 → ① 要等流真正发完才 add(后到)
# 所以两边都先暂存、谁发现对方已就绪就立刻拼成一行落库。
_PEND_ENTRY: dict[str, dict] = {} # rid -> 访问日志条目(等 meta)
_PEND_META: dict[str, dict] = {} # rid -> 响应元数据(等 entry)
_pend_lock = threading.Lock()
_PEND_MAX = 2000
_META_TTL = 300.0 # 元数据最多等 5 分钟,避免泄漏
def _trim(d: dict[str, dict]) -> None:
while len(d) > _PEND_MAX:
try:
d.pop(next(iter(d)), None)
except (StopIteration, KeyError):
break
def _row(entry: dict[str, Any], meta: dict[str, Any]) -> dict[str, Any]:
return {
"ts": entry.get("ts") or time.time(),
"rid": entry.get("rid") or meta.get("rid"),
"account": entry.get("account") or entry.get("caller") or meta.get("account"),
"key_id": meta.get("key_id", ""),
"kind": entry.get("kind") or meta.get("kind"),
"method": entry.get("method") or meta.get("method"),
"path": entry.get("path") or meta.get("path"),
"model": entry.get("model"),
"served": entry.get("served"),
"provider": entry.get("provider"),
"caller": entry.get("caller"),
"stream": 1 if entry.get("stream") else 0,
"ok": 1 if entry.get("ok") else 0,
"status": meta.get("status"),
"latency_ms": meta.get("latency_ms"),
"ip": meta.get("ip", ""),
"ua": _clip(meta.get("ua", ""), 200),
"prompt": _clip(entry.get("body"), CONTENT_MAX),
"completion": _clip(entry.get("response"), CONTENT_MAX),
"ver": meta.get("ver", ""),
}
def on_access(entry: dict[str, Any]) -> None:
"""`AccessLog.add()` 调用:拿到访问日志条目;若响应元数据已就绪则立即落库。"""
if not store.enabled:
return
rid = str(entry.get("rid") or "")
if not rid:
store.stage(_row(entry, {}))
return
with _pend_lock:
meta = _PEND_META.pop(rid, None)
if meta is None:
_PEND_ENTRY[rid] = entry
_trim(_PEND_ENTRY)
return
store.stage(_row(entry, meta))
def on_response(rid: str, *, status: int, latency_ms: float, ip: str = "", ua: str = "",
key_id: str = "", ver: str = "", method: str = "", path: str = "",
account: str = "", stream: bool = False, body: str = "") -> None:
"""请求中间件收尾调用:补上响应元数据;条目已就绪则立即落库。
`stream=True` 时条目一定还没到(SSE 还没开始发),只暂存等 `on_access`;
非流式且没有条目、又是 **/v1/ 下的失败**时,补一条"早失败"记录(例如未知模型 400、
鉴权 401/403、限流 429)—— 这类请求压根走不到 `_access.add`,但恰恰最需要留痕。
"""
if not store.enabled:
return
rid = str(rid or "")
meta = {"rid": rid, "status": int(status) if status else None,
"latency_ms": int(latency_ms), "ip": ip, "ua": ua, "key_id": key_id,
"ver": ver, "method": method, "path": path, "account": account,
"kind": "chat" if path.startswith("/v1/") else None,
"ts": time.time()}
if rid:
with _pend_lock:
entry = _PEND_ENTRY.pop(rid, None)
if entry is not None:
store.stage(_row(entry, meta))
return
if stream:
_PEND_META[rid] = meta # 等 event_stream 收尾时补条目
_trim(_PEND_META)
return
# 没有条目且不是流式:只有 /v1/ 下的失败才补记(成功一定有条目;其余端点不算"调用")
if status and int(status) >= 400 and path.startswith("/v1/"):
# 早失败(未知模型 / 鉴权失败 / 限流)走不到 AccessLog.add,这里用请求体补全信息,
# 否则日志里只有一行"400、无模型",排查时看不出客户端到底请求了什么。
model = ""
try:
import json as _json
model = str((_json.loads(body or "{}") or {}).get("model") or "")
except Exception: # noqa: BLE001
model = ""
store.stage(_row({"model": model or None, "body": body}, meta))
def sweep() -> int:
"""丢弃超时未配对的暂存项(由后台循环定期调用)。"""
now = time.time()
n = 0
with _pend_lock:
for d in (_PEND_META,):
for k in [k for k, v in d.items() if now - float(v.get("ts") or 0) > _META_TTL]:
d.pop(k, None)
n += 1
return n
+152 -1
View File
@@ -44,6 +44,7 @@ from app.circuit import CircuitBreakerRegistry
from app.config import (ConfigError, GatewayConfig, load_config, validate_warnings, from app.config import (ConfigError, GatewayConfig, load_config, validate_warnings,
PROTOCOL_CHAT, PROTOCOL_MESSAGES, PROTOCOL_RESPONSES, PROTOCOLS, PROTOCOL_CHAT, PROTOCOL_MESSAGES, PROTOCOL_RESPONSES, PROTOCOLS,
ALIAS_PROTOCOLS) ALIAS_PROTOCOLS)
from app import calllog
from app.metrics import (Metrics, AccessLog, RateLimiter, BODY_BY_RID, _BODY_MAX, from app.metrics import (Metrics, AccessLog, RateLimiter, BODY_BY_RID, _BODY_MAX,
video_poll_push, video_polls, video_poll_push, video_polls,
ACCOUNT, ROLE) ACCOUNT, ROLE)
@@ -220,6 +221,11 @@ async def lifespan(app: FastAPI):
hot_task = None hot_task = None
if config.server.hot_reload: if config.server.hot_reload:
hot_task = asyncio.create_task(_hot_reload_loop()) hot_task = asyncio.create_task(_hot_reload_loop())
# 调用日志落地(2026-10-23):SQLite + 后台批量写,失败不影响服务
try:
await calllog.store.start()
except Exception as e: # noqa: BLE001
log.warning("调用日志启动失败(不影响服务): %s", e)
yield yield
# 关闭阶段:先停热重载,再等在途请求(uvicorn 自身也做,这里补逻辑日志) # 关闭阶段:先停热重载,再等在途请求(uvicorn 自身也做,这里补逻辑日志)
if hot_task: if hot_task:
@@ -239,6 +245,11 @@ async def lifespan(app: FastAPI):
# 第六轮: 快照后再迭代(避免迭代中字典被改动 → RuntimeError) # 第六轮: 快照后再迭代(避免迭代中字典被改动 → RuntimeError)
for adapter in list(_state.get("adapters", {}).values()): for adapter in list(_state.get("adapters", {}).values()):
await adapter.close() await adapter.close()
# 调用日志:把队列里剩下的刷盘再关库(否则重启会丢最后不到 1 秒的记录)
try:
await calllog.store.close()
except Exception as e: # noqa: BLE001
log.warning("调用日志收尾失败: %s", e)
async def _apply_config(new: GatewayConfig, *, source: str = "热重载") -> None: async def _apply_config(new: GatewayConfig, *, source: str = "热重载") -> None:
@@ -340,7 +351,7 @@ _docs_on = config.server.docs_enabled # 安全默认关(report: /docs、/ope
# 2026-09-30: 2.6.1 → 2.7.0(按老板要求整个删除「火山方舟 Coding Plan」供应商:配置/模型/代码/文档/前端下拉/环境变量全清)。 # 2026-09-30: 2.6.1 → 2.7.0(按老板要求整个删除「火山方舟 Coding Plan」供应商:配置/模型/代码/文档/前端下拉/环境变量全清)。
# 2026-09-30: 2.7.0 → 2.7.1(界面提示白话化:熔断器→供应商自动保护、引擎→运行环境、slug→标识;删 120 个未引用死键)。 # 2026-09-30: 2.7.0 → 2.7.1(界面提示白话化:熔断器→供应商自动保护、引擎→运行环境、slug→标识;删 120 个未引用死键)。
# ★ 版本号纪律:每次改动都要往上加,小改动加小版本号、大改动加大版本号(README 有专章)。 # ★ 版本号纪律:每次改动都要往上加,小改动加小版本号、大改动加大版本号(README 有专章)。
VERSION = "2.7.3" VERSION = "2.8.0"
app = FastAPI( app = FastAPI(
title="API Gateway", title="API Gateway",
@@ -548,13 +559,65 @@ def _logsafe(v, maxlen: int = 200) -> str:
return "".join(ch if ch.isprintable() else "?" for ch in str(v))[:maxlen] return "".join(ch if ch.isprintable() else "?" for ch in str(v))[:maxlen]
def _client_ip(request: Request) -> str:
"""取客户端真实 IP:优先 X-Forwarded-For(EO/OpenResty 回源会带),否则取连接地址。
只给管理员看(/logs/calls 是 admin 端点),所以保留完整 IP 便于与 WAF/防火墙日志对账。
"""
xff = request.headers.get("x-forwarded-for") or ""
if xff:
first = xff.split(",")[0].strip()
if first:
return first
return getattr(request.client, "host", "") or ""
def _flush_calllog(request: Request, rid: str, status: int, latency_ms: float,
stream: bool = False) -> None:
"""请求收尾:把响应元数据交给 calllog,与访问日志条目拼成一行落库。"""
try:
# 只取前 8 位做标识,绝不落完整 Key。
# 鉴权失败(401)时 request.state.credential 是空的,但**恰恰这时最需要知道
# 客户端到底拿的是哪把 Key**(2026-10-23 真实工单:客户一直 401,靠这个前缀
# 才能判断他手上是不是旧 Key)—— 所以失败路径也记前缀。
hdr = request.headers.get("authorization") or ""
presented = (hdr[7:] if hdr.lower().startswith("bearer ")
else (request.headers.get("x-api-key") or "")).strip()
key_id = presented[:8] if presented else ""
calllog.on_response(
rid,
status=status,
latency_ms=latency_ms,
ip=_client_ip(request),
ua=request.headers.get("user-agent") or "",
key_id=key_id,
ver=VERSION,
method=request.method,
path=request.url.path,
account=getattr(request.state, "account", "") or "",
stream=stream,
body=BODY_BY_RID.pop(rid, "") or "",
)
except Exception as e: # noqa: BLE001
log.debug("调用日志落地失败(忽略): %s", e)
@app.middleware("http") @app.middleware("http")
async def add_request_id(request: Request, call_next): async def add_request_id(request: Request, call_next):
rid = _sanitize_rid(request.headers.get("X-Request-ID")) or str(uuid.uuid4()) rid = _sanitize_rid(request.headers.get("X-Request-ID")) or str(uuid.uuid4())
request.state.rid = rid request.state.rid = rid
t0 = time.perf_counter()
status = 500
is_stream = False
try:
response = await call_next(request) response = await call_next(request)
status = response.status_code
response.headers["X-Request-ID"] = rid response.headers["X-Request-ID"] = rid
# SSE 请求的访问日志条目要等流真正发完才产生 → 告知 calllog 先等条目
is_stream = str(response.headers.get("content-type", "")).startswith("text/event-stream")
return response return response
finally:
_flush_calllog(request, rid, status, (time.perf_counter() - t0) * 1000, is_stream)
# Fix 5(本轮): CORS 注册放在最后 → 位于最外层,覆盖 limit_body 等短路响应。 # Fix 5(本轮): CORS 注册放在最后 → 位于最外层,覆盖 limit_body 等短路响应。
@@ -3061,6 +3124,94 @@ async def logs(request: Request, n: int = 100, _auth=Depends(require_auth)):
return {"entries": _logs_for_account(request.state.account, n)} return {"entries": _logs_for_account(request.state.account, n)}
# ─────────────────────────────────────
# 调用日志(持久化,2026-10-23)—— 仅管理员
#
# 与上面的 /logs 的区别:
# /logs 内存环形缓冲(默认 500 条,重启即丢),**只看自己的**;
# /logs/calls SQLite 落盘(默认保留 90 天),管理员可分页/筛选/看**所有人**的,
# 含提示词与回复内容(截断记录)。
# ─────────────────────────────────────
def _require_admin_logs(request: Request) -> None:
"""调用日志含全站内容,仅管理员可读(口径与 /metrics/prometheus、/health 收敛一致)。"""
if getattr(request.state, "role", "") != "admin":
raise HTTPException(403, "仅管理员可查看调用日志")
def _as_ts(v: str | None) -> float | None:
"""时间参数:支持 epoch 秒、'2026-10-23'、'2026-10-23 03:00'、ISO 'T' 分隔。"""
if v is None or str(v).strip() == "":
return None
s2 = str(v).strip()
try:
return float(s2)
except ValueError:
pass
for fmt in ("%Y-%m-%d %H:%M:%S", "%Y-%m-%d %H:%M", "%Y-%m-%dT%H:%M:%S",
"%Y-%m-%dT%H:%M", "%Y-%m-%d"):
try:
return time.mktime(time.strptime(s2, fmt))
except ValueError:
continue
return None
@app.get("/logs/calls/stats")
async def logs_calls_stats(request: Request, days: int = 7, _auth=Depends(require_auth)):
_require_admin_logs(request)
return calllog.store.stats(days=days)
@app.get("/logs/calls/export")
async def logs_calls_export(request: Request, start: str | None = None, end: str | None = None,
account: str | None = None, model: str | None = None,
ok: int | None = None, status: int | None = None,
kind: str | None = None, q: str | None = None,
_auth=Depends(require_auth)):
"""导出 CSV(最多 5 万条,含提示词与回复内容)。"""
_require_admin_logs(request)
rows = calllog.store.export_rows(start=_as_ts(start), end=_as_ts(end), account=account,
model=model, ok=ok, status=status, kind=kind, q=q)
import csv as _csv
import io as _io
buf = _io.StringIO()
w = _csv.writer(buf)
w.writerow(["时间", "rid", "账号", "Key前8位", "类型", "模型", "实际模型", "供应商",
"流式", "成功", "状态码", "耗时ms", "IP", "UA", "提示词", "回复"])
for r in rows:
w.writerow([r.get("ts_iso"), r.get("rid"), r.get("account"), r.get("key_id"),
r.get("kind"), r.get("model"), r.get("served"), r.get("provider"),
"是" if r.get("stream") else "否", "是" if r.get("ok") else "否",
r.get("status"), r.get("latency_ms"), r.get("ip"), r.get("ua"),
(r.get("prompt") or "").replace("\n", "\\n"),
(r.get("completion") or "").replace("\n", "\\n")])
fname = time.strftime("calllog-%Y%m%d-%H%M%S.csv", time.localtime())
return Response(content=buf.getvalue().encode("utf-8-sig"), media_type="text/csv",
headers={"Content-Disposition": f'attachment; filename="{fname}"'})
@app.get("/logs/calls")
async def logs_calls(request: Request, start: str | None = None, end: str | None = None,
account: str | None = None, model: str | None = None,
ok: int | None = None, status: int | None = None,
kind: str | None = None, q: str | None = None,
limit: int = 50, offset: int = 0, content: int = 0,
_auth=Depends(require_auth)):
_require_admin_logs(request)
return calllog.store.query(start=_as_ts(start), end=_as_ts(end), account=account,
model=model, ok=ok, status=status, kind=kind, q=q,
limit=limit, offset=offset, with_content=bool(content))
@app.get("/logs/calls/{call_id}")
async def logs_call_detail(request: Request, call_id: int, _auth=Depends(require_auth)):
_require_admin_logs(request)
row = calllog.store.get(call_id)
if not row:
raise HTTPException(404, f"调用日志不存在: {call_id}")
return row
# ───────────────────────────────────── # ─────────────────────────────────────
# GET / — 端点导航(功能 7) # GET / — 端点导航(功能 7)
# ───────────────────────────────────── # ─────────────────────────────────────
+8
View File
@@ -25,6 +25,7 @@ ROLE: ContextVar[str] = ContextVar("gw_role", default="member")
# 附上「具体请求内容」,便于在日志页查看实际发了什么。有界缓存,防无界增长。 # 附上「具体请求内容」,便于在日志页查看实际发了什么。有界缓存,防无界增长。
BODY_BY_RID: dict[str, str] = {} BODY_BY_RID: dict[str, str] = {}
# ── 视频轮询历史(/logs 按 video_id 合成到「创建」那一条里,不再每个轮询各占一行) ── # ── 视频轮询历史(/logs 按 video_id 合成到「创建」那一条里,不再每个轮询各占一行) ──
VIDEO_POLLS: dict[str, "deque"] = {} VIDEO_POLLS: dict[str, "deque"] = {}
_vp_lock = threading.Lock() _vp_lock = threading.Lock()
@@ -272,6 +273,13 @@ class AccessLog:
entry["body"] = snip entry["body"] = snip
with self._lock: with self._lock:
self._buf.append(entry) self._buf.append(entry)
# 调用日志落地(2026-10-23):交给 app/calllog.py —— 它与请求中间件的响应元数据
# 做双向握手(谁后到谁拼行),流式请求的条目比响应晚到也能落到库里。
try:
from app import calllog
calllog.on_access(entry)
except Exception: # noqa: BLE001
pass
def tail(self, n: int = 100) -> list[dict[str, Any]]: def tail(self, n: int = 100) -> list[dict[str, Any]]:
with self._lock: with self._lock:
+84
View File
@@ -0,0 +1,84 @@
# 调用日志落地设计(2026-10-23)
> 方案对比与取舍记录。实现在 `app/calllog.py`,界面在「日志」页的「调用日志」页签。
## 为什么需要
`AccessLog` 是内存环形缓冲(默认 500 条,**重启即丢**),`/metrics` 只有聚合计数。
2026-10-22 的真实工单很说明问题:成员反馈「拿 apikey 请求 403」,最后是靠 grep 1Panel 的
WAF `access.log` 才定位 —— 网关这边**没有任何一条可查的调用记录**。调用日志要解决的就是这个。
## 方案对比
| | **A. SQLite(采用)** | B. JSONL 文件 | C. MySQL | D. Loki/ELK |
|---|---|---|---|---|
| 新增依赖 | **0**(标准库)| 0 | 驱动+连接池 | 整套平台 |
| 查询/聚合 | SQL ✅ | grep/jq | 强 | 最强 |
| 写入风险 | 低(WAL + 单写者 + 队列)| 最低 | 中(DB 挂了影响写)| 中 |
| 运维成本 | 极低(一个文件)| 极低 | 中 | 高 |
| 适合 | 单机、日调用 < 50 万 | 只求能查 | 多节点统一账单 | 规模化 |
选 A:项目依赖极简(7 个包),引入 MySQL 驱动破坏基调;MySQL 容器是 Gitea 在用,混用会把
「日志」和「业务库」绑死。
## 表结构要点
21 列,见 `app/calllog.py` 的 `SCHEMA`。索引覆盖四条最常用的查询路径:
`ts`、`(account, ts)`、`(model, ts)`、`(ok, ts)`、`(status, ts)`、`rid`。
## 三个关键设计
### 1. 双向握手(两半日志拼起来)
一条日志由两处产生,**先后顺序不确定**:
| 场景 | 先到 | 后到 |
|---|---|---|
| 非流式 | 访问日志条目(请求处理中途)| 响应元数据(状态码/耗时/IP/UA)|
| **流式** | 响应元数据(StreamingResponse 建好就返回)| 访问日志条目(SSE 真发完才 add)|
实现上两边都先暂存(`_PEND_ENTRY` / `_PEND_META`),谁发现对方已就绪就立刻拼成一行落库;
元数据最多等 5 分钟(`sweep()` 定期丢弃),避免泄漏。
### 2. 早失败也记
未知模型(400)、鉴权失败(401/403)、限流(429)这些请求**走不到** `AccessLog.add`。
中间件在收尾时若发现"没有条目 + `/v1/` 下 + status ≥ 400",就用请求体补一条记录,
并解析出 `model` 字段 —— 否则日志里只有一行"400、无模型",排查时没有任何信息量。
### 3. 失败也记 Key 前缀
401 时 `request.state.credential` 是空的,但**恰恰这时最需要知道客户端拿的是哪把 Key**。
所以从 `Authorization` / `x-api-key` 里取前 8 位落库(完整 Key 绝不落盘)。
> 2026-10-23 实测:靠这个字段抓到客户一直在发 `wjd-5Da_…`,而库里他的有效 Key 是
> `wjd-15zs…` —— 用的是旧 Key,与网关无关。
## 体量与保留
| 日调用量 | 单条约 | 90 天 |
|---|---|---|
| 100 次 | 400 B | ~4 MB |
| 1,000 次 | 400 B | ~36 MB |
| 10,000 次 | 400 B | ~360 MB |
默认保留 **90 天**(`CALLLOG_RETENTION_DAYS`),启动时 + 每 6 小时清理;`VACUUM` 未自动跑,
需要时手动执行(或等保留策略删够量后由 SQLite 复用页)。
## 配置
| 环境变量 | 默认 | 说明 |
|---|---|---|
| `CALLLOG_DB` | `data/calllog.db` | 库路径 |
| `CALLLOG_RETENTION_DAYS` | `90` | 保留天数 |
| `CALLLOG_QUEUE_MAX` | `10000` | 内存队列上限(满了丢弃并计数)|
| `CALLLOG_CONTENT_MAX` | `8000` | 提示词/回复各自截断长度 |
| `CALLLOG_DISABLED` | 空 | 设为 1 可整体关闭(不影响现有行为)|
## 已知限制
- 提示词内容受上游 `BODY_BY_RID` 的 2000 字符上限约束 —— 编码类长 prompt 会被截断。
想放长需要调 `app/main.py` 里 `limit_body` 中间件的 `raw[:2000]`。
- 流式请求的 `latency_ms` 是**响应头返回的时间**(≈首字节时间),不是流的总时长。
- 没有自动 `VACUUM`;库文件会随写入增长,90 天清理后页空间会被复用,不会无限膨胀。
+30
View File
@@ -169,6 +169,8 @@ export default {
byProvider: 'Calls by provider (success / fail)', byProvider: 'Calls by provider (success / fail)',
}, },
logs: { logs: {
tabLive: 'Realtime',
tabCalls: 'Call log',
all: 'All', all: 'All',
chat: 'Chat', chat: 'Chat',
image: 'Image', image: 'Image',
@@ -818,5 +820,33 @@ export default {
errAudioTooBig: '{name} is {size}, over the reference-audio limit of {max} (please trim or compress it)', errAudioTooBig: '{name} is {size}, over the reference-audio limit of {max} (please trim or compress it)',
errBodyBudget: 'This request body is about {size}, over the gateway upload limit (~{max}).', errBodyBudget: 'This request body is about {size}, over the gateway upload limit (~{max}).',
errBodyBudgetTip: 'Please use fewer materials, or smaller images/audio files.', errBodyBudgetTip: 'Please use fewer materials, or smaller images/audio files.',
calllogs: {
title: 'Call log (persisted)',
count: '{n} rows',
failN: '{n} failed',
avgMs: 'avg {n}ms',
startDate: 'Start',
endDate: 'End',
account: 'Account',
model: 'Model',
result: 'Result',
onlyOk: 'OK only',
onlyFail: 'Failed only',
keyword: 'Keyword (content/rid/IP/UA)',
colTime: 'Time',
colAccount: 'Account',
colModel: 'Model',
colLatency: 'Latency',
colStatus: 'Status',
colStream: 'Stream',
colUa: 'UA',
yes: 'yes',
no: 'no',
detailTitle: 'Call detail',
prompt: 'Prompt (request body, truncated)',
completion: 'Completion (truncated)',
loading: 'Loading…',
},
}, },
} }
+30
View File
@@ -169,6 +169,8 @@ export default {
byProvider: '按供应商调用(成功 / 失败)', byProvider: '按供应商调用(成功 / 失败)',
}, },
logs: { logs: {
tabLive: '实时',
tabCalls: '调用日志',
all: '全部', all: '全部',
chat: '对话', chat: '对话',
image: '图像', image: '图像',
@@ -817,5 +819,33 @@ export default {
errAudioTooBig: '{name} 有 {size},超过参考音频上限 {max}(请裁剪或压缩后再传)', errAudioTooBig: '{name} 有 {size},超过参考音频上限 {max}(请裁剪或压缩后再传)',
errBodyBudget: '本次请求体约 {size},超过网关上送上限(约 {max})。', errBodyBudget: '本次请求体约 {size},超过网关上送上限(约 {max})。',
errBodyBudgetTip: '请减少素材数量,或换体积更小的图片/音频。', errBodyBudgetTip: '请减少素材数量,或换体积更小的图片/音频。',
calllogs: {
title: '调用日志(持久化)',
count: '共 {n} 条',
failN: '失败 {n}',
avgMs: '平均 {n}ms',
startDate: '开始日期',
endDate: '结束日期',
account: '账号',
model: '模型',
result: '结果',
onlyOk: '仅成功',
onlyFail: '仅失败',
keyword: '关键词(内容/rid/IP/UA)',
colTime: '时间',
colAccount: '账号',
colModel: '模型',
colLatency: '耗时',
colStatus: '状态',
colStream: '流式',
colUa: 'UA',
yes: '是',
no: '否',
detailTitle: '调用详情',
prompt: '提示词(请求体,截断记录)',
completion: '回复(截断记录)',
loading: '加载中…',
},
}, },
} }
+239
View File
@@ -0,0 +1,239 @@
<template>
<div class="calllogs">
<!-- 筛选行 -->
<div class="filters">
<el-date-picker
v-model="range" type="daterange" size="small" class="w-range"
value-format="YYYY-MM-DD" :start-placeholder="t('calllogs.startDate')"
:end-placeholder="t('calllogs.endDate')" :clearable="true"
/>
<el-input v-model="account" size="small" class="w140" clearable
:placeholder="t('calllogs.account')" @keyup.enter="reload" />
<el-input v-model="model" size="small" class="w170" clearable
:placeholder="t('calllogs.model')" @keyup.enter="reload" />
<el-select v-model="okSel" size="small" class="w120" clearable :placeholder="t('calllogs.result')">
<el-option value="1" :label="t('calllogs.onlyOk')" />
<el-option value="0" :label="t('calllogs.onlyFail')" />
</el-select>
<el-input v-model="kw" size="small" class="w200" clearable
:placeholder="t('calllogs.keyword')" @keyup.enter="reload" />
<span class="spacer" />
<el-button size="small" :loading="loading" @click="reload">
<Icon name="refresh" :size="13" /><span>{{ t('common.refresh') }}</span>
</el-button>
<el-button size="small" :disabled="!total" @click="exportCsv">
<Icon name="dl" :size="13" /><span>{{ t('common.export') }}</span>
</el-button>
</div>
<div class="card">
<div class="card-h">
<b>{{ t('calllogs.title') }}</b>
<span class="sub">{{ t('calllogs.count', { n: total }) }}</span>
<div class="pager">
<span class="pill" :class="failCount ? 'warn' : 'ok'">
{{ t('calllogs.failN', { n: failCount }) }}
</span>
<span class="pill">{{ t('calllogs.avgMs', { n: avgMs }) }}</span>
</div>
</div>
<el-table :data="rows" size="small" v-loading="loading" max-height="580"
class="tbl" @row-click="openDetail" :row-class-name="rowCls">
<el-table-column prop="ts_iso" :label="t('calllogs.colTime')" width="152" />
<el-table-column prop="account" :label="t('calllogs.colAccount')" width="112" show-overflow-tooltip />
<el-table-column prop="model" :label="t('calllogs.colModel')" min-width="150" show-overflow-tooltip>
<template #default="{ row }">
<span>{{ row.model || '—' }}</span>
<span v-if="row.served && row.served !== row.model" class="sub"> → {{ row.served }}</span>
</template>
</el-table-column>
<el-table-column :label="t('calllogs.colLatency')" width="92">
<template #default="{ row }">{{ row.latency_ms != null ? row.latency_ms + 'ms' : '—' }}</template>
</el-table-column>
<el-table-column :label="t('calllogs.colStatus')" width="84">
<template #default="{ row }">
<span :class="row.ok ? 'ok-t' : 'err-t'">{{ row.status ?? '—' }}</span>
</template>
</el-table-column>
<el-table-column :label="t('calllogs.colStream')" width="66">
<template #default="{ row }">{{ row.stream ? t('calllogs.yes') : '—' }}</template>
</el-table-column>
<el-table-column prop="ua" :label="t('calllogs.colUa')" min-width="130" show-overflow-tooltip />
</el-table>
<div class="pager-bar">
<el-pagination
v-model:current-page="page" :page-size="pageSize" :total="total"
layout="prev, pager, next, total" small background @current-change="loadPage"
/>
</div>
</div>
<!-- 详情弹窗:完整提示词 + 回复 -->
<el-dialog v-model="dlg" :title="t('calllogs.detailTitle')" width="860px" top="6vh">
<div v-if="detail" class="det">
<div class="kv">
<span><i>{{ t('calllogs.colTime') }}</i>{{ detail.ts_iso }}</span>
<span><i>rid</i><code>{{ detail.rid || '—' }}</code></span>
<span><i>{{ t('calllogs.colAccount') }}</i>{{ detail.account || '—' }}</span>
<span><i>Key</i>{{ detail.key_id ? detail.key_id + '…' : '—' }}</span>
<span><i>{{ t('calllogs.colModel') }}</i>{{ detail.model || '—' }}</span>
<span><i>{{ t('calllogs.colStatus') }}</i>{{ detail.status ?? '—' }} · {{ detail.ok ? 'OK' : 'FAIL' }}</span>
<span><i>{{ t('calllogs.colLatency') }}</i>{{ detail.latency_ms != null ? detail.latency_ms + 'ms' : '—' }}</span>
<span><i>{{ t('calllogs.colStream') }}</i>{{ detail.stream ? t('calllogs.yes') : t('calllogs.no') }}</span>
<span><i>IP</i>{{ detail.ip || '—' }}</span>
<span><i>UA</i>{{ detail.ua || '—' }}</span>
<span><i>path</i>{{ detail.path || '—' }}</span>
<span><i>ver</i>{{ detail.ver || '—' }}</span>
</div>
<div class="sec">
<label>{{ t('calllogs.prompt') }}</label>
<pre>{{ detail.prompt || '—' }}</pre>
</div>
<div class="sec">
<label>{{ t('calllogs.completion') }}</label>
<pre>{{ detail.completion || '—' }}</pre>
</div>
</div>
<div v-else class="sub" style="padding: 20px 0">{{ t('calllogs.loading') }}</div>
</el-dialog>
</div>
</template>
<script setup lang="ts">
import { computed, onMounted, ref } from 'vue'
import { useI18n } from 'vue-i18n'
import { api, getToken } from '../api'
import Icon from '../components/Icon.vue'
import { filenameFromHeader } from '../lib/download'
const { t } = useI18n()
const rows = ref<any[]>([])
const total = ref(0)
const loading = ref(false)
const page = ref(1)
const pageSize = 50
const range = ref<[string, string] | null>(null)
const account = ref('')
const model = ref('')
const okSel = ref('')
const kw = ref('')
const dlg = ref(false)
const detail = ref<any>(null)
const failCount = computed(() => rows.value.filter((r) => !r.ok).length)
const avgMs = computed(() => {
const v = rows.value.map((r) => Number(r.latency_ms)).filter((n) => Number.isFinite(n) && n > 0)
if (!v.length) return 0
return Math.round(v.reduce((a, b) => a + b, 0) / v.length)
})
function qs(extra: Record<string, string> = {}): string {
const p = new URLSearchParams()
if (range.value && range.value[0]) p.set('start', `${range.value[0]} 00:00`)
if (range.value && range.value[1]) p.set('end', `${range.value[1]} 23:59`)
if (account.value.trim()) p.set('account', account.value.trim())
if (model.value.trim()) p.set('model', model.value.trim())
if (okSel.value !== '') p.set('ok', okSel.value)
if (kw.value.trim()) p.set('q', kw.value.trim())
Object.entries(extra).forEach(([k, v]) => p.set(k, v))
return p.toString()
}
async function loadPage() {
loading.value = true
try {
const data = await api(`/logs/calls?${qs({ limit: String(pageSize), offset: String((page.value - 1) * pageSize) })}`)
rows.value = data.items || []
total.value = data.total || 0
} catch (e: any) {
rows.value = []
total.value = 0
} finally {
loading.value = false
}
}
function reload() {
page.value = 1
loadPage()
}
async function openDetail(row: any) {
detail.value = null
dlg.value = true
try {
detail.value = await api(`/logs/calls/${row.id}`)
} catch {
detail.value = row
}
}
async function exportCsv() {
const q = qs()
const resp = await fetch(`/logs/calls/export?${q}`, {
headers: { Authorization: `Bearer ${getToken()}` },
})
if (!resp.ok) return
const blob = await resp.blob()
const name = filenameFromHeader(resp.headers.get('content-disposition') || '') ||
`calllog-${new Date().toISOString().slice(0, 19).replace(/[:T]/g, '')}.csv`
const href = URL.createObjectURL(blob)
const a = document.createElement('a')
a.href = href
a.download = name
a.style.display = 'none'
document.body.appendChild(a)
a.click()
a.remove()
setTimeout(() => URL.revokeObjectURL(href), 10_000)
}
function rowCls({ row }: { row: any }) {
return row.ok ? '' : 'row-fail'
}
onMounted(loadPage)
</script>
<style scoped>
.calllogs { display: flex; flex-direction: column; gap: 12px; }
.filters { display: flex; align-items: center; gap: 8px; flex-wrap: wrap; }
.w-range { width: 250px; }
.w140 { width: 140px; }
.w170 { width: 170px; }
.w120 { width: 120px; }
.w200 { width: 200px; }
.spacer { flex: 1; }
.card { background: var(--surface); border: 1px solid var(--line); border-radius: 12px; padding: 12px 14px; }
.card-h { display: flex; align-items: center; gap: 10px; margin-bottom: 10px; }
.card-h b { font-size: 13.5px; }
.pager { display: flex; gap: 6px; margin-left: auto; }
.pill { font-size: 11.5px; padding: 2px 8px; border-radius: 999px; background: var(--chip); color: var(--muted-2); }
.pill.ok { color: var(--ok); }
.pill.warn { color: var(--err); }
.tbl :deep(.row-fail) { background: color-mix(in srgb, var(--err) 6%, transparent); }
.ok-t { color: var(--ok); font-variant-numeric: tabular-nums; }
.err-t { color: var(--err); font-weight: 600; font-variant-numeric: tabular-nums; }
.pager-bar { display: flex; justify-content: flex-end; padding-top: 10px; }
.sub { font-size: 11.5px; color: var(--muted-2); }
.kv { display: grid; grid-template-columns: repeat(3, minmax(0, 1fr)); gap: 6px 14px; font-size: 12.5px; }
.kv span { display: flex; gap: 6px; align-items: baseline; overflow: hidden; }
.kv i { font-style: normal; color: var(--muted-2); min-width: 52px; }
.kv code { font-size: 11.5px; }
.sec { margin-top: 12px; }
.sec label { font-size: 12px; color: var(--muted-2); }
.sec pre {
margin: 4px 0 0; padding: 10px 12px; max-height: 260px; overflow: auto;
background: var(--surface-2, #f7f8fa); border: 1px solid var(--line); border-radius: 8px;
font-size: 12px; line-height: 1.5; white-space: pre-wrap; word-break: break-word;
}
@media (max-width: 768px) {
.kv { grid-template-columns: 1fr; }
.w-range, .w140, .w170, .w120, .w200 { width: 100%; }
}
</style>
+28
View File
@@ -1,5 +1,14 @@
<template> <template>
<div class="page"> <div class="page">
<!-- 2026-10-23 调用日志落地:两页签 —— 实时(内存环形缓冲,重启即丢)/ 调用日志(SQLite 持久化 90 天,仅管理员) -->
<div class="tabs">
<span class="tab" :class="{ on: tab === 'live' }" @click="tab = 'live'">{{ t('logs.tabLive') }}</span>
<span v-if="isAdmin" class="tab" :class="{ on: tab === 'calls' }" @click="tab = 'calls'">
{{ t('logs.tabCalls') }}
</span>
</div>
<div v-show="tab === 'live'">
<!-- 筛选行:类型胶囊 + 原有筛选控件 + 刷新/导出 --> <!-- 筛选行:类型胶囊 + 原有筛选控件 + 刷新/导出 -->
<div class="filters"> <div class="filters">
<!-- 第十九轮:类型胶囊与「失败」互斥(原来两个维度叠加,「失败」亮着时点类型只剩空列表, <!-- 第十九轮:类型胶囊与「失败」互斥(原来两个维度叠加,「失败」亮着时点类型只剩空列表,
@@ -103,6 +112,9 @@
</div> </div>
</div> </div>
</div> </div>
<CallLogs v-if="tab === 'calls' && isAdmin" />
</div>
</template> </template>
<script setup lang="ts"> <script setup lang="ts">
@@ -112,6 +124,13 @@ import { useI18n } from 'vue-i18n'
import { api } from '../api' import { api } from '../api'
import Icon from '../components/Icon.vue' import Icon from '../components/Icon.vue'
import { stamp } from '../lib/download' import { stamp } from '../lib/download'
import { useSessionStore } from '../store/session'
import CallLogs from './CallLogs.vue'
const store = useSessionStore()
const isAdmin = computed(() => store.isAdmin)
// 'live' = 内存实时日志(原有页面);'calls' = SQLite 持久化的调用日志(仅管理员)
const tab = ref<'live' | 'calls'>('live')
const router = useRouter() const router = useRouter()
const { t } = useI18n() const { t } = useI18n()
@@ -335,6 +354,15 @@ onUnmounted(() => clearInterval(timer))
.sub { font-size: 11.5px; color: var(--muted-2); } .sub { font-size: 11.5px; color: var(--muted-2); }
.err-text { color: var(--err); font-size: 12px; } .err-text { color: var(--err); font-size: 12px; }
/* 2026-10-23: 顶部页签(实时 / 调用日志) */
.tabs { display: flex; gap: 6px; margin-bottom: 10px; border-bottom: 1px solid var(--line); }
.tabs .tab {
padding: 7px 14px; font-size: 13px; cursor: pointer; color: var(--muted-2);
border-bottom: 2px solid transparent; margin-bottom: -1px; user-select: none;
}
.tabs .tab:hover { color: var(--text); }
.tabs .tab.on { color: var(--primary); border-bottom-color: var(--primary); font-weight: 600; }
/* 类型胶囊上的条数:小一号、淡一档;胶囊选中时跟着主色走(.on 已给 font-weight) */ /* 类型胶囊上的条数:小一号、淡一档;胶囊选中时跟着主色走(.on 已给 font-weight) */
.chip-sel .cn { .chip-sel .cn {
font-style: normal; font-size: 11px; font-variant-numeric: tabular-nums; font-style: normal; font-size: 11px; font-variant-numeric: tabular-nums;