✨ feat: WP-14 运营收口——诊断双授权、管理面治理、支持包、观测与自托管

- 迁移 16:diagnostic_grant(token 哈希唯一、15 分钟、单次、绑定 Run)+ user.role(user/admin 检查约束)
- 诊断双授权(ADR-004):本人签发 grant(明文 token 只返回一次,库存 SHA-256);读取需登录态 + Hunjing-Diagnostic-Token + Purpose 三重匹配;单次消耗(重放 401)、用途不符 403、响应 no-store/private;返回观测摘要(usage/TTFT/工具轨迹)不含消息正文
- 管理面:pending 司议案列表与处置(content_removed 联动下架目标内容——治理闭环);非管理员 403
- 脱敏支持包:统计聚合(runs/messages/tool/media 按状态),无正文无密钥
- 观测:GET /readyz(DB 连通)与 GET /metrics(Prometheus 文本,真实 SQL 聚合)
- 自托管:docker/Dockerfile.backend(uv 构建)+ docker-compose.app.yml(hunjing-api + hunjing-postgres 健康依赖、密码经 .env);Makefile 新增 selfhost-up/down、db-dump/db-restore(备份到 backups/)
- 测试 94 项全绿(+3):诊断全路径(无令牌/用途不符/成功/单次重放)、管理面处置联动、支持包与指标
This commit is contained in:
chenxin committed 2026-09-12 11:20:15 +08:00
1 parent 701c20c783
commit 21eacc8c75
10 files changed
+643

No files matched your search

+147
View File
@@ -0,0 +1,147 @@
"""运营域路由:诊断、管理面、支持包与观测(WP-14)。"""
from typing import Annotated, Any
from fastapi import APIRouter, Depends, Header, HTTPException, Request, Response, status
from sqlalchemy import func, select
from sqlalchemy import text as sql_text
from app.auth.dependencies import get_current_user, require_state_change
from app.conversations.routes import _domain_of
from app.db.session import DbSession
from app.models import MediaJob, Message, Run, ToolCall, User
from app.ops import service
ops_router = APIRouter(prefix="/v1", tags=["ops-extended"])
admin_router = APIRouter(prefix="/v1/admin", tags=["admin"])
@ops_router.post("/runs/{run_id}/diagnostic-grants", status_code=201)
async def issue_grant(
run_id: str,
db: DbSession,
user: Annotated[User, Depends(require_state_change)],
request: Request,
body: dict[str, str],
) -> dict[str, Any]:
"""签发诊断授权:明文 token 只返回一次;默认 15 分钟、单次、绑定 Run。"""
domain = await _domain_of(db, user)
grant, token = await service.issue_grant(
db, domain.id, run_id, body.get("purpose", ""),
getattr(request.state, "trace_id", None),
)
return {
"grant_id": grant.id, "token": token, # 唯一一次明文返回
"expires_at": grant.expires_at.isoformat(),
}
@ops_router.get("/runs/{run_id}/diagnostics")
async def read_diagnostics(
run_id: str,
db: DbSession,
user: Annotated[User, Depends(get_current_user)],
response: Response,
diag_token: Annotated[
str | None, Header(alias="Hunjing-Diagnostic-Token")
] = None,
diag_purpose: Annotated[
str | None, Header(alias="Hunjing-Diagnostic-Purpose")
] = None,
) -> dict[str, Any]:
"""双重授权诊断:登录态 + 诊断令牌;响应不落缓存(ADR-004)。"""
domain = await _domain_of(db, user)
if not diag_token or not diag_purpose:
raise HTTPException(status.HTTP_401_UNAUTHORIZED, "缺少诊断令牌或用途")
result = await service.read_diagnostics(
db, domain.id, run_id, diag_token, diag_purpose
)
response.headers["Cache-Control"] = "no-store, private"
return result
@ops_router.post("/support/bundles")
async def support_bundle(
db: DbSession,
user: Annotated[User, Depends(require_state_change)],
) -> dict[str, Any]:
"""脱敏支持包:统计聚合,无正文与密钥。"""
domain = await _domain_of(db, user)
return await service.build_support_bundle(db, domain.id)
@admin_router.get("/moderation-cases")
async def list_cases(
db: DbSession,
user: Annotated[User, Depends(get_current_user)],
) -> dict[str, Any]:
await service.require_admin(db, user.id)
from app.models import Report
cases = await service.list_pending_cases(db)
out = []
for case in cases:
report = await db.get(Report, case.report_id)
out.append({
"case_id": case.id, "status": case.status,
"report": {
"target_type": report.target_type if report else None,
"target_id": report.target_id if report else None,
"reason": report.reason if report else None,
} if report else None,
})
return {"data": out}
@admin_router.post("/moderation-cases/{case_id}/resolutions")
async def resolve_case(
case_id: str,
db: DbSession,
user: Annotated[User, Depends(require_state_change)],
request: Request,
body: dict[str, str],
) -> dict[str, Any]:
domain = await _domain_of(db, user)
await service.require_admin(db, user.id)
case = await service.resolve_case(
db, domain.id, case_id, body.get("action", ""),
getattr(request.state, "trace_id", None),
)
return {"case_id": case.id, "status": case.status, "action_taken": case.action_taken}
@ops_router.get("/readyz", tags=["ops"])
async def readyz(db: DbSession) -> dict[str, str]:
await db.execute(sql_text("SELECT 1"))
return {"status": "ready"}
@ops_router.get("/metrics", tags=["ops"])
async def metrics(db: DbSession) -> Response:
"""Prometheus 文本格式:真实 SQL 聚合(无正文)。"""
run_rows = (await db.execute(select(Run.status, func.count()).group_by(Run.status))).all()
msg_total = (await db.execute(select(func.count()).select_from(Message))).scalar_one()
tool_rows = (
await db.execute(select(ToolCall.status, func.count()).group_by(ToolCall.status))
).all()
media_rows = (
await db.execute(select(MediaJob.status, func.count()).group_by(MediaJob.status))
).all()
lines = ["# HELP hunjing_runs_total Run 总数(按状态)", "# TYPE hunjing_runs_total counter"]
for st, cnt in run_rows:
lines.append(f'hunjing_runs_total{{status="{st}"}} {int(cnt)}')
lines += [
"# HELP hunjing_messages_total 消息总数",
"# TYPE hunjing_messages_total counter",
f"hunjing_messages_total {int(msg_total)}",
]
groups = [
("hunjing_tool_calls_total", tool_rows),
("hunjing_media_jobs_total", media_rows),
]
for prefix, rows in groups:
lines.append(f"# HELP {prefix} 计数(按状态)")
lines.append(f"# TYPE {prefix} counter")
for st, cnt in rows:
lines.append(f'{prefix}{{status="{st}"}} {int(cnt)}')
return Response(content="\n".join(lines) + "\n", media_type="text/plain")