feat(alerts): алерты всей команде — детекторы, дедупликация, список (#15)
CI / ruff · mypy · pytest (pull_request) Successful in 4m49s

- детекторы по состоянию после каждого раунда снапшотов: краш, OOM,
  цикл рестартов, unhealthy, перегрузка ≥ 1 мин, мало места, Docker
  недоступен; пороги в config/alerts.yaml
- AlertService: ключ kind+service, повтор в 10 мин — счётчик в исходном
  сообщении, «✅ восстановлено» ответом; разовые события (🛡) со счётчиком
- рассылка всем активным (🛡 — только админам), темп ~25/с, RetryAfter,
  заблокировавший бота пропускается; alert_deliveries для правок и ответов
- «🚨 Алерты»: активные + 7 дней по 10, карточка; кнопки «📜 Логи»,
  «🩺 Статус» под алертом открывают новое сообщение, а не затирают алерт
- 🛡 атаки через агента идут через общий сервис алертов
- убраны заглушки «Раздел в разработке» — у всех разделов свои роутеры

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-10-17 15:36:09 +03:00
co-authored by Claude Opus 5.5
parent 65acf3abf1
commit 6c6f766abc
24 changed files with 1355 additions and 107 deletions
+15 -22
View File
@@ -6,19 +6,19 @@ import logging
from datetime import timedelta
from enum import StrEnum
from aiogram import Bot, F, Router
from aiogram import F, Router
from aiogram.enums import ParseMode
from aiogram.exceptions import TelegramAPIError
from aiogram.filters.callback_data import CallbackData
from aiogram.fsm.context import FSMContext
from aiogram.fsm.state import State, StatesGroup
from aiogram.types import BufferedInputFile, CallbackQuery, InlineKeyboardMarkup, Message
from aiogram.utils.keyboard import InlineKeyboardBuilder
from app.access.service import AccessService
from app.agent.service import Agent, Outcome
from app.alerts.detectors import Condition
from app.alerts.service import AlertService
from app.bot.menu import Nav, NavAction, free_input
from app.db.models import LogMode, Role, User, utcnow
from app.db.models import AlertKind, LogMode, User, utcnow
from app.logs.grep import GrepError, GrepErrorCode, GrepResult, parse_grep, run, to_command
from app.logs.journal import LogJournal
from app.runtime.docker import Container, DockerClient, DockerError
@@ -146,7 +146,7 @@ def build_router(
docker: DockerClient | None,
journal: LogJournal,
agent: Agent | None = None,
access: AccessService | None = None,
alerts: AlertService | None = None,
) -> Router:
router = Router(name="logs")
t = texts.logs
@@ -156,23 +156,16 @@ def build_router(
raise DockerError("docker is not configured (DOCKER_URL)")
return docker
async def alert_admins(bot: Bot, member: User, prompt: str, stage: str, reason: str) -> None:
"""🛡 to the admins only (SPEC AL-2, L-12). The asker sees just a polite refusal."""
if access is None:
async def alert_admins(member: User, prompt: str, stage: str, reason: str) -> None:
"""🛡 to the admins only (SPEC AL-2, L-12); repeats by the same member bump one alert's counter."""
if alerts is None:
return
text = t.attack_alert.format(
who=html.escape(who(member)),
stage=stage,
reason=html.escape(reason),
prompt=html.escape(prompt[:1000]),
condition = Condition(
AlertKind.AGENT_ATTACK,
f"user {member.tg_id}",
{"who": who(member), "stage": stage, "reason": reason, "log": prompt[:1000]},
)
for admin in await access.list_members():
if admin.role is not Role.ADMIN:
continue
try:
await bot.send_message(admin.tg_id, text, parse_mode=ParseMode.HTML)
except TelegramAPIError as e: # one blocked admin must not hide the alert from the others
log.warning("attack alert to %s failed: %s", admin.tg_id, e)
await alerts.event(condition, admins_only=True)
@router.message(F.text == texts.menu.logs)
async def open_logs(message: Message, state: FSMContext) -> None:
@@ -275,7 +268,7 @@ def build_router(
await deliver(message, texts, head, query.count, result, container_keyboard(texts, c.short_id), c.name)
@router.message(LogsFlow.agent, F.text, free_input(texts))
async def got_question(message: Message, state: FSMContext, member: User, bot: Bot) -> None:
async def got_question(message: Message, state: FSMContext, member: User) -> None:
if agent is None: # pragma: no cover - the button doesn't open this state without an agent
await state.clear()
return
@@ -296,7 +289,7 @@ def build_router(
match answer.outcome:
case Outcome.ATTACK:
await message.answer(t.agent_refused, reply_markup=kb)
await alert_admins(bot, member, prompt, answer.stage, answer.reason)
await alert_admins(member, prompt, answer.stage, answer.reason)
return
case Outcome.QUOTA:
await message.answer(t.agent_quota.format(quota=agent.quota), reply_markup=kb)