feat(runner): tgit-runner — сервисы в песочнице на хосте, деплой после push (#13)
CI / ruff · mypy · pytest (pull_request) Successful in 4m3s
CI / ruff · mypy · pytest (pull_request) Successful in 4m3s
- infra/runner/tgit-runner: deploy/restart/remove/list по SSH forced command; аргументы не исполняются и не глобятся; архив — только файлы и папки, без «..», ≤ 20 МБ, есть main.py; health_url — только localhost - песочница: uid 10001, read-only, tmpfs noexec, cap-drop ALL, no-new-privileges, 128 МБ без swap, 0.25 CPU, 64 PID, --network none (сеть — только по egress-allow админа), лимит 10 сервисов, 2 образа - новая версия стартует рядом (-next) и заменяет старую, только став healthy; иначе удаляется, старая работает, exit 3 - бот: после push git archive изменённых services/<name>/ → runner; автору — «🚀 работает» или «⛔ не развёрнут» с логом, команде — алерт ⛔; удалённая папка → remove; «♻️ Перезапустить» в «Статусе» для админа (S-5) - examples/escape_attempts: все попытки выбраться — BLOCKED на живом Docker - install.sh, authorized_keys.example; CD бота обновляет runner на хосте Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
@@ -4,6 +4,7 @@ import base64
|
||||
import html
|
||||
import io
|
||||
import logging
|
||||
from collections.abc import Awaitable, Callable
|
||||
from dataclasses import dataclass
|
||||
from enum import StrEnum
|
||||
from typing import Any
|
||||
@@ -33,6 +34,8 @@ from app.repo.validate import (
|
||||
python_syntax,
|
||||
strip_fence,
|
||||
)
|
||||
from app.runner.deployer import Deployer
|
||||
from app.runner.deployer import Outcome as DeployOutcome
|
||||
from app.texts import CommitTexts, Texts
|
||||
|
||||
log = logging.getLogger(__name__)
|
||||
@@ -212,7 +215,12 @@ def to_dir_keyboard(texts: Texts) -> InlineKeyboardMarkup:
|
||||
|
||||
|
||||
def build_router(
|
||||
texts: Texts, repo: UploadsRepo, drafts: DraftService, publisher: Publisher, commit_url_template: str = ""
|
||||
texts: Texts,
|
||||
repo: UploadsRepo,
|
||||
drafts: DraftService,
|
||||
publisher: Publisher,
|
||||
commit_url_template: str = "",
|
||||
deployer: Deployer | None = None,
|
||||
) -> Router:
|
||||
router = Router(name="commit")
|
||||
t = texts.commit
|
||||
@@ -275,6 +283,21 @@ def build_router(
|
||||
ops = {ChangeOp.ADD: t.op_add, ChangeOp.MODIFY: t.op_modify, ChangeOp.DELETE: t.op_delete}
|
||||
return "\n".join(f"{ops[c.op]} /{c.path}" for c in draft.changes)
|
||||
|
||||
def deploy_report(msg: Message, sha: str) -> Callable[[DeployOutcome], Awaitable[None]]:
|
||||
async def report(o: DeployOutcome) -> None:
|
||||
name = html.escape(o.service)
|
||||
if o.result.ok and o.removed:
|
||||
text = t.removed.format(service=name)
|
||||
elif o.result.ok:
|
||||
text = t.deployed.format(service=name, sha=sha[:7])
|
||||
else:
|
||||
text = t.deploy_failed.format(service=name, reason=html.escape(o.result.summary))
|
||||
if o.result.tail:
|
||||
text += f"\n<pre>{html.escape(o.result.tail[-3000:])}</pre>"
|
||||
await msg.answer(text, parse_mode=ParseMode.HTML)
|
||||
|
||||
return report
|
||||
|
||||
async def do_push(msg: Message, state: FSMContext, member: User, *, overwrite: bool) -> None:
|
||||
draft = await drafts.get(member.tg_id)
|
||||
if draft is None or not draft.changes or not draft.message:
|
||||
@@ -291,6 +314,12 @@ def build_router(
|
||||
if commit_url_template:
|
||||
lines.append(commit_url_template.format(sha=result.sha))
|
||||
lines.append(t.ci_pending)
|
||||
if deployer is not None:
|
||||
services = deployer.schedule(
|
||||
result.sha, [c.path for c in draft.changes], deploy_report(msg, result.sha)
|
||||
)
|
||||
if services:
|
||||
lines.append(t.deploying.format(services=", ".join(services)))
|
||||
await msg.edit_text("\n".join(lines))
|
||||
await publisher.record(member, result, draft.message, draft, msg.chat.id, msg.message_id)
|
||||
await drafts.clear(member.tg_id)
|
||||
|
||||
@@ -13,7 +13,8 @@ from aiogram.utils.keyboard import InlineKeyboardBuilder
|
||||
|
||||
from app.bot.menu import Nav, NavAction
|
||||
from app.bot.routers.commit import human_size
|
||||
from app.db.models import utcnow
|
||||
from app.db.models import Role, User, utcnow
|
||||
from app.runner.client import RunnerClient
|
||||
from app.runtime.checks import Checker, CheckName, ContainerReport
|
||||
from app.runtime.docker import Container, Details, DockerClient, DockerError, Stats
|
||||
from app.texts import StatusTexts, Texts
|
||||
@@ -27,6 +28,7 @@ class StatusAct(StrEnum):
|
||||
LIST = "l"
|
||||
CARD = "c"
|
||||
CHECK = "k"
|
||||
RESTART = "r"
|
||||
|
||||
|
||||
class StatusCb(CallbackData, prefix="s"):
|
||||
@@ -75,7 +77,10 @@ def render_list(texts: Texts, containers: list[Container], now: datetime) -> tup
|
||||
return _clip("\n".join(lines)), kb.as_markup()
|
||||
|
||||
|
||||
def render_card(texts: Texts, d: Details, stats: Stats | None, now: datetime) -> tuple[str, InlineKeyboardMarkup]:
|
||||
def render_card(
|
||||
texts: Texts, d: Details, stats: Stats | None, now: datetime, can_restart: bool = False
|
||||
) -> tuple[str, InlineKeyboardMarkup]:
|
||||
"""can_restart: an admin looking at a service deployed by tgit-runner (S-5)."""
|
||||
t = texts.status
|
||||
running = d.state == "running"
|
||||
uptime = duration(t, (now - d.started_at).total_seconds()) if running and d.started_at else t.none
|
||||
@@ -105,11 +110,18 @@ def render_card(texts: Texts, d: Details, stats: Stats | None, now: datetime) ->
|
||||
kb = InlineKeyboardBuilder()
|
||||
kb.button(text=t.check_button, callback_data=StatusCb(a=StatusAct.CHECK, id=d.id[:12]))
|
||||
kb.button(text=t.refresh, callback_data=StatusCb(a=StatusAct.CARD, id=d.id[:12]))
|
||||
if can_restart and runner_service(d):
|
||||
kb.button(text=t.restart, callback_data=StatusCb(a=StatusAct.RESTART, id=d.id[:12]))
|
||||
kb.button(text=t.to_list, callback_data=StatusCb(a=StatusAct.LIST))
|
||||
kb.adjust(2, 1)
|
||||
kb.adjust(2, 1, 1)
|
||||
return text, kb.as_markup()
|
||||
|
||||
|
||||
def runner_service(d: Details) -> str | None:
|
||||
"""Only services tgit-runner deployed can be restarted through it — not the bot, its DB or the proxy."""
|
||||
return d.labels.get("tgit.service") or None
|
||||
|
||||
|
||||
def render_report(
|
||||
texts: Texts, reports: list[ContainerReport], now: datetime, back_id: str = ""
|
||||
) -> tuple[str, InlineKeyboardMarkup]:
|
||||
@@ -141,7 +153,9 @@ def render_report(
|
||||
return _clip("\n".join(lines)), kb.as_markup()
|
||||
|
||||
|
||||
def build_router(texts: Texts, docker: DockerClient | None, checker: Checker | None) -> Router:
|
||||
def build_router(
|
||||
texts: Texts, docker: DockerClient | None, checker: Checker | None, runner: RunnerClient | None = None
|
||||
) -> Router:
|
||||
router = Router(name="status")
|
||||
t = texts.status
|
||||
|
||||
@@ -150,7 +164,7 @@ def build_router(texts: Texts, docker: DockerClient | None, checker: Checker | N
|
||||
raise DockerError("docker is not configured (DOCKER_URL)")
|
||||
return render_list(texts, await docker.containers(), utcnow())
|
||||
|
||||
async def card_screen(short_id: str) -> tuple[str, InlineKeyboardMarkup]:
|
||||
async def card_screen(short_id: str, is_admin: bool = False) -> tuple[str, InlineKeyboardMarkup]:
|
||||
if docker is None:
|
||||
raise DockerError("docker is not configured (DOCKER_URL)")
|
||||
cid = await docker.resolve(short_id)
|
||||
@@ -161,7 +175,7 @@ def build_router(texts: Texts, docker: DockerClient | None, checker: Checker | N
|
||||
stats = await docker.stats(cid)
|
||||
except DockerError:
|
||||
log.warning("stats %s unavailable", short_id)
|
||||
return render_card(texts, details, stats, utcnow())
|
||||
return render_card(texts, details, stats, utcnow(), can_restart=is_admin and runner is not None)
|
||||
|
||||
@router.message(F.text == texts.menu.status)
|
||||
async def open_status(message: Message, state: FSMContext) -> None:
|
||||
@@ -175,7 +189,7 @@ def build_router(texts: Texts, docker: DockerClient | None, checker: Checker | N
|
||||
await message.answer(text, reply_markup=kb)
|
||||
|
||||
@router.callback_query(StatusCb.filter())
|
||||
async def on_action(query: CallbackQuery, callback_data: StatusCb) -> None:
|
||||
async def on_action(query: CallbackQuery, callback_data: StatusCb, member: User) -> None:
|
||||
if not isinstance(query.message, Message):
|
||||
await query.answer()
|
||||
return
|
||||
@@ -185,7 +199,23 @@ def build_router(texts: Texts, docker: DockerClient | None, checker: Checker | N
|
||||
case StatusAct.LIST:
|
||||
text, kb = await list_screen()
|
||||
case StatusAct.CARD:
|
||||
text, kb = await card_screen(callback_data.id)
|
||||
text, kb = await card_screen(callback_data.id, member.role is Role.ADMIN)
|
||||
case StatusAct.RESTART:
|
||||
if docker is None or runner is None or member.role is not Role.ADMIN:
|
||||
await query.answer()
|
||||
return
|
||||
service = runner_service(await docker.inspect(await docker.resolve(callback_data.id)))
|
||||
if service is None:
|
||||
await query.answer()
|
||||
return
|
||||
result = await runner.restart(service)
|
||||
if not result.ok:
|
||||
await query.answer(t.restart_failed.format(reason=result.summary)[:200], show_alert=True)
|
||||
return
|
||||
await query.answer(t.restarted)
|
||||
text, kb = await card_screen(callback_data.id, True)
|
||||
await _edit(msg, text, kb)
|
||||
return
|
||||
case StatusAct.CHECK:
|
||||
if docker is None or checker is None:
|
||||
raise DockerError("docker is not configured (DOCKER_URL)")
|
||||
|
||||
Reference in New Issue
Block a user