fix: слоты, безопасность, единый формат ответа и валидация

- слоты браузеров: asyncio.Semaphore, размер из BROWSER_SLOTS, при нехватке — 503 + Retry-After;
  устранена утечка счётчика при отказе (#12, #49)
- session_id принимается только как UUID, путь профиля проверяется (#5)
- секреты клиента в логах маскируются, полный дамп — только на DEBUG (#6)
- ответ: status всегда "ok"/"err", поле code (no_slots / request_failed / challenge_failed /
  action_failed), response_model (#7)
- непройденный антибот → challenge_failed, действия не выполняются (#8)
- таймауты провайдеров в мс вынесены в константы; detect переживает навигацию посреди
  проверки; пауза между неудачными solve вместо холостого цикла (#9)
- обязательные поля действий и формат прокси проверяются схемой → 422 до запуска браузера (#13)
- прокси необязателен (#15)
- user_agent сохраняется в профиле сессии и реально передаётся в браузер (#16)
- duration_ms у действий и у запроса (#18), опция screenshot (#19)
- общий конфиг логирования (LOG_LEVEL), print → логгер, CloudflareProvider,
  удалён cached_persistent, GET /health (#22)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

dot_env

@@ -1,4 +1,4 @@
 API_PORT=
 API_CPUS=1
 API_MEM_LIMIT=1g
-API_MEM_LIMIT=1g
+BROWSER_SLOTS=1

env.example

@@ -1 +1,9 @@
 # Project Env
+# порт, CPU и память контейнера (docker-compose)
+API_PORT=8000
+API_CPUS=4
+API_MEM_LIMIT=8g
+# сколько Chrome параллельно, ≈ числу ядер (docs/Deployment.md)
+BROWSER_SLOTS=4
+NO_SLOTS_RETRY_AFTER=5
+LOG_LEVEL=INFO

src/antibot/humanize.py

@@ -85,7 +85,6 @@ async def move_mouse_human(
 		if i < steps:
 			px += random.uniform(-1.5, 1.5)
 			py += random.uniform(-1.5, 1.5)
-		print("Move:", px, py)
 		await page.mouse.move(px, py)
 		await page.wait_for_timeout(step_delay_ms * random.uniform(0.7, 1.3))
 
@@ -103,13 +102,9 @@ async def click_human(
 	"""move_mouse_human + клик с человеческой задержкой между наведением и down/up, вместо
 	мгновенного page.mouse.click(x, y)."""
 	await move_mouse_human(page, x, y, **move_kwargs)
-	print('wait for timeout')
 	await page.wait_for_timeout(random.uniform(50, 180))
-	print('key down')
 	await page.mouse.down(button=button)
-	print('sleep')
 	await page.wait_for_timeout(random.uniform(40, 99))
-	print('key up')
 	await page.mouse.up(button=button)
 
 

src/antibot/known_providers/cloudflare.py

@@ -72,8 +72,6 @@ from ..registry import register
 from ..humanize import wait_for_page_stability, wait_for_networkidle, click_human
 
 log = logging.getLogger(__name__)
-log.setLevel(logging.DEBUG)
-log.addHandler(logging.StreamHandler())
 
 
 class CloudflareDetector(Detector):
@@ -185,9 +183,12 @@ __CF_PATTERN__ = re.compile(r"^https://challenges\.cloudflare\.com/cdn-cgi/chall
 
 
 class CloudflareSolver(Solver):
+	# опрос исчезновения маркера челленджа после клика / на interstitial
+	POLL_INTERVAL_MS = 250
+	WAIT_CLEARED_S = 10
 
 	async def solve(self, page: Page, info: ChallengeInfo):
-		print("start solving")
+		log.debug("start solving")
 		iframe = page.frame(url=__CF_PATTERN__)
 		if iframe:
 			await page.wait_for_timeout(randint(500, 1000))
@@ -212,11 +213,11 @@ class CloudflareSolver(Solver):
 				finally:
 					return
 			else:
-				deadline = time.time() + 10
+				deadline = time.time() + self.WAIT_CLEARED_S
 				while time.time() < deadline:
 					if not await page.query_selector(info.detect_locator):
 						return
-					await page.wait_for_timeout(1)
+					await page.wait_for_timeout(self.POLL_INTERVAL_MS)
 			return
 		elif info.stage == Stage.INTERSTITIAL:
 			if info.success_locator:
@@ -227,11 +228,11 @@ class CloudflareSolver(Solver):
 				else:
 					return
 			else:
-				deadline = time.time() + 10
+				deadline = time.time() + self.WAIT_CLEARED_S
 				while time.time() < deadline:
 					if not await page.query_selector(info.detect_locator):
 						return
-					await page.wait_for_timeout(1)
+					await page.wait_for_timeout(self.POLL_INTERVAL_MS)
 				return
 		else:
 
@@ -257,7 +258,7 @@ class CloudflareSolver(Solver):
 
 
 @register
-class ReCaptchaProvider(AntiBotProvider):
+class CloudflareProvider(AntiBotProvider):
 	provider = 'cloudflare'
 	_solver_cls = CloudflareSolver
 	_detector_cls = CloudflareDetector

src/antibot/known_providers/recaptcha.py

@@ -11,8 +11,6 @@ from ..registry import register
 from ..exceptions import IframeLoadError
 
 log = logging.getLogger(__name__)
-log.setLevel(logging.DEBUG)
-log.addHandler(logging.StreamHandler())
 
 
 JS_CHALLENGE_SUBMIT = """(token) => {
@@ -36,6 +34,8 @@ JS_CHALLENGE_SUBMIT = """(token) => {
 
 class ReCaptchaSolver(Solver):
 	page: Page
+	# пауза на стадии INTERSTITIAL перед повторным detect
+	INTERSTITIAL_WAIT_MS = 15_000
 
 	def __init__(self, captcha_solver: Captcha, **kwargs) -> None:
 		super().__init__(captcha_solver=captcha_solver, **kwargs)
@@ -49,8 +49,7 @@ class ReCaptchaSolver(Solver):
 		self.page = page
 		log.debug("Начинаю решение recaptcha антибот преграды")
 		if info.stage == Stage.INTERSTITIAL:
-			# TODO: вытащить
-			await page.wait_for_timeout(15)
+			await page.wait_for_timeout(self.INTERSTITIAL_WAIT_MS)
 			return
 		elif info.stage == Stage.CHECKBOX:
 			log.debug("Нужно тыкнуть чекбокс")

src/antibot/orchestrator.py

@@ -3,9 +3,10 @@
 """
 import logging
 import time
-import traceback
 from dataclasses import dataclass
 
+from patchright.async_api import Error as PlaywrightError
+
 from captcha import load_captcha
 
 from .base import Stage
@@ -13,8 +14,8 @@ from .registry import select_one
 from .exceptions import ProviderNotDetected
 
 log = logging.getLogger(__name__)
-log.setLevel(logging.DEBUG)
-log.addHandler(logging.StreamHandler())
+
+SOLVE_RETRY_DELAY_MS = 1000
 
 
 @dataclass
@@ -53,7 +54,14 @@ async def pass_challenges(
 
 	info = None
 	while time.monotonic() < deadline:
-		info = await provider.detect(page)  # noqa
+		try:
+			info = await provider.detect(page)  # noqa
+		except PlaywrightError as e:
+			# челлендж сам переходит на целевую страницу посреди detect()
+			# («Execution context was destroyed») — даём навигации закончиться и смотрим заново
+			log.debug("%s: detect interrupted (%s), retrying", provider.provider, e.message.splitlines()[0])
+			await page.wait_for_load_state("load")
+			continue
 		# 3. проверяем что решение успешно
 		if info.success:  # noqa
 			return ChallengeOutcome(True, reason=(info.meta or {}).get("reason"))
@@ -61,7 +69,10 @@ async def pass_challenges(
 		try:
 			await provider.solve(page, info)
 		except Exception as e:
-			traceback.print_exc()
+			log.warning("%s: solve failed (%s), retrying", provider.provider, str(e).splitlines()[0] if str(e) else type(e).__name__)
+			log.debug("solve traceback", exc_info=True)
+			# пауза, иначе при стабильно падающем solve цикл крутится вхолостую до дедлайна
+			await page.wait_for_timeout(SOLVE_RETRY_DELAY_MS)
 			continue
 	else:
 		return ChallengeOutcome(

src/api/actions.py

@@ -1,8 +1,8 @@
 import asyncio
+import time
 
 from patchright.async_api import Page
 
-from antibot.base import click_and_race
 from antibot.orchestrator import pass_challenges
 from captcha import load_captcha
 from api.schemas import ActionRequest
@@ -10,50 +10,52 @@ from api.schemas import ActionRequest
 _actions: dict = {}
 
 
+class ChallengeFailed(Exception):
+	pass
+
+
 def register(fun) -> None:
 	_actions[fun.__name__] = fun
 	return fun
 
 
+async def _query(page: Page, selector: str):
+	element = await page.query_selector(selector)
+	if element is None:
+		raise LookupError(f"selector not found: {selector}")
+	return element
+
+
 @register
 async def fill(page: Page, act: ActionRequest, **kwargs) -> str | None:
-	assert act.selector
-	assert act.value
 	await page.fill(act.selector, act.value)
 
 
 @register
 async def captcha(page: Page, act: ActionRequest, captcha_key: str, **kwargs) -> str | None:
-	assert act.selector
-	assert act.value
-	assert captcha_key
 	captcha_solver = load_captcha(captcha_key, act.value)
 	await captcha_solver.solve(act.selector, page)
 
 
 @register
 async def delay(page: Page, act: ActionRequest, **kwargs) -> str | None:
-	assert act.value
 	await asyncio.sleep(int(act.value))
 
 
 @register
 async def mouse_move(page: Page, act: ActionRequest, **kwargs) -> str | None:
-	assert act.selector
-	element = await page.query_selector(act.selector)
+	element = await _query(page, act.selector)
 	await element.hover(scroll="auto")
 
 
 @register
 async def click(page: Page, act: ActionRequest, **kwargs) -> str | None:
-	assert act.selector
 	element = page.locator(act.selector)
 	await element.click()
 
 
 @register
 async def submit(page: Page, act: ActionRequest, **kwargs) -> str | None:
-	assert act.selector
 	element = page.locator(act.selector)
 
 	async with page.expect_event(
@@ -65,15 +67,12 @@ async def submit(page: Page, act: ActionRequest, **kwargs) -> str | None:
 
 @register
 async def inner_html(page: Page, act: ActionRequest, **kwargs) -> str | None:
-	assert act.selector
-	element = await page.query_selector(act.selector)
+	element = await _query(page, act.selector)
 	return await element.inner_html()
 
 
 @register
 async def wait_for(page: Page, act: ActionRequest, **kwargs) -> str | None:
-	assert act.selector
-	assert 1 <=  act.timeout <= 7200
 	await page.wait_for_selector(act.selector, timeout=act.timeout * 1000)
 
 
@@ -99,7 +98,10 @@ async def execute(
 	captcha_key: str | None = None, antibot_name: str | None = None,
 	success_locator: str | None = None,
 	detect_locator: str | None = None) -> list[dict]:
-	"""Выполнение очереди автоматизации"""
+	"""Выполнение очереди автоматизации.
+
+	Если антибот не пройден — бросает ChallengeFailed, действия не выполняются: на странице-заглушке
+	они бы упали с невнятной ошибкой селектора."""
 	actions_result = []
 	ctx = {
 		"captcha_key": captcha_key,
@@ -107,27 +109,32 @@ async def execute(
 		"success_locator": success_locator,
 		"detect_locator": detect_locator,
 	}
-	await pass_challenges(
+	outcome = await pass_challenges(
 		antibot_name, page,
 		captcha=captcha_key,
 		success_locator=success_locator,
 		detect_locator=detect_locator,
 	)
+	if not outcome.cleared:
+		stage = outcome.last_stage.value if outcome.last_stage else None
+		raise ChallengeFailed(f"{antibot_name} not passed: {outcome.reason} (stage: {stage})")
+
 	for act in actions:
 		new_line = {
 			"type": act.type,
 		}
+		started = time.monotonic()
 		try:
-			# TODO: добавить время выполнения
 			result = await _actions[act.type](page, act, **ctx)
 		except Exception as e:
 			new_line["ok"] = False
-			new_line["result"] = str(e)
-			actions_result.append(new_line)
-			break
+			new_line["result"] = str(e) or type(e).__name__
 		else:
 			new_line["ok"] = True
 			if result:
 				new_line["result"] = result
-			actions_result.append(new_line)
-	return actions_result if len(actions_result) else [{"status": "pass"}]
+		new_line["duration_ms"] = int((time.monotonic() - started) * 1000)
+		actions_result.append(new_line)
+		if not new_line["ok"]:
+			break
+	return actions_result

src/api/schemas.py

@@ -1,25 +1,70 @@
-from typing import Any
+from typing import Any, Literal
+from urllib.parse import urlsplit
 
-from pydantic import BaseModel
+from pydantic import BaseModel, field_validator, model_validator
+
+CAPTCHA_TYPES = ("cloudflare", "recaptcha", "hcaptcha")
+PROXY_SCHEMES = ("http", "https", "socks5")
+
+# тип действия -> обязательные поля (кроме type)
+_REQUIRED_FIELDS: dict[str, tuple[str, ...]] = {
+	"fill": ("selector", "value"),
+	"captcha": ("selector", "value"),
+	"delay": ("value",),
+	"mouse_move": ("selector",),
+	"click": ("selector",),
+	"submit": ("selector",),
+	"inner_html": ("selector",),
+	"wait_for": ("selector",),
+	"pass_challenge": (),
+}
+WAIT_FOR_DEFAULT_TIMEOUT = 30
 
 
 class ActionRequest(BaseModel):
 	"""
 	Действие автоматизации. Характеризуется типом и производится над селектором.
 	Набор параметров зависит от типа, этот класс перечисляет все параметры.
+	Обязательные поля для каждого типа проверяются здесь же — неверное действие
+	отклоняется с 422 до запуска браузера.
 	"""
 	type: str
 	selector: str | None = None
 	value: str | None = None
 	timeout: int | None = None
 
+	@model_validator(mode="after")
+	def check_fields(self):
+		if self.type not in _REQUIRED_FIELDS:
+			raise ValueError(f"unknown action type {self.type!r}, expected one of: {', '.join(_REQUIRED_FIELDS)}")
+
+		missing = [f for f in _REQUIRED_FIELDS[self.type] if getattr(self, f) in (None, "")]
+		if missing:
+			raise ValueError(f"action {self.type!r} requires: {', '.join(missing)}")
+
+		if self.type == "wait_for":
+			if self.timeout is None:
+				self.timeout = WAIT_FOR_DEFAULT_TIMEOUT
+			if not 1 <= self.timeout <= 7200:
+				raise ValueError("wait_for timeout must be between 1 and 7200 seconds")
+		elif self.type == "delay":
+			try:
+				delay = int(self.value)
+			except ValueError:
+				raise ValueError("delay value must be an integer number of seconds") from None
+			if not 0 <= delay <= 600:
+				raise ValueError("delay value must be between 0 and 600 seconds")
+		elif self.type == "captcha" and self.value not in CAPTCHA_TYPES:
+			raise ValueError(f"captcha value must be one of: {', '.join(CAPTCHA_TYPES)}")
+		return self
+
 
 class SolveRequest(BaseModel):
 	"""
 	Объект запроса для автоматизации на сайте.
 	"""
 	url: str
-	proxy: str
+	proxy: str | None = None
 	timeout: int | None = None
 	user_agent: str | None = None
 	cookies: list[dict[str, Any]] | None = None
@@ -28,7 +73,7 @@ class SolveRequest(BaseModel):
 	actions: list[ActionRequest]
 
 	html: bool = False
-	# TODO: screenshot: bool = False
+	screenshot: bool = False
 
 	captcha: str | None = None
 	captcha_detect_locator: str | None = None
@@ -43,3 +88,48 @@ class SolveRequest(BaseModel):
 	retry_delay: float = 1.0
 	wait: int = 0
 	google_search: bool = False
+
+	@field_validator("proxy")
+	@classmethod
+	def check_proxy(cls, proxy: str | None) -> str | None:
+		if not proxy:
+			return None
+		try:
+			parts = urlsplit(proxy)
+			port = parts.port
+		except ValueError as e:
+			raise ValueError(f"invalid proxy url: {e}") from None
+		if parts.scheme not in PROXY_SCHEMES:
+			raise ValueError(f"proxy scheme must be one of: {', '.join(PROXY_SCHEMES)}")
+		if not parts.hostname or not port:
+			raise ValueError("proxy must be in form scheme://[user:pass@]host:port")
+		return proxy
+
+	@model_validator(mode="after")
+	def check_captcha_key(self):
+		if not self.captcha and any(a.type == "captcha" for a in self.actions):
+			raise ValueError("action 'captcha' requires request field 'captcha' (2captcha key)")
+		return self
+
+
+class ActionResult(BaseModel):
+	type: str
+	ok: bool
+	result: str | None = None
+	duration_ms: int
+
+
+class SolveResponse(BaseModel):
+	"""
+	Ответ /v1/solve. status — "ok" (антибот пройден, все действия выполнены) или "err".
+	code — машиночитаемая причина ошибки, error — текст для человека.
+	"""
+	status: Literal["ok", "err"]
+	code: Literal["no_slots", "request_failed", "challenge_failed", "action_failed"] | None = None
+	error: str | None = None
+	user_agent: str | None = None
+	cookies: list[dict[str, Any]] | None = None
+	actions: list[ActionResult] = []
+	html: str | None = None
+	screenshot: str | None = None
+	duration_ms: int | None = None

src/captcha/base.py

@@ -1,48 +1,9 @@
 import asyncio
-import json
-import hashlib
-import time
-from pathlib import Path
 
 from patchright.async_api import Page
 from twocaptcha import TwoCaptcha
 
 
-def cached_persistent(file_name=None):
-
-	def key(*args, **kwargs):
-		return hashlib.sha256(f'{str(args)}-{str(kwargs)}'.encode()).hexdigest()
-
-	def wrapper(fun):
-
-		def clear_cache():
-			cache_file.write_text(json.dumps({}))
-
-		cache_file = Path(file_name)
-		if not cache_file.exists():
-			cache_file.write_text(json.dumps({}))
-
-		async def wrapped(*args, **kwargs):
-			data = json.loads(cache_file.read_text())
-			_key = key(fun.__name__, *args, **kwargs)
-
-			if cached_value := data.get(_key):
-				cached_value, cache_time = cached_value
-				if time.time() - cache_time > 300:
-					cached_value = None
-
-			value = cached_value or await fun(*args, **kwargs)
-
-			data[_key] = [value, time.time()]
-			cache_file.write_text(json.dumps(data))
-			return value
-
-		setattr(wrapped, '_clear_cache', clear_cache)
-		return wrapped
-
-	return wrapper
-
-
 class Captcha:
 	ctype = "captcha"
 

src/captcha/hcaptcha.py

@@ -16,8 +16,6 @@ import logging
 from .base import Captcha, Page
 
 logger = logging.getLogger(__name__)
-logger.setLevel(logging.DEBUG)
-logger.addHandler(logging.StreamHandler())
 
 
 class SolverHCaptcha(Captcha):

src/captcha/recaptcha.py

@@ -7,8 +7,6 @@ from urllib.parse import urlparse, parse_qs
 from .base import Captcha, Page
 
 logger = logging.getLogger(__name__)
-logger.setLevel(logging.DEBUG)
-logger.addHandler(logging.StreamHandler())
 
 
 class SolverReCaptcha(Captcha):

src/captcha/turnstile.py

@@ -16,8 +16,6 @@ import logging
 from .base import Captcha, Page
 
 logger = logging.getLogger(__name__)
-logger.setLevel(logging.DEBUG)
-logger.addHandler(logging.StreamHandler())
 
 
 class SolverTurnstile(Captcha):

src/main.py

@@ -1,6 +1,9 @@
+import asyncio
+import base64
 import json
 import logging
-import traceback
+import os
+import time
 import uuid
 from pathlib import Path
 from typing import Annotated
@@ -8,19 +11,29 @@ from urllib.parse import urlsplit
 
 from fastapi import Cookie
 from fastapi import FastAPI
-from fastapi import Request
 from fastapi import Response
+from fastapi.responses import JSONResponse
 from patchright.async_api import Page, ProxySettings
 
-from api import actions
-from api.geoip import resolve_geo
-from api.schemas import SolveRequest
-from engine import AsyncStealthySession
+logging.basicConfig(
+	level=os.getenv("LOG_LEVEL", "INFO").upper(),
+	format="%(asctime)s %(levelname)s %(name)s: %(message)s",
+)
+
+from api import actions  # noqa: E402 — модули ниже логируют при импорте, конфиг логов должен быть раньше
+from api.geoip import resolve_geo  # noqa: E402
+from api.schemas import SolveRequest, SolveResponse  # noqa: E402
+from engine import AsyncStealthySession  # noqa: E402
 
 logger = logging.getLogger(__name__)
-logger.setLevel(logging.DEBUG)
-logger.addHandler(logging.StreamHandler())
-BROWSER_SLOTS = 10
+
+# сколько Chrome запускается параллельно; подбирать ≈ числу ядер (см. docs/Deployment.md)
+BROWSER_SLOTS = int(os.getenv("BROWSER_SLOTS", "10"))
+RETRY_AFTER = os.getenv("NO_SLOTS_RETRY_AFTER", "5")
+_slots = asyncio.Semaphore(BROWSER_SLOTS)
+
+USER_DATA_ROOT = (Path(__file__).parent / "../extra/user_data_dir").resolve()
+SESSION_SETTINGS = "robotex_session.json"
 
 
 def parse_proxy(proxy_url: str) -> ProxySettings:
@@ -34,23 +47,97 @@ def parse_proxy(proxy_url: str) -> ProxySettings:
 	return settings
 
 
+def mask_proxy(proxy_url: str | None) -> str | None:
+	if not proxy_url:
+		return proxy_url
+	parts = urlsplit(proxy_url)
+	return f"{parts.scheme}://{'***@' if parts.username else ''}{parts.hostname}:{parts.port}"
+
+
+def safe_request_dump(payload: SolveRequest) -> str:
+	"""Запрос для лога без секретов клиента: креды прокси, ключ 2captcha, вводимые значения, cookies."""
+	data = payload.model_dump(exclude_none=True)
+	data["proxy"] = mask_proxy(payload.proxy)
+	if payload.captcha:
+		data["captcha"] = "***"
+	if payload.cookies:
+		data["cookies"] = [c.get("name") for c in payload.cookies]
+	for act in data["actions"]:
+		if act["type"] == "fill" and "value" in act:
+			act["value"] = "***"
+	return json.dumps(data, indent=2, ensure_ascii=False)
+
+
+def safe_response_dump(result: dict) -> str:
+	"""Ответ для лога: без значений cookies, html и скриншота."""
+	data = dict(result)
+	if data.get("cookies"):
+		data["cookies"] = [c.get("name") for c in data["cookies"]]
+	for key in ("html", "screenshot"):
+		if data.get(key):
+			data[key] = f"<{len(data[key])} chars>"
+	return json.dumps(data, indent=2, ensure_ascii=False)
+
+
+def session_dir(session_id: str | None) -> tuple[str, Path]:
+	"""Каталог профиля по session_id из cookie. Принимается только UUID — иначе (в т.ч. попытка
+	подсунуть путь вида ../..) выдаётся новая сессия."""
+	try:
+		session_id = str(uuid.UUID(session_id)) if session_id else None
+	except ValueError:
+		logger.warning("invalid session_id cookie, issuing a new session")
+		session_id = None
+	session_id = session_id or str(uuid.uuid4())
+
+	user_data_dir = (USER_DATA_ROOT / session_id).resolve()
+	if user_data_dir.parent != USER_DATA_ROOT:
+		raise ValueError("invalid session path")
+	return session_id, user_data_dir
+
+
+def load_session_settings(user_data_dir: Path) -> dict:
+	try:
+		return json.loads((user_data_dir / SESSION_SETTINGS).read_text())
+	except (OSError, json.JSONDecodeError):
+		return {}
+
+
+def save_session_settings(user_data_dir: Path, settings: dict) -> None:
+	(user_data_dir / SESSION_SETTINGS).write_text(json.dumps(settings, ensure_ascii=False))
+
+
 def make_page_action(payload: SolveRequest):
 	"""Замыкание для AsyncStealthySession.fetch(page_action=...): выполняет очередь действий
 	(включая прохождение капчи/антибота через api.actions.execute) и достаёт из уже
 	навигированной страницы всё, что нужно для ответа API. Возврат этой функции — то, что
-	вернёт fetch() (см. engine/stealthy.py)."""
+	вернёт fetch() (см. engine/stealthy.py).
+
+	Непройденный антибот не бросается наружу, а возвращается в challenge_error — чтобы в ответ
+	всё равно попали cookies, html и скриншот страницы-заглушки."""
 
 	async def page_action(page: Page) -> dict:
-		actions_result = await actions.execute(
-			page, payload.actions, payload.captcha,
-			antibot_name=payload.captcha_type,
-			success_locator=payload.captcha_success_locator,
-			detect_locator=payload.captcha_detect_locator,
-		)
+		challenge_error = None
+		try:
+			actions_result = await actions.execute(
+				page, payload.actions, payload.captcha,
+				antibot_name=payload.captcha_type,
+				success_locator=payload.captcha_success_locator,
+				detect_locator=payload.captcha_detect_locator,
+			)
+		except actions.ChallengeFailed as e:
+			challenge_error = str(e)
+			actions_result = []
+
+		screenshot = None
+		if payload.screenshot:
+			screenshot = base64.b64encode(await page.screenshot(type="png")).decode()
+
 		return {
+			"challenge_error": challenge_error,
 			"actions": actions_result,
 			"cookies": await page.context.cookies(),
 			"html": await page.inner_html("html"),
+			"screenshot": screenshot,
 			"user_agent": await page.evaluate("() => navigator.userAgent"),
 		}
 
@@ -61,100 +148,105 @@ app = FastAPI()
 
 
 @app.get("/")
-async def root():
-	return {"message": "Hello World"}
+@app.get("/health")
+async def health():
+	return {"status": "ok", "browser_slots": BROWSER_SLOTS, "free_slots": _slots._value}  # noqa
 
 
-@app.post("/v1/solve")
+@app.post("/v1/solve", response_model=SolveResponse, response_model_exclude_none=True)
 async def solve(
-	request: Request,
 	payload: SolveRequest,
 	response: Response,
 	session_id: Annotated[str | None, Cookie()] = None
 ):
-	# TODO: добавить поддержку api ключей
-	api_key = request.headers.get('X-API-Key')
+	# TODO: добавить поддержку api ключей (#4)
+	started = time.monotonic()
 
-	logger.info("Incoming payload: [%s]\n%s", session_id, payload.model_dump_json(indent=2, ensure_ascii=True))
-
-	global BROWSER_SLOTS
-
-	try:
-		assert BROWSER_SLOTS > 0 and (BROWSER_SLOTS := BROWSER_SLOTS.__sub__(1)) >= 0, "No slots available"
-
-		proxy = parse_proxy(payload.proxy)
-
-		if not session_id:
-			session_id = str(uuid.uuid4())
-			response.set_cookie("session_id", session_id)
-			logger.info("Setup new session with session_id: %s", session_id)
-
-		# профиль (cookies/local storage/fingerprint.json) хранится на диске по
-		# session_id — если он уже приходил в куках, ниже подгружаем существующий
-		# fingerprint из этой папки; если нет (новый session_id) — генерируем новый
-		# и сохраняем туда же (см. api/fingerprint.py:load_or_create_fingerprint)
-		user_data_dir = Path(__file__).parent / f"../extra/user_data_dir/{session_id}"
-		if not user_data_dir.exists():
-			user_data_dir.mkdir(parents=True)
-
-		# geo — до создания fingerprint'а, чтобы производная от прокси locale
-		# (см. api/geoip.py) попала в сам fingerprint, а не подменяла его
-		# задним числом
-		geo = await resolve_geo(proxy)
-		locale = geo.get("locale") if geo else None
-
-		fp_kwargs = {}
-		if payload.user_agent:
-			fp_kwargs["user_agent"] = payload.user_agent
-		# TODO: если есть session_id, то user_agent нужно взять из настроек пользователя
-		if locale:
-			fp_kwargs["locale"] = locale
-
-		nav_timeout_ms = payload.timeout * 1000 if payload.timeout else 60_000
-		blocked_domains = set(payload.blocked_domains) if payload.blocked_domains else None
-
-		kwargs = dict(
-			headless=False,
-			proxy=proxy,
-			user_data_dir=str(user_data_dir),
-			real_chrome=True,
-			locale=locale,
-			timezone_id=geo.get("timezone_id") if geo else None,
-			cookies=payload.cookies,
-			disable_resources=payload.disable_resources,
-			blocked_domains=blocked_domains,
-			timeout=nav_timeout_ms,
-			retries=max(1, payload.retries),
-			retry_delay=payload.retry_delay,
-			wait=payload.wait,
-			google_search=payload.google_search
+	if _slots.locked():
+		logger.warning("no free browser slots (%d), rejecting %s", BROWSER_SLOTS, payload.url)
+		return JSONResponse(
+			status_code=503,
+			headers={"Retry-After": RETRY_AFTER},
+			content={"status": "err", "code": "no_slots", "error": "No slots available"},
 		)
 
-		async with AsyncStealthySession(**kwargs) as engine:
-			result = await engine.fetch(payload.url, page_action=make_page_action(payload))
-	except Exception as e:
-		traceback.print_exc()
-		result_payload = {
-			"error": str(e) or type(e).__name__,
-			"status": False
-		}
-	else:
-		failed = [a for a in result["actions"] if a.get("ok") is False]
-		status = not failed
-		error = None if status else f"{failed[0]['type']}: {failed[0].get('result')}"
+	async with _slots:
+		session_id, user_data_dir = session_dir(session_id)
+		response.set_cookie("session_id", session_id)
+		logger.info("Incoming request [%s]: %s", session_id, payload.url)
+		logger.debug("Incoming payload [%s]:\n%s", session_id, safe_request_dump(payload))
 
-		result_payload = {
-			"status": ["err", "ok"][status],
-			"user_agent": result["user_agent"],
-			"cookies": result["cookies"],
-			"actions": result["actions"],
-		}
-		if payload.html:
-			result_payload["html"] = result["html"]
-		if error:
-			result_payload["error"] = error
-	finally:
-		BROWSER_SLOTS += 1
+		try:
+			user_data_dir.mkdir(parents=True, exist_ok=True)
 
-	logger.info("Outgoing payload:\n%s", json.dumps(result_payload, indent=2, sort_keys=True, ensure_ascii=True))
+			# user_agent сессии сохраняется рядом с профилем: при повторном запросе с тем же
+			# session_id без user_agent отпечаток не должен меняться
+			settings = load_session_settings(user_data_dir)
+			if payload.user_agent:
+				settings["user_agent"] = payload.user_agent
+				save_session_settings(user_data_dir, settings)
+			user_agent = settings.get("user_agent")
+
+			proxy = parse_proxy(payload.proxy) if payload.proxy else None
+
+			# geo — по exit-IP прокси (см. api/geoip.py); без прокси оставляем настройки браузера
+			geo = await resolve_geo(proxy) if proxy else None
+			locale = geo.get("locale") if geo else None
+
+			nav_timeout_ms = payload.timeout * 1000 if payload.timeout else 60_000
+			blocked_domains = set(payload.blocked_domains) if payload.blocked_domains else None
+
+			kwargs = dict(
+				headless=False,
+				proxy=proxy,
+				user_data_dir=str(user_data_dir),
+				real_chrome=True,
+				useragent=user_agent,
+				locale=locale,
+				timezone_id=geo.get("timezone_id") if geo else None,
+				cookies=payload.cookies,
+				disable_resources=payload.disable_resources,
+				blocked_domains=blocked_domains,
+				timeout=nav_timeout_ms,
+				retries=max(1, payload.retries),
+				retry_delay=payload.retry_delay,
+				wait=payload.wait,
+				google_search=payload.google_search
+			)
+
+			async with AsyncStealthySession(**kwargs) as engine:
+				result = await engine.fetch(payload.url, page_action=make_page_action(payload))
+		except Exception as e:
+			logger.exception("request failed [%s]", session_id)
+			result_payload = {
+				"status": "err",
+				"code": "request_failed",
+				"error": str(e) or type(e).__name__,
+			}
+		else:
+			failed = [a for a in result["actions"] if a.get("ok") is False]
+			result_payload = {
+				"status": "ok",
+				"user_agent": result["user_agent"],
+				"cookies": result["cookies"],
+				"actions": result["actions"],
+			}
+			if result["challenge_error"]:
+				result_payload.update(status="err", code="challenge_failed", error=result["challenge_error"])
+			elif failed:
+				result_payload.update(
+					status="err", code="action_failed",
+					error=f"{failed[0]['type']}: {failed[0].get('result')}",
+				)
+			if payload.html:
+				result_payload["html"] = result["html"]
+			if payload.screenshot:
+				result_payload["screenshot"] = result["screenshot"]
+
+	result_payload["duration_ms"] = int((time.monotonic() - started) * 1000)
+	logger.info(
+		"Outgoing [%s]: %s %s %dms", session_id, result_payload["status"],
+		result_payload.get("error", ""), result_payload["duration_ms"],
+	)
+	logger.debug("Outgoing payload:\n%s", safe_response_dump(result_payload))
 	return result_payload

tests/loadtest.py

@@ -35,15 +35,16 @@ def classify(resp: httpx.Response | Exception) -> str:
 	if isinstance(resp, Exception):
 		return f"exc:{type(resp).__name__}"
 	data = resp.json()
-	if "No slots" in str(data.get("error", "")):
+	code = data.get("code")
+	if code == "no_slots":
 		return "no_slots"
-	actions = data.get("actions") or []
-	if not actions:
-		return f"error:{str(data.get('error'))[:60]}"
-	last = actions[-1]
-	if not last.get("ok") or last.get("result"):
+	if code == "challenge_failed":
+		return "cf_blocked"
+	if code == "action_failed":
 		html = data.get("html") or ""
 		return "cf_blocked" if any(m in html for m in CF_MARKERS) else "action_failed"
+	if data.get("status") != "ok":
+		return f"error:{str(data.get('error'))[:60]}"
 	return "ok"