scrapely-python 0.1.5__tar.gz → 0.1.7__tar.gz

{scrapely_python-0.1.5 → scrapely_python-0.1.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: scrapely-python
-Version: 0.1.5
+Version: 0.1.7
 Summary: Scrapely Python SDK - A toolkit for building Actors on the Scrapely platform
 Author-email: Scrapely <support@scrape.ly>
 License-Expression: MIT

{scrapely_python-0.1.5 → scrapely_python-0.1.7}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "scrapely-python"
-version = "0.1.5"
+version = "0.1.7"
 description = "Scrapely Python SDK - A toolkit for building Actors on the Scrapely platform"
 readme = "README.md"
 license = "MIT"

{scrapely_python-0.1.5 → scrapely_python-0.1.7}/src/scrapely/__init__.py

@@ -132,7 +132,7 @@ from .errors import (
     raise_for_status,
 )
 
-__version__ = '0.1.5'
+__version__ = '0.1.7'
 __all__ = [
     # Core
     'Actor',

{scrapely_python-0.1.5 → scrapely_python-0.1.7}/src/scrapely/actor.py

@@ -18,7 +18,7 @@ import warnings
 from collections.abc import MutableMapping
 from datetime import datetime, timedelta
 from decimal import Decimal
-from typing import TYPE_CHECKING, Any, Callable, Dict, List, Optional, Union, cast
+from typing import TYPE_CHECKING, Any, Callable, Dict, List, Optional, Tuple, Union, cast
 
 from crawlee.configuration import Configuration as CrawleeConfiguration
 from crawlee.storage_clients._base import StorageClient
@@ -27,6 +27,7 @@ from crawlee.storages import Dataset, KeyValueStore, RequestQueue
 from .charging import ChargingManager, is_default_dataset_item_event, push_data_and_charge
 from .configuration import Configuration
 from .crypto import decrypt_secret_inputs
+from .proxy_configuration import with_proxy_key_aliases
 from crawlee.events import EventManager
 
 from .events import Event, LocalEventManager, ScrapelyEventManager
@@ -287,6 +288,7 @@ class Actor(metaclass=_ActorMeta):
             cls._exit_code = self._instance_exit_code
         if self._instance_status_message:
             cls._status_message = self._instance_status_message
+            await cls._persist_terminal_status_message(self._instance_status_message)
         await cls._shutdown(timeout_secs=int(self._cleanup_timeout.total_seconds()))
 
     @classmethod
@@ -483,8 +485,10 @@ class Actor(metaclass=_ActorMeta):
                     flush = getattr(cls._charging_manager, 'flush_pending', None)
                     if flush is not None:
                         deadline_ms = min(max(1, timeout_secs) * 1000, 30_000)
+                        # The flush's own deadline ends its retries (charge POSTs back off
+                        # for up to ~45 s) a little before wait_for would cancel it.
                         await asyncio.wait_for(
-                            flush(max_intents=64, deadline_ms=deadline_ms),
+                            flush(max_intents=64, deadline_ms=max(500, deadline_ms - 2_000)),
                             timeout=max(1.0, deadline_ms / 1000.0),
                         )
                 except BaseException as e:
@@ -589,6 +593,7 @@ class Actor(metaclass=_ActorMeta):
         cls._exit_code = opts.exit_code
         cls._status_message = opts.status_message
 
+        await cls._persist_terminal_status_message(opts.status_message)
         await cls._shutdown(timeout_secs=opts.timeout_secs)
 
         if opts.exit:
@@ -685,8 +690,8 @@ class Actor(metaclass=_ActorMeta):
                 kvs = await cls.open_key_value_store()
                 metamorph_input = await kvs.get_value(f'INPUT-METAMORPH-{count}')
                 if metamorph_input is not None:
-                    cls._cached_input = metamorph_input
-                    return cast(Dict[str, Any], metamorph_input)
+                    cls._cached_input = with_proxy_key_aliases(metamorph_input)
+                    return cast(Dict[str, Any], cls._cached_input)
 
         raw_input = os.environ.get('ACTOR_INPUT')
         if raw_input:
@@ -699,7 +704,8 @@ class Actor(metaclass=_ActorMeta):
             # path is decrypted, mirroring Actor.getInput() in the Node SDK)
             if isinstance(parsed, dict):
                 parsed = decrypt_secret_inputs(parsed)
-            cls._cached_input = parsed
+            # Proxy objects carry both namings (scrapelyProxy* and apifyProxy*).
+            cls._cached_input = with_proxy_key_aliases(parsed)
             return cast(Dict[str, Any], cls._cached_input)
 
         kvs = await cls.open_key_value_store()
@@ -707,8 +713,8 @@ class Actor(metaclass=_ActorMeta):
         result = await kvs.get_value(input_key)
         if isinstance(result, dict):
             result = decrypt_secret_inputs(result)
-        cls._cached_input = result
-        return cast(Optional[Dict[str, Any]], result)
+        cls._cached_input = with_proxy_key_aliases(result)
+        return cast(Optional[Dict[str, Any]], cls._cached_input)
 
     @classmethod
     async def get_input_or_throw(cls) -> Dict[str, Any]:
@@ -737,7 +743,7 @@ class Actor(metaclass=_ActorMeta):
 
         # Decrypt secret inputs after loading (JS parity with Actor.input getter)
         if isinstance(cls._cached_input, dict):
-            cls._cached_input = decrypt_secret_inputs(cls._cached_input)
+            cls._cached_input = with_proxy_key_aliases(decrypt_secret_inputs(cls._cached_input))
 
         return cls._cached_input
 
@@ -1021,13 +1027,32 @@ class Actor(metaclass=_ActorMeta):
         else:
             logger.info(logged)
 
+        result = await cls._put_run_status_message(message, is_terminal=is_terminal)
+        if result is None:
+            return ActorRun(id='', actor_id='', status='')
+        run_id, data = result
+
+        return ActorRun(
+            id=data.get('id', run_id),
+            actor_id=data.get('actorId', ''),
+            status=data.get('status', ''),
+            started_at=data.get('startedAt'),
+            finished_at=data.get('finishedAt'),
+            status_message=data.get('statusMessage'),
+        )
+
+    @classmethod
+    async def _put_run_status_message(
+        cls, message: str, *, is_terminal: bool
+    ) -> Optional[Tuple[str, Dict[str, Any]]]:
+        """PUT the status message to the current run. Returns ``None`` when not on the platform."""
         config = cls.get_config()
         api_url = config.get('api_base_url')
         token = config.get('token')
         run_id = config.get('actor_run_id')
 
         if not api_url or not token or not run_id:
-            return ActorRun(id='', actor_id='', status='')
+            return None
 
         client = cls.new_client(token=token, base_url=api_url)
         response = await client.put(
@@ -1041,15 +1066,27 @@ class Actor(metaclass=_ActorMeta):
         if response.content:
             body = response.json()
         data = body.get('data', body) if isinstance(body, dict) else {}
+        return run_id, data if isinstance(data, dict) else {}
 
-        return ActorRun(
-            id=data.get('id', run_id),
-            actor_id=data.get('actorId', ''),
-            status=data.get('status', ''),
-            started_at=data.get('startedAt'),
-            finished_at=data.get('finishedAt'),
-            status_message=data.get('statusMessage'),
-        )
+    @classmethod
+    async def _persist_terminal_status_message(cls, message: Optional[str], timeout_secs: float = 10.0) -> None:
+        """Store the exit/fail status message on the run as terminal (Apify parity).
+
+        Runs before teardown so it lands while the run still exists and the
+        process is alive, on success and failure alike. Best effort: a failed
+        update is logged and never blocks the exit.
+        """
+        if not message or not cls._initialized or cls._config is None:
+            return
+        try:
+            await asyncio.wait_for(
+                cls._put_run_status_message(message, is_terminal=True),
+                timeout=max(1.0, timeout_secs),
+            )
+        except asyncio.CancelledError:
+            raise
+        except BaseException as e:
+            logger.warning('Failed to store the final status message on the run: %s', e)
 
     @_ClassProperty
     def log(cls) -> logging.Logger:
@@ -1139,6 +1176,11 @@ class Actor(metaclass=_ActorMeta):
         """Return the Configuration instance."""
         return cls.get_config()
 
+    @_ClassProperty
+    def config(cls) -> Configuration:
+        """Return the Configuration instance (Apify SDK ``Actor.config`` parity)."""
+        return cls.get_config()
+
     @_ClassProperty
     def event_manager(cls) -> Optional[EventManager]:
         """Return the EventManager instance."""

scrapely_python-0.1.7/src/scrapely/cdp_retry.py

@@ -0,0 +1,226 @@
+"""Retry remote-browser (CDP) connects that fail for transient reasons.
+
+The remote browser answers ``429`` when it is at capacity and ``503`` when it
+cannot start a browser or validate the session right now, both with a
+``Retry-After`` delay (also repeated in the response body, because Playwright
+only puts the status line and body into its error message). Connection errors
+and timeouts are transient too. Those are retried with exponential backoff,
+waiting at least the ``Retry-After`` delay when the error carries one. Any
+other status (``400``/``401``/``402``/``403``/``404``) fails at once.
+
+Tuning (environment):
+
+- ``SCRAPELY_CDP_CONNECT_ATTEMPTS``: total attempts, default ``5`` (``1`` disables retries).
+- ``SCRAPELY_CDP_RETRY_BASE_SECS``: first backoff delay, default ``1``; doubles per attempt.
+- ``SCRAPELY_CDP_RETRY_MAX_SECS``: cap for one backoff delay, default ``30``.
+- ``SCRAPELY_CDP_RETRY_AFTER_MAX_SECS``: longest ``Retry-After`` the SDK waits
+  out, default ``120``. A longer ``Retry-After`` ends the retries (the SDK never
+  retries earlier than the server asked).
+"""
+
+from __future__ import annotations
+
+import asyncio
+import functools
+import logging
+import os
+import random
+import re
+import time
+from typing import Any, Awaitable, Callable, Optional, Tuple, TypeVar
+from urllib.parse import urlsplit
+
+logger = logging.getLogger('scrapely.browser')
+
+T = TypeVar('T')
+
+_RETRY_WRAPPED_ATTR = '_scrapely_cdp_retry_wrapped'
+
+_REASONS = (
+    'Too Many Requests|Internal Server Error|Bad Gateway|Service Unavailable|Gateway Timeout|'
+    'Request Timeout|Bad Request|Unauthorized|Payment Required|Forbidden|Not Found'
+)
+_STATUS_RE = re.compile(rf'\b([1-5]\d\d) (?:{_REASONS})\b')
+_WS_STATUS_RE = re.compile(r'Unexpected server response:?\s*(\d{3})', re.IGNORECASE)
+_RETRY_AFTER_RE = re.compile(r'Retry-After:?\s*(\d+(?:\.\d+)?)', re.IGNORECASE)
+_TRANSIENT_MARKERS = (
+    'econnrefused',
+    'econnreset',
+    'etimedout',
+    'epipe',
+    'eai_again',
+    'ehostunreach',
+    'enetunreach',
+    'socket hang up',
+    'timeout',
+    'timed out',
+    'connection closed',
+    'connection refused',
+    'connection reset',
+    'websocket error',
+    'net::err_',
+    'target closed',
+)
+
+
+def _env_float(name: str, default: float) -> float:
+    try:
+        value = float(os.environ.get(name, ''))
+    except ValueError:
+        return default
+    return value if value > 0 else default
+
+
+def _max_attempts() -> int:
+    return max(1, int(_env_float('SCRAPELY_CDP_CONNECT_ATTEMPTS', 5)))
+
+
+def parse_status(message: str) -> Optional[int]:
+    """HTTP status from a Playwright/Puppeteer connect error message, if any."""
+    match = _WS_STATUS_RE.search(message) or _STATUS_RE.search(message)
+    return int(match.group(1)) if match else None
+
+
+def parse_retry_after(message: str) -> Optional[float]:
+    match = _RETRY_AFTER_RE.search(message)
+    if not match:
+        return None
+    return float(match.group(1))
+
+
+def classify_connect_error(err: BaseException) -> Tuple[bool, Optional[int], Optional[float]]:
+    """Return ``(retryable, status, retry_after_secs)`` for a connect error."""
+    message = f'{type(err).__name__}: {err}'
+    status = parse_status(message)
+    retry_after = parse_retry_after(message)
+    if status is not None:
+        return (status == 408 or status == 429 or status >= 500), status, retry_after
+    if isinstance(err, (asyncio.TimeoutError, TimeoutError, ConnectionError, OSError)):
+        return True, None, retry_after
+    lowered = message.lower()
+    return any(marker in lowered for marker in _TRANSIENT_MARKERS), None, retry_after
+
+
+def backoff_delay(attempt: int, retry_after: Optional[float]) -> float:
+    """Delay before retry number ``attempt`` (1-based), honoring Retry-After."""
+    base = _env_float('SCRAPELY_CDP_RETRY_BASE_SECS', 1.0)
+    cap = _env_float('SCRAPELY_CDP_RETRY_MAX_SECS', 30.0)
+    delay = min(cap, base * (2 ** (attempt - 1)))
+    delay = delay * (0.75 + random.random() * 0.5)
+    if retry_after is not None:
+        delay = max(delay, retry_after)
+    return delay
+
+
+def _retry_after_too_long(retry_after: Optional[float]) -> bool:
+    return retry_after is not None and retry_after > _env_float('SCRAPELY_CDP_RETRY_AFTER_MAX_SECS', 120.0)
+
+
+def _describe(err: BaseException, status: Optional[int]) -> str:
+    if status is not None:
+        return f'HTTP {status}'
+    # First line only: connect errors can carry the session URL with ids.
+    text = str(err).strip().splitlines()[0] if str(err).strip() else type(err).__name__
+    return re.sub(r'(wss?|https?)://\S+', '<cdp-url>', text)[:200]
+
+
+def _log_retry(err: BaseException, status: Optional[int], attempt: int, attempts: int, delay: float) -> None:
+    logger.warning(
+        'Remote browser connect failed (%s); retrying in %.1fs (attempt %d/%d)',
+        _describe(err, status),
+        delay,
+        attempt + 1,
+        attempts,
+    )
+
+
+async def connect_with_retry_async(connect: Callable[[], Awaitable[T]]) -> T:
+    """Run an async CDP connect, retrying transient failures."""
+    attempts = _max_attempts()
+    attempt = 1
+    while True:
+        try:
+            return await connect()
+        except asyncio.CancelledError:
+            raise
+        except Exception as err:
+            retryable, status, retry_after = classify_connect_error(err)
+            # Never retry earlier than the server asked; give up instead of waiting that long.
+            if not retryable or attempt >= attempts or _retry_after_too_long(retry_after):
+                raise
+            delay = backoff_delay(attempt, retry_after)
+            _log_retry(err, status, attempt, attempts, delay)
+            await asyncio.sleep(delay)
+            attempt += 1
+
+
+def connect_with_retry_sync(connect: Callable[[], T]) -> T:
+    """Run a blocking CDP connect, retrying transient failures."""
+    attempts = _max_attempts()
+    attempt = 1
+    while True:
+        try:
+            return connect()
+        except Exception as err:
+            retryable, status, retry_after = classify_connect_error(err)
+            # Never retry earlier than the server asked; give up instead of waiting that long.
+            if not retryable or attempt >= attempts or _retry_after_too_long(retry_after):
+                raise
+            delay = backoff_delay(attempt, retry_after)
+            _log_retry(err, status, attempt, attempts, delay)
+            time.sleep(delay)
+            attempt += 1
+
+
+def is_platform_cdp_url(url: Any, platform_cdp_url: Optional[str]) -> bool:
+    """True when ``url`` points at the same host as the platform's CDP_URL."""
+    if not isinstance(url, str) or not platform_cdp_url:
+        return False
+    try:
+        target, platform = urlsplit(url), urlsplit(platform_cdp_url)
+    except ValueError:
+        return False
+    return bool(target.netloc) and target.netloc.lower() == platform.netloc.lower()
+
+
+def wrap_connect_over_cdp(chromium: Any, is_async: bool, get_platform_cdp_url: Callable[[], Optional[str]]) -> None:
+    """Make ``chromium.connect_over_cdp`` retry transient failures for the platform CDP host.
+
+    Covers actor code that connects to ``CDP_URL`` directly as well as the
+    SDK's own redirected launches. Connects to other hosts are left alone.
+    Idempotent.
+    """
+    original = getattr(chromium, 'connect_over_cdp', None)
+    if original is None or getattr(original, _RETRY_WRAPPED_ATTR, False):
+        return
+
+    def _target(args: Tuple[Any, ...], kwargs: dict) -> Any:
+        if args:
+            return args[0]
+        return kwargs.get('endpoint_url') or kwargs.get('endpointURL')
+
+    if is_async:
+
+        @functools.wraps(original)
+        async def connect_over_cdp(*args: Any, **kwargs: Any) -> Any:
+            if not is_platform_cdp_url(_target(args, kwargs), get_platform_cdp_url()):
+                return await original(*args, **kwargs)
+            return await connect_with_retry_async(lambda: original(*args, **kwargs))
+
+    else:
+
+        @functools.wraps(original)
+        def connect_over_cdp(*args: Any, **kwargs: Any) -> Any:  # type: ignore[misc]
+            if not is_platform_cdp_url(_target(args, kwargs), get_platform_cdp_url()):
+                return original(*args, **kwargs)
+            return connect_with_retry_sync(lambda: original(*args, **kwargs))
+
+    setattr(connect_over_cdp, _RETRY_WRAPPED_ATTR, True)
+    try:
+        chromium.connect_over_cdp = connect_over_cdp
+    except (AttributeError, TypeError):
+        pass
+
+
+def is_retry_wrapped(fn: Any) -> bool:
+    return bool(getattr(fn, _RETRY_WRAPPED_ATTR, False))

{scrapely_python-0.1.5 → scrapely_python-0.1.7}/src/scrapely/charging.py

@@ -47,8 +47,16 @@ CHARGE_IDEMPOTENCY_SAFE_AGE_SECS = 30 * 24 * 60 * 60
 # Soft cap on open unresolved intents. Refuse NEW billable charges when full.
 _PENDING_MAX_ENTRIES = 1024
 
-# Immediate same-key POST attempts before leaving intent pending / raising.
-_POST_ATTEMPTS = 3
+# Same-key POST attempts before leaving an intent pending / raising. The API
+# client itself retries 5xx answers and refused or reset connections of these
+# POSTs with its long backoff (0.5 s doubling to 8 s, about 45 s): re-sending
+# is safe because every attempt carries the same Idempotency-Key. These outer
+# attempts cover what the client gives up on quickly (timeouts, other
+# transport errors), with the same backoff, within _RETRY_BUDGET_SECS.
+_POST_ATTEMPTS = 8
+
+# Outer retries stop once this much time has passed since the first attempt.
+_RETRY_BUDGET_SECS = 60.0
 
 # Local-only dataset collecting one entry per logical charge operation
 # (JS ChargingManager LOCAL_CHARGING_LOG_DATASET_NAME parity).
@@ -160,6 +168,7 @@ class ChargingManager:
     _PENDING_MAX_ENTRIES = _PENDING_MAX_ENTRIES
     _IDEMPOTENCY_SAFE_AGE_SECS = CHARGE_IDEMPOTENCY_SAFE_AGE_SECS
     _POST_ATTEMPTS = _POST_ATTEMPTS
+    _RETRY_BUDGET_SECS = _RETRY_BUDGET_SECS
 
     def __init__(self, config: 'Configuration', client: 'ApiClient'):
         """Initialize the charging manager.
@@ -458,12 +467,14 @@ class ChargingManager:
         """
         if self._client is None or not self._is_at_home or not self._actor_run_id:
             return False
+        started = time.monotonic()
         for attempt in range(self._POST_ATTEMPTS):
             try:
                 await self._client.post(
                     f'/v2/actor-runs/{self._actor_run_id}/charge/void',
                     json_data={},
                     headers={'Idempotency-Key': operation_id},
+                    idempotent=True,
                 )
                 self._unwind_reservation(event_name, reserved)
                 return True
@@ -478,8 +489,8 @@ class ChargingManager:
                 # 404: nothing to void; 409: already settled; 4xx auth: permanent.
                 if status is not None and 400 <= status < 500:
                     return False
-                if attempt + 1 < self._POST_ATTEMPTS:
-                    await asyncio.sleep(0.25 * (attempt + 1))
+                if not await self._backoff_before_retry(attempt, started, err):
+                    break
         # Outcome unknown: keep the reservation and retry before the next push / at exit.
         self._pending_voids.append((event_name, operation_id, reserved, not_found_means_voided))
         logger.error('Void of charge %s (%r) not confirmed yet; will retry', operation_id, event_name)
@@ -528,6 +539,7 @@ class ChargingManager:
         event_name: str,
         count: int,
         idempotency_key: str,
+        deadline: Optional[float] = None,
     ) -> Dict[str, Any]:
         if self._client is None:
             raise RuntimeError('ChargingManager has no API client; cannot post charge intent')
@@ -538,6 +550,8 @@ class ChargingManager:
                 'count': count,
             },
             headers={'Idempotency-Key': idempotency_key},
+            idempotent=True,
+            **({'retry_deadline': deadline} if deadline is not None else {}),
         )
         if isinstance(response, dict):
             return response
@@ -550,25 +564,58 @@ class ChargingManager:
             pass
         return {}
 
+    async def _backoff_before_retry(
+        self, attempt: int, started: float, err: BaseException, deadline: Optional[float] = None
+    ) -> bool:
+        """Sleep before the next same-key attempt; False when no attempt is left.
+
+        A 5xx answer reaching this point was already retried by the API client
+        with its long backoff, so it ends the attempts. Timeouts and other
+        errors without an answer back off the same way (0.5 s doubling to 8 s)
+        while attempts and the retry budget last.
+        """
+        from .storage.api_client import server_retry_delay
+
+        if attempt + 1 >= self._POST_ATTEMPTS:
+            return False
+        status = _http_status_of(err)
+        if status is not None and status >= 500:
+            return False
+        delay = server_retry_delay(attempt, None)
+        if time.monotonic() - started + delay > self._RETRY_BUDGET_SECS:
+            return False
+        if deadline is not None and time.monotonic() + delay > deadline:
+            return False
+        logger.warning('Charge request failed (%s); retrying in %.1fs', type(err).__name__, delay)
+        await asyncio.sleep(delay)
+        return True
+
     async def _post_charge_with_retries(
         self,
         *,
         event_name: str,
         count: int,
         idempotency_key: str,
+        deadline: Optional[float] = None,
     ) -> Optional[int]:
         """POST charge; return acceptedCount on success, None if uncertain.
 
+        ``deadline`` (``time.monotonic()``) bounds every retry, the API
+        client's included, so a flush with a time budget ends with the intent
+        still pending instead of being cancelled mid-request.
+
         Raises ChargeRejectedError for definitive platform rejections.
         Same idempotency key on every attempt.
         """
         last_error: Optional[BaseException] = None
+        started = time.monotonic()
         for attempt in range(self._POST_ATTEMPTS):
             try:
                 body = await self._post_charge_once(
                     event_name=event_name,
                     count=count,
                     idempotency_key=idempotency_key,
+                    deadline=deadline,
                 )
                 return _read_accepted_count(body, count)
             except Exception as e:
@@ -581,8 +628,8 @@ class ChargingManager:
                 )
                 if rejected is not None:
                     raise rejected from e
-                if attempt + 1 < self._POST_ATTEMPTS:
-                    await asyncio.sleep(0.25 * (attempt + 1))
+                if not await self._backoff_before_retry(attempt, started, e, deadline):
+                    break
         if last_error is not None:
             logger.warning(
                 'Failed to record charge for event %s (count=%s, key=%s): %s',
@@ -638,6 +685,7 @@ class ChargingManager:
                             event_name=event_name,
                             count=count,
                             idempotency_key=idempotency_key,
+                            deadline=deadline,
                         )
                     except ChargeRejectedError as err:
                         self._unwind_reservation(event_name, count)

{scrapely_python-0.1.5 → scrapely_python-0.1.7}/src/scrapely/configuration.py

@@ -145,6 +145,8 @@ class Configuration:
         'proxy_hostname': 'proxy.scrape.ly',
         'proxy_port': 8000,
         'container_port': 4321,
+        # Apify parity: standby runs listen on ACTOR_STANDBY_PORT (4321 unless set).
+        'standby_port': 4321,
         'container_url': 'http://localhost:4321',
         'metamorph_after_sleep_millis': 300000,
         'persist_state_interval_millis': 60000,
@@ -245,6 +247,25 @@ class Configuration:
         """
         self._config[key] = value
     
+    def __getattr__(self, name: str) -> Any:
+        """Attribute access to any known setting (Apify ``Actor.config.<field>`` parity).
+
+        ``Actor.config.standby_port``, ``Actor.config.meta_origin`` and the other
+        keys this class reads from the environment resolve like ``get(name)``.
+        """
+        if name.startswith('_'):
+            raise AttributeError(name)
+        known = name in self.DEFAULTS or name in self.ENV_MAP.values() or name in self.__dict__.get('_config', {})
+        if not known:
+            raise AttributeError(f'{type(self).__name__!r} object has no attribute {name!r}')
+        return self.get(name)
+
+    @property
+    def standby_port(self) -> int:
+        """Port a standby run serves HTTP on (ACTOR_STANDBY_PORT, default 4321)."""
+        value = self.get('standby_port')
+        return value if isinstance(value, int) else 4321
+
     @property
     def token(self) -> Optional[str]:
         """Get the authentication token."""

{scrapely_python-0.1.5 → scrapely_python-0.1.7}/src/scrapely/patch_browser_pool.py

@@ -70,6 +70,7 @@ from typing import Any, List, Optional, Tuple
 from urllib.parse import SplitResult, parse_qsl, unquote, urlencode, urlsplit, urlunsplit
 
 from ._browser_locations import subdivision_to_state_name
+from .cdp_retry import connect_with_retry_async, wrap_connect_over_cdp
 
 logger = logging.getLogger(__name__)
 
@@ -427,7 +428,9 @@ def _apply_patch() -> None:
         cdp_url = apply_proxy_info_to_cdp_url(cdp_url, proxy_info, browser_configuration)
 
         # Remote browser infrastructure is Chromium CDP (same as Node connectOverCDP).
-        browser = await self._playwright.chromium.connect_over_cdp(cdp_url)
+        chromium = self._playwright.chromium
+        wrap_connect_over_cdp(chromium, True, get_cdp_url)
+        browser = await chromium.connect_over_cdp(cdp_url)
 
         controller = PlaywrightBrowserController(
             browser,
@@ -518,7 +521,9 @@ def _apply_camoufox_patch() -> None:
         # Same as the PlaywrightBrowserPlugin patch above: the first page's
         # proxy info / browser configuration pins the connection's location.
         cdp_url = apply_proxy_info_to_cdp_url(cdp_url, _launch_proxy_info.get(), browser_configuration)
-        return await playwright.chromium.connect_over_cdp(cdp_url)
+        chromium = playwright.chromium
+        wrap_connect_over_cdp(chromium, True, get_cdp_url)
+        return await chromium.connect_over_cdp(cdp_url)
 
     camoufox.AsyncNewBrowser = _patched_async_new_browser
     camoufox._scrapely_cdp_patched = True
@@ -726,6 +731,10 @@ def _patch_playwright_object(playwright: Any, is_async: bool) -> Any:
     chromium = getattr(playwright, 'chromium', None)
     if chromium is None or getattr(chromium, 'connect_over_cdp', None) is None:
         return playwright
+    # Retry transient connect failures (429/5xx/network) for the platform CDP
+    # host: covers direct chromium.connect_over_cdp(CDP_URL) calls in actor
+    # code and the redirected launches below, which call this same attribute.
+    wrap_connect_over_cdp(chromium, is_async, get_cdp_url)
     for name in ('chromium', 'firefox', 'webkit'):
         browser_type = getattr(playwright, name, None)
         if browser_type is None:
@@ -845,7 +854,9 @@ def _apply_pyppeteer_patch() -> None:
                 'ignored: %s (the remote browser manages headless mode, binary and flags).',
                 ', '.join(sorted(merged.keys())),
             )
-        return await pyppeteer_launcher.connect(browserWSEndpoint=cdp_url)
+        return await connect_with_retry_async(
+            lambda: pyppeteer_launcher.connect(browserWSEndpoint=cdp_url)
+        )
 
     pyppeteer_launcher.launch = _patched_launch
     if hasattr(pyppeteer_mod, 'launch'):