dubizzle project

dubizzle_scraper/worker/__init__.py

@@ -0,0 +1,3 @@
+from .celery_app import celery_app
+
+__all__ = ["celery_app"]

dubizzle_scraper/worker/celery_app.py

@@ -0,0 +1,148 @@
+# Инициализация Celery-приложения и периодических задач.
+
+import logging
+
+from celery import Celery
+from celery.signals import worker_process_init, worker_ready, setup_logging as celery_setup_logging
+from redis import Redis
+
+from ..core.config import settings
+from ..core.logs import setup_logging
+
+logger = logging.getLogger("dubizzle_scraper.worker.celery_app")
+
+
+@celery_setup_logging.connect
+def _configure_logging(loglevel=None, **kwargs):
+    # Перехватываем логирование Celery и пишем только в stderr (Docker logs).
+    level = settings.log_level if settings.log_level else "INFO"
+    setup_logging(level, None)
+
+
+@worker_process_init.connect
+def _on_worker_process_init(**kwargs):
+    # Повторно настраиваем логирование в каждом дочернем prefork-процессе,
+    # чтобы StreamHandler(stderr) корректно работал после fork.
+    level = settings.log_level if settings.log_level else "INFO"
+    setup_logging(level, None)
+
+
+def _broker_url() -> str:
+    return settings.celery.broker_url or settings.redis.url
+
+
+def _result_backend() -> str:
+    return settings.celery.result_backend or settings.redis.url
+
+
+celery_app = Celery(
+    "dubizzle_scraper",
+    broker=_broker_url(),
+    backend=_result_backend(),
+)
+
+# Auto-clamp: если hard limit слишком далёк от soft (> soft + 120),
+# ограничиваем, чтобы зависший worker не жил вечно.
+_soft = settings.celery.task_soft_time_limit
+_hard = settings.celery.task_time_limit
+_max_hard = _soft + 120 if _soft else _hard
+if _hard > _max_hard:
+    logger.warning(
+        "CELERY_TASK_TIME_LIMIT=%d too far from CELERY_TASK_SOFT_TIME_LIMIT=%d; "
+        "clamping hard limit to %d",
+        _hard, _soft, _max_hard,
+    )
+    _hard = _max_hard
+
+celery_app.conf.update(
+    task_serializer="json",
+    accept_content=["json"],
+    result_serializer="json",
+    timezone="UTC",
+    enable_utc=True,
+    task_soft_time_limit=_soft,
+    task_time_limit=_hard,
+    task_acks_late=True,
+    task_reject_on_worker_lost=True,
+    task_track_started=True,
+    worker_concurrency=settings.celery.worker_concurrency,
+    worker_max_tasks_per_child=settings.celery.worker_max_tasks_per_child,
+    worker_pool="solo",
+    worker_prefetch_multiplier=1,
+    broker_connection_retry_on_startup=True,
+    broker_transport_options={
+        "visibility_timeout": settings.celery.broker_visibility_timeout,
+    },
+    result_expires=86400,
+    worker_redirect_stdouts=False,
+    worker_hijack_root_logger=False,
+    beat_schedule={
+        "periodic-sync-listing": {
+            "task": "dubizzle_scraper.worker.tasks.sync_listing_task",
+            "schedule": settings.celery.beat_sync_interval_minutes * 60.0,
+            "args": (),
+            "kwargs": {
+                "limit": settings.celery.beat_sync_limit,
+                "only_new": False if settings.discovery.always_full_scan else True,
+            },
+            "options": {
+                "queue": "scraping",
+                "expires": settings.celery.beat_sync_interval_minutes * 60.0,
+            },
+        }
+    },
+    task_routes={
+        "dubizzle_scraper.worker.tasks.*": {"queue": "scraping"}
+    },
+)
+
+celery_app.autodiscover_tasks(["dubizzle_scraper.worker"])
+
+
+@worker_ready.connect
+def _on_worker_ready(**kwargs):
+    """При старте worker отправляем первый sync_listing, если очередь пуста."""
+    redis_client = None
+    try:
+        redis_client = Redis.from_url(
+            settings.redis.url,
+            decode_responses=True,
+            socket_connect_timeout=settings.redis.socket_connect_timeout_seconds,
+            socket_timeout=settings.redis.socket_timeout_seconds,
+            health_check_interval=settings.redis.health_check_interval_seconds,
+            retry_on_timeout=True,
+        )
+
+        for stale_key in ("dubizzle:locks:sync_listing",):
+            try:
+                ttl = redis_client.ttl(stale_key)
+                if ttl is not None and ttl != -2:
+                    redis_client.delete(stale_key)
+                    logger.warning("Cleared stale lock on startup: %s (ttl was %s)", stale_key, ttl)
+            except Exception:
+                logger.warning("Failed to clear stale lock %s on startup", stale_key, exc_info=True)
+
+        try:
+            queue_len = int(redis_client.llen("scraping") or 0)
+        except Exception:
+            queue_len = 0
+        if queue_len > 0:
+            logger.info("Worker ready: scraping queue already has %d task(s); skip startup dispatch", queue_len)
+            return
+    except Exception:
+        logger.warning("Worker ready startup sync check failed; skipping immediate dispatch", exc_info=True)
+        return
+    finally:
+        if redis_client is not None:
+            try:
+                redis_client.close()
+            except Exception:
+                pass
+
+    logger.info("Worker ready — dispatching initial sync_listing task")
+    celery_app.send_task(
+        "dubizzle_scraper.worker.tasks.sync_listing_task",
+        kwargs={"limit": settings.celery.beat_sync_limit, "only_new": False},
+        queue="scraping",
+        expires=settings.celery.beat_sync_interval_minutes * 60.0,
+    )

dubizzle_scraper/worker/self_heal.py

@@ -0,0 +1,209 @@
+import json
+import logging
+import os
+import random
+import signal
+import time
+
+from redis import Redis
+
+
+logger = logging.getLogger("dubizzle_scraper.worker.self_heal")
+
+GLOBAL_PROGRESS_TS_KEY = "dubizzle:state:last_progress_ts"
+SELF_HEAL_RESTART_LOCK_KEY = "dubizzle:state:self_heal_restart_in_progress"
+
+
+def _env_str(name: str, default: str) -> str:
+	value = os.getenv(name)
+	if value is None and name.startswith("DUBIZZLE_"):
+		value = os.getenv("DUBIZZLE_" + name[len("DUBIZZLE_"):])
+	return value if value is not None else default
+
+
+def _env_bool(name: str, default: bool) -> bool:
+	value = _env_str(name, "true" if default else "false")
+	if value is None:
+		return default
+	return value.strip().lower() in {"1", "true", "yes", "on"}
+
+
+def _env_int(name: str, default: int) -> int:
+	value = _env_str(name, str(default))
+	if value is None:
+		return default
+	try:
+		return int(value.strip())
+	except Exception:
+		return default
+
+
+def _get_redis() -> Redis:
+	url = _env_str("DUBIZZLE_REDIS_URL", "redis://redis:6379/0")
+	return Redis.from_url(
+		url,
+		decode_responses=True,
+		socket_connect_timeout=5.0,
+		socket_timeout=10.0,
+		health_check_interval=30,
+		retry_on_timeout=True,
+	)
+
+
+def _safe_int(value: str | None, default: int = 0) -> int:
+	if value is None:
+		return default
+	try:
+		return int(str(value).strip())
+	except Exception:
+		return default
+
+
+def _read_last_progress_ts(redis_client: Redis) -> int | None:
+	raw = redis_client.get(GLOBAL_PROGRESS_TS_KEY)
+	if raw:
+		ts = _safe_int(raw)
+		if ts > 0:
+			return ts
+
+	# Fallback: если глобальный ключ не найден, берём max(ts) из task_progress:*.
+	# Это дороже, но выполняется только при отсутствии основного маркера.
+	max_ts = 0
+	for key in redis_client.scan_iter(match="dubizzle:state:task_progress:*"):
+		try:
+			payload = redis_client.get(key)
+			if not payload:
+				continue
+			data = json.loads(payload)
+			ts = _safe_int(data.get("ts"), 0)
+			if ts > max_ts:
+				max_ts = ts
+		except Exception:
+			continue
+	return max_ts or None
+
+
+def _kill_worker_process() -> None:
+	pid_file = "/tmp/celery-worker.pid"
+	pid: int | None = None
+	try:
+		with open(pid_file, "r", encoding="utf-8") as f:
+			pid = int(f.read().strip())
+	except Exception:
+		pid = None
+
+	if not pid:
+		logger.error("Self-heal: failed to read worker pid from %s", pid_file)
+		return
+
+	logger.error("Self-heal: terminating stuck worker process pid=%s", pid)
+	try:
+		os.kill(pid, signal.SIGTERM)
+	except Exception:
+		logger.exception("Self-heal: failed to send SIGTERM to pid=%s", pid)
+		return
+
+	time.sleep(20)
+	try:
+		# Если процесс ещё жив — принудительно убиваем.
+		os.kill(pid, 0)
+		logger.error("Self-heal: worker pid=%s did not stop after SIGTERM; sending SIGKILL", pid)
+		kill_signal = getattr(signal, "SIGKILL", signal.SIGTERM)
+		os.kill(pid, kill_signal)
+	except ProcessLookupError:
+		pass
+	except Exception:
+		logger.exception("Self-heal: failed to send SIGKILL to pid=%s", pid)
+
+
+def main() -> None:
+	if not _env_bool("DUBIZZLE_SELF_HEAL_ENABLED", True):
+		logger.info("Self-heal watchdog disabled via DUBIZZLE_SELF_HEAL_ENABLED/DUBIZZLE_SELF_HEAL_ENABLED")
+		return
+
+	queue_name = _env_str("DUBIZZLE_CELERY_QUEUE", "scraping")
+	check_interval = max(5, _env_int("DUBIZZLE_SELF_HEAL_CHECK_INTERVAL_SECONDS", 30))
+	stall_seconds = max(180, _env_int("DUBIZZLE_SELF_HEAL_STALL_SECONDS", 720))
+	startup_grace = max(30, _env_int("DUBIZZLE_SELF_HEAL_STARTUP_GRACE_SECONDS", 300))
+	restart_cooldown = max(60, _env_int("DUBIZZLE_SELF_HEAL_RESTART_COOLDOWN_SECONDS", 300))
+
+	logger.warning(
+		"Self-heal watchdog enabled: queue=%s check_interval=%ss stall=%ss startup_grace=%ss cooldown=%ss",
+		queue_name,
+		check_interval,
+		stall_seconds,
+		startup_grace,
+		restart_cooldown,
+	)
+
+	started_at = time.time()
+	redis_client: Redis | None = None
+
+	while True:
+		try:
+			if redis_client is None:
+				redis_client = _get_redis()
+				redis_client.ping()
+
+			queue_len = _safe_int(redis_client.llen(queue_name), 0)
+			if queue_len <= 0:
+				time.sleep(check_interval)
+				continue
+
+			last_progress_ts = _read_last_progress_ts(redis_client)
+			now_ts = int(time.time())
+			age = None if last_progress_ts is None else max(0, now_ts - int(last_progress_ts))
+
+			if age is None:
+				if now_ts - int(started_at) < startup_grace:
+					time.sleep(check_interval)
+					continue
+				logger.warning(
+					"Self-heal: queue=%d but no progress timestamp found after startup grace",
+					queue_len,
+				)
+				age = stall_seconds + 1
+
+			if age <= stall_seconds:
+				time.sleep(check_interval)
+				continue
+
+			# Глобальный anti-storm lock: чтобы много воркеров не рестартились одновременно.
+			acquired = bool(
+				redis_client.set(
+					SELF_HEAL_RESTART_LOCK_KEY,
+					str(now_ts),
+					nx=True,
+					ex=restart_cooldown,
+				)
+			)
+			if not acquired:
+				time.sleep(check_interval)
+				continue
+
+			logger.error(
+				"Self-heal: detected global stall (queue=%d, progress_age=%ss > %ss). Restarting worker process...",
+				queue_len,
+				age,
+				stall_seconds,
+			)
+			# Небольшой джиттер, чтобы при одинаковом событии у разных контейнеров
+			# перезапуск был не строго одновременно.
+			time.sleep(random.uniform(0.3, 2.0))
+			_kill_worker_process()
+			# После kill pid1 контейнер будет перезапущен Docker restart-policy.
+			# На случай неуспеха не молотим цикл.
+			time.sleep(check_interval)
+
+		except Exception:
+			logger.exception("Self-heal watchdog iteration failed")
+			redis_client = None
+			time.sleep(check_interval)
+
+
+if __name__ == "__main__":
+	logging.basicConfig(
+		level=_env_str("DUBIZZLE_LOG_LEVEL", "INFO"),
+		format="%(asctime)s | %(levelname)s | %(name)s | %(message)s",
+	)
+	main()