diff --git a/app/processing_watchdog.py b/app/processing_watchdog.py
new file mode 100644
index 0000000..945c657
--- /dev/null
+++ b/app/processing_watchdog.py
@@ -0,0 +1,106 @@
+"""
+    Watchdog для мониторинга и прерывания длительных обработок попыток и их прерывания
+    Запускать как отдельный сервис
+    
+    Два значения читаются из конфига - макс. время попытки и интервал проверок
+    [constants]
+    processing_limit = ()
+    interval_time = ()
+    
+    Взаимодействие с базой данных:
+    - Использует TrainingsDBManager для доступа к тренировкам
+    - Находит все тренировки, у которых задано поле `processing_start_timestamp`
+    - Если прошло больше processing_limit секунд:
+        * Добавляет в базу вердикт с комментарием "обратиться к администраторам" через append_verdict()
+        * Устанавливает оценку 0 через set_score()
+        * Меняет статусы presentation и audio на PROCESSING_FAILED, если они ещё не финальные
+    - После каждой итерации ждёт interval_time секунд и повторяет проверку
+"""
+
+import time
+from datetime import datetime, timezone
+
+from app.config import Config
+from app.mongo_odm import TrainingsDBManager
+from app.root_logger import get_root_logger
+from app.status import TrainingStatus, PresentationStatus, AudioStatus
+
+logger = get_root_logger(service_name='processing_watchdog')
+
+DEFAULT_MAX_SECONDS = 300
+DEFAULT_INTERVAL_SECONDS = 30
+
+def get_config_values():
+    try:
+        return int(Config.c.constants.processing_limit), int(Config.c.constants.interval_time)
+    except Exception as e:
+        logger.warning("Failed to read config values, using defaults: %s", e)
+        return DEFAULT_MAX_SECONDS, DEFAULT_INTERVAL_SECONDS
+    
+def time_now():
+    return datetime.now(timezone.utc)
+
+def run_once(max_seconds):
+    """
+        Ищет все тренировки, у которых задан processing_start_timestamp
+        и проверяет их статус
+    """
+    now = time_now()
+    trainings = TrainingsDBManager().get_trainings()
+    candidates = []
+    for training in trainings:
+        process_started = getattr(training, "processing_start_timestamp", None)
+        if process_started is None:
+            continue
+        try:
+            started = datetime.fromtimestamp(process_started.time, timezone.utc)
+        except Exception as e:
+            logger.warning("Invalid timestamp in training %s: %s", getattr(training, "_id", "?"), e)
+        elapsed = (now - started).total_seconds()
+        if elapsed <= max_seconds:
+            continue
+        status = getattr(training, "status", None)
+        if status in [TrainingStatus.PROCESSED, TrainingStatus.PROCESSING_FAILED, TrainingStatus.PREPARATION_FAILED]:
+            continue
+        candidates.append((training, elapsed))
+    for training, elapsed in candidates:
+        training_id = getattr(training, "_id", None)
+        if training_id is None:
+            continue
+        try:
+            msg = ("Техническая ошибка: время обработки превысило лимит {:.0f} секунд. "
+                   "Оценка выставлена автоматически как 0. Пожалуйста, обратитесь к администраторам."
+                   ).format(elapsed)
+            logger.warning("Training %s exceeded processing timeout (%.0f s)", training_id, elapsed)
+            TrainingsDBManager().append_verdict(training_id, msg)
+            TrainingsDBManager().set_score(training_id, 0)
+            try:
+                pres_status = getattr(training, "presentation_status", None)
+                if pres_status not in [PresentationStatus.PROCESSED, PresentationStatus.PROCESSING_FAILED]:
+                    TrainingsDBManager().change_presentation_status(training_id, PresentationStatus.PROCESSING_FAILED)
+            except Exception as e:
+                logger.exception("Failed to mark presentation status as failed for training %s: %s", training_id, e)
+            try:
+                audio_status = getattr(training, "audio_status", None)
+                if audio_status not in [AudioStatus.PROCESSED, AudioStatus.PROCESSING_FAILED]:
+                    TrainingsDBManager().change_audio_status(training_id, AudioStatus.PROCESSING_FAILED)
+            except Exception as e:
+                logger.exception("Failed to mark audio status as failed for training %s: %s", training_id, e)
+        except Exception as e:
+            logger.exception("Error while handling timeout for training %s: %s", training_id, e)
+            
+def run():
+    max_seconds, interval_time = get_config_values()
+    logger.info("Processing watchdog started. Timeout = %s s, interval time = %s s", max_seconds, interval_time)
+    while True:
+        try:
+            run_once(max_seconds)
+        except Exception as e:
+            logger.exception("Unhandled error in processing watchdog main loop (%s)", e)
+        time.sleep(interval_time)
+        
+if __name__ == "__main__":
+    import sys
+    if len(sys.argv) >= 2:
+        Config.init_config(sys.argv[1])
+    run()
\ No newline at end of file
diff --git a/app_conf/config.ini b/app_conf/config.ini
index 553e675..97c80a7 100644
--- a/app_conf/config.ini
+++ b/app_conf/config.ini
@@ -6,6 +6,8 @@ lti_consumer_secret=supersecretconsumersecret
 version_file=VERSION.json
 backup_path=../dump/database-dump/
 storage_max_size_mbytes=20000
+processing_limit = 600
+interval_time = 30
 
 [mongodb]
 url=mongodb://db:27017/
diff --git a/app_conf/testing.ini b/app_conf/testing.ini
index dd228f4..75c2290 100644
--- a/app_conf/testing.ini
+++ b/app_conf/testing.ini
@@ -6,6 +6,8 @@ lti_consumer_secret=testing_lti_consumer_secret
 version_file=VERSION.json
 backup_path=../dump/database-dump/
 storage_max_size_mbytes=20000
+processing_limit = 600
+interval_time = 30
 
 [mongodb]
 url=mongodb://db:27017/
diff --git a/docker-compose.yml b/docker-compose.yml
index 303a30e..4c07ba0 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -86,6 +86,13 @@ services:
     volumes:
       - whisper_models:/root/.cache/whisper
 
+  processing_watchdog:
+    image: wst-image:v0.2
+    command: python3 -m processing_watchdog $APP_CONF
+    restart: always
+    depends_on: 
+      - db
+
 volumes:
   whisper_models:
   nltk_data: