improve monitoring resilience

config.py

@@ -12,6 +12,8 @@ class Settings:
     minecraft_port: int
     poll_interval_seconds: int
     status_file_path: str
+    request_timeout_seconds: int
+    offline_after_failures: int
 
 
 def load_settings() -> Settings:
@@ -23,6 +25,8 @@ def load_settings() -> Settings:
     port = os.getenv("MINECRAFT_PORT", "25565")
     interval = os.getenv("POLL_INTERVAL_SECONDS", "30")
     status_file_path = os.getenv("STATUS_FILE_PATH", "data/status.json")
+    request_timeout = os.getenv("REQUEST_TIMEOUT_SECONDS", "5")
+    offline_after_failures = os.getenv("OFFLINE_AFTER_FAILURES", "2")
 
     if not token:
         raise RuntimeError("TELEGRAM_BOT_TOKEN is required")
@@ -41,6 +45,16 @@ def load_settings() -> Settings:
     except ValueError as exc:
         raise RuntimeError("POLL_INTERVAL_SECONDS must be an integer") from exc
 
+    try:
+        parsed_timeout = int(request_timeout)
+    except ValueError as exc:
+        raise RuntimeError("REQUEST_TIMEOUT_SECONDS must be an integer") from exc
+
+    try:
+        parsed_offline_after = int(offline_after_failures)
+    except ValueError as exc:
+        raise RuntimeError("OFFLINE_AFTER_FAILURES must be an integer") from exc
+
     return Settings(
         telegram_bot_token=token,
         telegram_chat_id=int(chat_id),
@@ -48,4 +62,6 @@ def load_settings() -> Settings:
         minecraft_port=parsed_port,
         poll_interval_seconds=parsed_interval,
         status_file_path=status_file_path,
+        request_timeout_seconds=parsed_timeout,
+        offline_after_failures=parsed_offline_after,
     )