MQTT trouble

gauge.py

@@ -877,6 +877,28 @@ def _subscribe_all(c):
         c.subscribe(f"{prefix}/status_led/green/set")
 
 
+def _mqtt_connect_with_timeout(client, timeout_s=10):
+    """Call client.connect() with a socket-level timeout to avoid blocking forever."""
+    try:
+        import usocket as _sm
+    except ImportError:
+        import socket as _sm
+    _orig = _sm.socket
+    def _timed(*a, **kw):
+        s = _orig(*a, **kw)
+        s.settimeout(timeout_s)
+        return s
+    _sm.socket = _timed
+    try:
+        client.connect()
+    finally:
+        _sm.socket = _orig
+        try:
+            client.sock.settimeout(None)
+        except Exception:
+            pass
+
+
 def connect_mqtt():
     global client_ref, _mqtt_connected
     info(f"Connecting to MQTT broker {MQTT_BROKER}:{MQTT_PORT} ...")
@@ -896,10 +918,9 @@ def connect_mqtt():
                 user=MQTT_USER,
                 password=MQTT_PASSWORD,
                 keepalive=30,
-                socket_timeout=10,
             )
             client.set_callback(on_message)
-            client.connect()
+            _mqtt_connect_with_timeout(client)
             client_ref = client
             _mqtt_connected = True
             info(f"MQTT connected  client_id={MQTT_CLIENT_ID}")
@@ -962,10 +983,9 @@ def check_mqtt():
                 user=MQTT_USER,
                 password=MQTT_PASSWORD,
                 keepalive=30,
-                socket_timeout=10,
             )
             client_ref.set_callback(on_message)
-            client_ref.connect()
+            _mqtt_connect_with_timeout(client_ref)
             _mqtt_connected = True
             info("MQTT reconnected!")
             schedule_discovery()
@@ -1304,7 +1324,21 @@ def main():
 
     connect_wifi(WIFI_SSID, WIFI_PASSWORD, force_reconnect=False)
 
+    mqtt_attempts = 0
+    while True:
+        try:
             connect_mqtt()
+            break
+        except Exception as e:
+            mqtt_attempts += 1
+            log_err(f"MQTT connect failed: {e} (attempt {mqtt_attempts})")
+            if mqtt_attempts % 3 == 0:
+                log_err("WiFi may be stale — forcing reconnect...")
+                try:
+                    connect_wifi(WIFI_SSID, WIFI_PASSWORD, force_reconnect=True)
+                except Exception as we:
+                    log_err(f"WiFi reconnect failed: {we}")
+            utime.sleep_ms(5000)
     _subscribe_all(client_ref)
     schedule_discovery()