How to Fix Modbus “Queue Full” & Dropouts in Solar Assistant / HA

This is a frustratingly familiar situation. When the system load is low or at night, your local monitoring setup works well. But around 1:00 PM, when the sun is at its peak and your inverters and BMS are working hard, your Solar Assistant or Home Assistant dashboard goes dead.

2026-07-15 13:02:14 ERROR (MainThread) [modbus] Pymodbus: Modbus Error: [Connection] ModbusTcpClient failed
2026-07-15 13:02:16 WARNING (MainThread) [modbus] Modbus update failed: Queue Full
2026-07-15 13:02:18 ERROR (MainThread) [modbus] Dropping packets: RS485 Bus Timeout

You haven’t wired anything wrong, and your equipment isn’t broken. You are experiencing a Modbus RTU Channel Flood.

The Root Cause: The Baud Rate Bottleneck

RS485 is a half-duplex communication standard. This means only one device can talk at a time. Most legacy solar equipment (like Pylontech batteries, Deye/Growatt inverters) communicate at a sluggish 9600 bps.

During peak solar generation, the internal state of these devices changes wildly every millisecond. If your software (like Home Assistant) is aggressively polling 50 different registers across 3 different devices every second, the physical wire literally cannot carry the data fast enough.

The queries pile up in the software’s queue until the memory buffer overflows, causing the OS to panic and drop the serial connection entirely.

Option A: The Free Software Fix (Tuning the Polling Cycle)

To stop the crashing, you must force your monitoring software to “breathe” by adding mandatory delays between queries. This prevents the queue from filling up.

# In your configuration.yaml, add these critical delays to your modbus hub
modbus:
  - name: solar_hub
    type: serial
    port: /dev/ttyUSB0
    baudrate: 9600
    bytesize: 8
    method: rtu
    parity: N
    stopbits: 1
    timeout: 5                     # Increase timeout to 5 seconds
    message_wait_milliseconds: 200 # Force a 200ms pause between EVERY query
    
    sensors:
      - name: "Grid Power"
        address: 86
        scan_interval: 10          # Do not poll faster than every 10 seconds