Issue related to
Sunsynk Home Assistant Add-On
Describe the issue/bug
AInverter.read_identity() builds an Identity Component over self.inv.unit and calls
async_update(). That reads holding registers 0-7 directly off the unit, so it never goes through
Sunsynk.read_holding_registers() — the only place READ_ATTEMPTS is applied. There is no retry
anywhere in modbus_connection.model, so the identity read gets exactly one attempt no matter what
READ_ATTEMPTS is set to.
Because main_loop() turns any ConnectionError out of ist.connect() into return 2, one
dropped reply on one inverter aborts the whole add-on.
The asymmetry is visible in the log — every other read failure carries [attempt n/3], the fatal
one does not:
[11:33:14] ERROR Read register 552 x7: ModbusTimeoutError: read_holding_registers(552, 7): Response timeout after 2 seconds [attempt 1/3]
[11:33:19] INFO No response on the Modbus interface /dev/ttyACM1: read_holding_registers(0, 8): Response timeout after 2 seconds
[11:33:19] CRITICAL This Add-On will terminate in 30 seconds, use the Supervisor Watchdog to restart automatically.
With three inverters this becomes a restart loop: whichever port was left mid-transaction by the
previous SIGKILL is the one that misses next, and the ~6 s watchdog restart never gives the port
time to settle. Ports rotate between runs — 11:33:02 /dev/ttyACM2, 11:33:19 /dev/ttyACM1.
Lowering TIMEOUT makes it much more likely, since the single-shot read has no retries to absorb a
miss, but it also happens at TIMEOUT: 3.
Expected behavior The identity read honours READ_ATTEMPTS like every other holding-register
read, including the serial flush between attempts.
Your environment
- Addon:
- Name: sunsynk-edge
- Version: 246721b
- Inverter:
- Make: Sunsynk
- Model: high-voltage three-phase hybrid, 3 x 50 kW (identity reports device type
0x6,
protocol 1.4)
- Adaptor details:
- RS485 over USB CDC,
/dev/ttyACM0 .. /dev/ttyACM2, one adaptor per inverter
Your configuration
INVERTERS:
- SERIAL_NR: "**********"
HA_PREFIX: SS1
MODBUS_ID: 1
PORT: /dev/ttyACM0
- SERIAL_NR: "**********"
HA_PREFIX: SS2
MODBUS_ID: 2
PORT: /dev/ttyACM1
- SERIAL_NR: "**********"
HA_PREFIX: SS3
MODBUS_ID: 3
PORT: /dev/ttyACM2
SENSOR_DEFINITIONS: three-phase-hv
READ_ALLOW_GAP: 2
READ_SENSORS_BATCH_SIZE: 20
READ_MESSAGE_SPACING: 0.05
READ_ATTEMPTS: 3
TIMEOUT: 2Logs
[11:33:10] INFO sunsynk library - commit 246721b - last stable version 1.2.0
[11:33:10] INFO Connecting to /dev/ttyACM0
[11:33:11] INFO Identity: device_type=High-voltage three-phase hybrid (0x6) protocol=1.4 serial=****18594
[11:33:14] ERROR Read register 552 x7: ModbusTimeoutError: read_holding_registers(552, 7): Response timeout after 2 seconds [attempt 1/3]
[11:33:17] INFO Connecting to /dev/ttyACM1
[11:33:17] INFO Reading inverter identity (registers 0-7)
[11:33:19] INFO No response on the Modbus interface /dev/ttyACM1: read_holding_registers(0, 8): Response timeout after 2 seconds
[11:33:19] CRITICAL This Add-On will terminate in 30 seconds, use the Supervisor Watchdog to restart automatically.
[09:33:19] WARNING: Addon resource not ready 2
s6-supervise sunsynk: warning: finish script lifetime reached maximum value - sending it a SIGKILL
[11:33:25] INFO: Starting the Sunsynk add-on
Issue related to
Sunsynk Home Assistant Add-On
Describe the issue/bug
AInverter.read_identity()builds anIdentityComponent overself.inv.unitand callsasync_update(). That reads holding registers 0-7 directly off the unit, so it never goes throughSunsynk.read_holding_registers()— the only placeREAD_ATTEMPTSis applied. There is no retryanywhere in
modbus_connection.model, so the identity read gets exactly one attempt no matter whatREAD_ATTEMPTSis set to.Because
main_loop()turns anyConnectionErrorout ofist.connect()intoreturn 2, onedropped reply on one inverter aborts the whole add-on.
The asymmetry is visible in the log — every other read failure carries
[attempt n/3], the fatalone does not:
With three inverters this becomes a restart loop: whichever port was left mid-transaction by the
previous SIGKILL is the one that misses next, and the ~6 s watchdog restart never gives the port
time to settle. Ports rotate between runs — 11:33:02
/dev/ttyACM2, 11:33:19/dev/ttyACM1.Lowering
TIMEOUTmakes it much more likely, since the single-shot read has no retries to absorb amiss, but it also happens at
TIMEOUT: 3.Expected behavior The identity read honours
READ_ATTEMPTSlike every other holding-registerread, including the serial flush between attempts.
Your environment
0x6,protocol 1.4)
/dev/ttyACM0../dev/ttyACM2, one adaptor per inverterYour configuration
Logs