Polling profiles

By default every discovered model is read on every poll cycle. Models that rarely change, such as common (1), nameplate (120) and settings (121), can be read less often, and models you do not need can be skipped. Both cut Modbus traffic and Kafka volume.

[poller]
poll_interval_ms = 1000
exclude_models = [123]            # never read the immediate controls model

[poller.model_intervals_s]        # model ID = seconds
1 = 3600                          # common: hourly
120 = 3600                        # nameplate: hourly
121 = 600                         # settings: every 10 minutes

The same in the environment (for the systemd environment file or Docker):

SUNSPEC_MODEL_INTERVALS_S=1=3600,120=3600,121=600
SUNSPEC_EXCLUDE_MODELS=123

How the schedule works:

  • Every polled model is read on the first cycle of a session, so a device's identity and settings are published as soon as it connects. After a successful read, a model with an interval is read again on the first cycle at least that long later. A failed read is retried on the next cycle.
  • Intervals are counted in whole poll cycles: with poll_interval_ms = 1000, 121 = 600 reads model 121 every 600th cycle.
  • A new session (after a reconnect, a stall or repeated failures) reads every model again, so a slow model can be read more often than its interval while a device is unstable.
  • A cycle in which nothing is due does not contact the device and does not count towards the stall timeout.
  • include_models and exclude_models filter the discovered chain. Listing a model that a device does not have is not an error. If nothing on a device is selected, its session ends with none of the discovered models [...] is selected for polling and is retried with the usual backoff; the chain is rediscovered each time, so a firmware update that adds a model is noticed.
  • Discovery still reads every model header. The discovered SunSpec models log line lists the whole chain (models), the models that are polled (polled) and the slow ones (slow).

Per-device settings

A [[discovery.static_devices]] entry in a config file can override these settings for that device:

KeyOverridesNotes
poll_interval_mspoller.poll_interval_ms≥ 1. The device's default stall timeout follows it.
request_timeout_mspoller.request_timeout_ms≥ 1.
base_addresssunspec.base_addressProbed first; the standard bases follow.
max_batch_sizemodbus.max_batch_size1–125.
framingmodbus.framingtcp or rtu.
include_modelspoller.include_modelsReplaces the global list.
exclude_modelspoller.exclude_modelsReplaces the global list; [] clears it for this device.
model_intervals_spoller.model_intervals_sMerged model by model; the device's entries win.
[poller.model_intervals_s]
1 = 3600
120 = 3600

[[discovery.static_devices]]
ip = "192.168.1.30"               # a slow meter behind a serial gateway
unit_id = 2
poll_interval_ms = 5000
request_timeout_ms = 3000
max_batch_size = 60

[[discovery.static_devices]]
ip = "192.168.1.40"               # an inverter at unit 126 with MPPT data every 10 s
unit_id = 126
include_models = [1, 120, 160]
model_intervals_s = { 1 = 86400, 160 = 10 }   # merged: 1 daily, 120 hourly, 160 every 10 s
  • Per-device settings exist only in the config file. SUNSPEC_STATIC_DEVICES replaces the file's device list together with these settings, and logs a warning when it does.
  • Devices found by a subnet scan use the global settings.
  • After its overrides are applied, a device's include_models and exclude_models may not share a model; the error names the device.