Hi,
I recently realized an important Home Assistant limitation that may affect
EMHASS users relying on the websocket connection for historical data fitting.
For context: during the development and testing of the websocket-based history
retrieval, my Home Assistant instance still had the relevant data in
statistics_short_term. This meant 5-minute resolution data was still available
via the websocket API, so the approach worked as expected at the time.
I recently changed purge_keep_days: 360 -> purge_keep_days: 10 in HA and ran into issues.
If Home Assistant’s purge_keep_days is smaller than EMHASS’
historic_days_to_retrieve, the websocket API no longer returns high-resolution
(5-minute) state history. Instead, it falls back to long-term statistics with
~1 value per hour.
Key points:
- recorder purging removes high-resolution history
- websocket returns long-term statistics once raw data is gone
- long-term statistics are hourly aggregates
- this makes websocket-based fitting unsuitable for longer historical horizons
InfluxDB does not have this limitation and remains a reliable data source for
model fitting (I'm so happy I had an influx backup).
The websocket connection still makes sense for:
- MPC / real-time optimization
- short-term history windows
- potential future live sensor tracking
This is not a bug in EMHASS, but an HA architectural constraint that may be worth:
- documenting clearly
- warning about in the config
- or validating against
purge_keep_days
Posting this mainly to prevent others from unknowingly relying on websocket
history for fitting over long periods.
Hi,
I recently realized an important Home Assistant limitation that may affect
EMHASS users relying on the websocket connection for historical data fitting.
For context: during the development and testing of the websocket-based history
retrieval, my Home Assistant instance still had the relevant data in
statistics_short_term. This meant 5-minute resolution data was still availablevia the websocket API, so the approach worked as expected at the time.
I recently changed
purge_keep_days: 360->purge_keep_days: 10in HA and ran into issues.If Home Assistant’s
purge_keep_daysis smaller than EMHASS’historic_days_to_retrieve, the websocket API no longer returns high-resolution(5-minute) state history. Instead, it falls back to long-term statistics with
~1 value per hour.
Key points:
InfluxDB does not have this limitation and remains a reliable data source for
model fitting (I'm so happy I had an influx backup).
The websocket connection still makes sense for:
This is not a bug in EMHASS, but an HA architectural constraint that may be worth:
purge_keep_daysPosting this mainly to prevent others from unknowingly relying on websocket
history for fitting over long periods.