Home Assistant 2026.8x and Fixes
V42.0.4 – Bug Fixes - based on findings from users (GitHub + Simon42 Forum)
Fuel my late-night ideas with a coffee? I'd really appreciate it — keep this project running!
Here are the most important changes:
Home Assistant 2026.8.x ready!
Astronomy Data & Startup Issues
- The error “SFML astronomy database is unavailable or invalid” / “database is locked” has been fixed.
The integration now accesses the astronomy data cleanly and in a controlled way. - Short delays when accessing the database no longer cause the morning forecast to abort.
- The misleading message “refresh was coalesced” no longer appears.
- Normal database activities during startup are no longer falsely reported as errors.
- When reloading or unloading the integration, no delayed background tasks keep running.
- Database timing control is more robust and now also works with Home Assistant 2026.8.x.
AI Model Management
- The first AI model can now be activated more reliably as soon as enough comparison data is available.
- Selection of the best model candidates has been improved:
– Candidates with already collected shadow evidence are still preferred.
– If no candidate has useful evidence, the qualitatively best model is chosen (instead of simply the oldest one). - Previously deferred model candidates can now be re-evaluated in a controlled way without rewriting the immutable model history.
Shading Detection
- Complete group data is no longer incorrectly discarded as “incomplete”.
- Under rain and heavy cloud cover, reduced production is no longer so quickly classified as “local shading” (trees, buildings, etc.) – especially when no local irradiance or lux sensor is available.
- The diagnosis now more clearly distinguishes between:
– general loss compared to an ideal clear sky and
– actual local shading.
MPPT Detection
- Battery limitations, zero-export situations, unexpected cloud cover, etc. are no longer shown as “MPPT throttling” by default.
Only hours that were actually classified as MPPT throttling now appear as such.
Diagnostic Attributes
- The displayed values (feature count, training data, last training time, model status, current shading, etc.) are now always taken from the current runtime and database state and no longer from a possibly outdated cache.
If you have any questions or notice further issues, just let me know.
Credits:
- Greennothing for the excellent and thorough SQL review
- Joachim-xo for testing the UI
- Everyone who actively contributes and supports in the forum!
Disclaimer:
SFML and its components are a private development and independent of the forum owner Simon42.