Three fixes merged into Home Assistant Core on 21 October 2026 deal with the hour that happens twice when daylight saving time ends. They touch the schedule helper, time_pattern triggers and long-term statistics. All three came from Raman Gupta, were reviewed by Paulus Schoutsen, and were found with property-based testing.
The timing matters for Europe: clocks go back on Sunday 25 October 2026, and in the US on 1 November. The fixes went into the dev branch, which becomes Home Assistant 2026.11 in early November. The pull requests have no milestone for a 2026.10 patch release, and the current stable version, 2026.10.2 from 16 October, does not include them.
The schedule helper could spin thousands of times per second
The worst of the three affects the schedule helper. Take a Sunday block from 01:00 to 01:45. During the second pass through 01:30, the helper picked the first 01:45 as its next event, even though that moment was already 45 minutes in the past. The timer fired at once and re-armed the same past event, thousands of times per second, until 01:45 came around again.
The cause was Python's datetime comparison. Times in the same time zone are compared without looking at the fold flag that marks the second pass through a repeated hour, so the first 01:45 looked later than the second 01:30. The helper now builds candidate times with the same fold as the current moment, so inside the repeated hour it picks the upcoming occurrence.
Only schedules with a block that starts or ends inside the repeated hour are affected.
Time pattern triggers in unusual time zones
The second fix concerns time_pattern triggers in places where the clock moves back by half an hour or by an amount that does not land on a full hour. The pull request uses Pacific/Chatham, where clocks go back from 03:45 to 02:45, and mentions Lord Howe Island with its 30-minute change. A trigger set to hour 2 should fire again at 02:45 that night, but Home Assistant jumped to 02:00 the next day. The search for the next match now starts at the beginning of the repeated period. Whole-hour changes, as in Europe and the US, were not affected by this one.
Statistics lost the last day, week or month
The third fix is in the recorder. When long-term statistics were read by day, week or month, a period that contained the fall-back hour could vanish from the result if the data ended in the first hour of that period. A day with the clock change lasts 25 hours, and the code closed periods with a fixed length, so the final row was never emitted. It now uses the real end of the period. That happens, for example, when the statistics are read soon after midnight on the day of the change.