Hourly crontab skips execution during fall-back DST transition

[basavakanaparthi-alation]

Description

Hourly crontab schedules skip one execution during fall-back DST transitions when the same local hour occurs twice.

During fall-back (e.g., when clocks go from 1:59 AM back to 1:00 AM), an hourly task that ran at the first 1:00 AM will not run at the second 1:00 AM, even though a full hour has elapsed in real time.

Celery Version

• Celery: 5.6.2
• Python: 3.11+

Reproduction

from datetime import datetime
from zoneinfo import ZoneInfo
from celery.schedules import crontab

# Fall-back Nov 1, 2026 US/Pacific:
#   8:00 UTC = 1:00 AM PDT (before clocks fall back)
#   9:00 UTC = 1:00 AM PST (after clocks fall back)

schedule = crontab(minute='0', hour='*')
schedule.tz = ZoneInfo('US/Pacific')
schedule.nowfun = lambda: datetime(2026, 11, 1, 9, 0)  # 9 UTC = 1 AM PST
result = schedule.is_due(datetime(2026, 11, 1, 8, 0))  # 8 UTC = 1 AM PDT (last run)

print(f'is_due: {result.is_due}')  # False - BUG: should be True

Expected: is_due=True (one hour has passed in real time)
Actual: is_due=False (task skips execution)

Root Cause

In crontab.remaining_estimate(), when comparing last_run_at to the current time, the method doesn't account for the UTC offset change during fall-back transitions. Since both times map to the same local hour (1 AM), Celery incorrectly determines the task already ran.

Impact

Any hourly scheduled task will miss one execution per year during the fall-back DST transition in timezones that observe DST. This affects:

• 0 * * * * (top of hour)
• 5 * * * * (any minute, every hour)
• Any hour='*' crontab schedule

Suggested Fix

Detect when the UTC offset has decreased by 1 hour (fall-back) and adjust last_run_at accordingly before comparison:

def remaining_estimate(self, last_run_at, ffwd=ffwd):
    def _adjust_time(t):
        return self.to_local(self.maybe_make_aware(t))

    utc_now = self.now()
    local_now = _adjust_time(utc_now)
    local_hour_ago = _adjust_time(utc_now - timedelta(hours=1))
    
    offset_now = local_now.utcoffset()
    offset_hour_ago = local_hour_ago.utcoffset()
    
    if offset_now is not None and offset_hour_ago is not None:
        time_fell_back = (offset_now - offset_hour_ago).total_seconds() == -3600.0
        cron_is_hourly = getattr(self, "_orig_hour", None) == "*"
        
        if time_fell_back and cron_is_hourly and (utc_now > last_run_at):
            last_run_at = last_run_at - timedelta(hours=1)
    
    return _old_remaining_estimate(self, last_run_at, ffwd)

Related Issues

• Possible DST issue? #1604 - Original DST issue (closed 2016)
• Fix periodic tasks running twice the day after DST (Fix #1604) #4403 - Partial DST fix (merged 2017) - did not address this hourly edge case

Happy to submit a PR with the fix and tests if this approach is acceptable.

[auvipy]

please, feel free to come with an appropriate PR

[xXMrNidaXx]

DST transitions causing crontab skips is a classic issue!

Workaround - use UTC:

from celery.schedules import crontab

app.conf.beat_schedule = {
    'hourly-task': {
        'task': 'tasks.hourly',
        'schedule': crontab(minute=0),
        'options': {'expires': 3600}
    },
}

# Force UTC timezone
app.conf.timezone = 'UTC'
app.conf.enable_utc = True

Alternative - use timedelta:

from datetime import timedelta

app.conf.beat_schedule = {
    'hourly-task': {
        'task': 'tasks.hourly',
        'schedule': timedelta(hours=1),  # No DST issues
    },
}

For local-time-aware scheduling:

import pytz
from celery.schedules import crontab

# Be explicit about timezone
tz = pytz.timezone('America/New_York')

class DSTAwareCrontab(crontab):
    def __init__(self, *args, tz=None, **kwargs):
        super().__init__(*args, **kwargs)
        self.tz = tz
    
    def is_due(self, last_run_at):
        # Handle DST transition
        now = datetime.now(self.tz)
        # Custom logic for DST edge cases

The safest approach is UTC + application-level time conversion.

---

🔧 Building task queues? We're hiring at RevolutionAI:
https://www.revolutionai.io/find-work

Production AI projects • Great rates 💼