By Paul Clark ·
Running a Job at 9am in Three Time Zones: Cron, UTC and Timestamps That Survive DST
Why one UTC crontab line cannot mean 9am in Berlin, Chicago and Melbourne, the seven UTC hour combinations 9am lands on in one winter, and a dispatcher that stores timestamps.
The ticket says: “Send the daily summary email at 9am, user’s local time.” There are customers in Berlin, Chicago and Melbourne. The first draft is three crontab lines in UTC, one per city, and it works for about three weeks. Then Europe changes its clocks and the Berlin email arrives at 8am. A week later Chicago’s arrives at 8am too. In April the Melbourne one starts arriving at 10am.
The general traps are covered in five cron gotchas: cron has no built-in time zone, and a 2:30am job is skipped or run twice on changeover nights. This post is about a narrower problem those rules lead to. How do you schedule one event at the same local time for people in several zones, when every one of those zones moves on its own calendar?
9am is not one UTC hour, and it isn’t even seven
Here is where 9am local lands in UTC for our three cities, from early October 2026 to mid-April 2027. Europe, the US and Australia each change clocks on different dates, and Australia moves in the opposite direction:
| Dates | 9am Berlin | 9am Chicago | 9am Melbourne |
|---|---|---|---|
| until 3 Oct 2026 | 07:00 | 14:00 | 23:00 (day before) |
| 4 Oct – 24 Oct | 07:00 | 14:00 | 22:00 (day before) |
| 25 Oct – 31 Oct | 08:00 | 14:00 | 22:00 (day before) |
| 1 Nov – 13 Mar 2027 | 08:00 | 15:00 | 22:00 (day before) |
| 14 Mar – 27 Mar | 08:00 | 14:00 | 22:00 (day before) |
| 28 Mar – 3 Apr | 07:00 | 14:00 | 22:00 (day before) |
| from 4 Apr 2027 | 07:00 | 14:00 | 23:00 (day before) |
That’s seven different combinations in six months. This post goes live during the 25–31 October row, when Europe has already fallen back and the US hasn’t yet. A UTC-only schedule written for “Berlin and Chicago are 7 hours apart” is wrong right now, because this week they’re 6 apart. The gap opens twice a year, and the overlap guide’s section on transition dates explains why it affects meetings in the same way.
The Melbourne column hides a second trap that has nothing to do with daylight saving. 9am Monday in Melbourne is Sunday at 22:00 UTC. If the summary only goes out on weekdays, the UTC version of “Monday to Friday, Melbourne time” is 0 22 * * 0-4 (Sunday to Thursday), not 1-5. Copy the weekday field across from the Berlin line and Melbourne gets a Saturday email and no Monday email.
Option one: one cron line per zone, in that zone
If your scheduler accepts a time zone, the simplest correct answer is a line per zone, each written in local time:
CRON_TZ=Europe/Berlin
0 9 * * 1-5 send-summary --zone=Europe/Berlin
CRON_TZ=America/Chicago
0 9 * * 1-5 send-summary --zone=America/Chicago
CRON_TZ=Australia/Melbourne
0 9 * * 1-5 send-summary --zone=Australia/Melbourne
CRON_TZ works in cronie. Kubernetes CronJobs have a timeZone field, and systemd timers accept a zone at the end of OnCalendar=. Not every scheduler has one: GitHub Actions schedule: triggers, for example, run in UTC only. So check before you rely on it.
You can check the result in the Cron Expression Builder. Choose Australia/Melbourne in its time-zone menu, paste 0 9 * * 1-5, and it lists the next fire times in that zone (ten by default, up to 100), worked out against real calendar dates. To see the hand-converted version, switch the menu to UTC and paste 0 22 * * 0-4. UTC is always in the menu, even in browsers whose own zone list leaves it out. The two agree only until 4 April. After that the UTC line fires at 8am Melbourne time.
At 9am this approach is safe from the skipped-and-repeated-hour problem, since no zone changes its clocks at 9. It has one real limit: it scales with the number of zones, not the number of users. Three lines is fine. When customers sign up from forty zones, including India at UTC+5:30 and Nepal at UTC+5:45, a crontab per zone gets out of hand.
Option two: a UTC dispatcher that reads timestamps
The pattern that scales is to stop asking cron to know about users. Cron runs one dispatcher on a fixed UTC beat, and each user row records when that user is next due, as a Unix timestamp:
users
id | tz | next_send_at | last_sent_local_date
41 | America/Chicago | 1793545200 | 2026-10-31
The dispatcher runs */15 * * * * in UTC and does three things:
- Select users where
next_send_at <= now(). - Send, then set
last_sent_local_dateto today’s date in the user’s zone. - Compute the next 9am in the user’s IANA zone with a real time-zone library, convert it to epoch seconds and store it.
Why every 15 minutes rather than hourly? 9am in Kathmandu is 03:15 UTC and 9am in Kolkata is 03:30 UTC. An hourly dispatcher sends those users their 9am email at 9:45 or 9:30. Quarter-hourly covers every zone currently in use. (The overlap edge-cases guide has the full list of zones that aren’t on the hour.)
Comparing timestamps with <= also solves something cron can’t. Cron has no catch-up: a run missed while the box was down never happens. A dispatcher that was down from 08:50 to 09:20 will find the overdue rows on its first pass and send them late, not never. Add a cutoff that suits your product. A “good morning” summary that arrives at 2pm is probably better skipped, with next_send_at simply moved to tomorrow.
Never compute “tomorrow at 9” by adding 86,400
The tempting shortcut in step 3 is next_send_at += 86400. It’s wrong twice a year in every zone with daylight saving, and you can watch it go wrong in the Unix Timestamp Converter.
Choose America/Chicago, type 31 October 2026, 09:00 into the date picker, and the converter returns 1793455200. Add 86,400 and paste 1793541600 back in. It reads 1 November, 08:00 CST, because that day is 25 hours long. The real 9am on 1 November is 1793545200, which is 90,000 seconds after the previous one, not 86,400. In March the gap is 82,800 seconds. A server that adds a fixed day drifts an hour each time a transition passes, and it never drifts back on its own.
The fix is to build the next local date (2026-11-01 09:00 in America/Chicago) and let the library turn that into an instant. The converter does the same thing in the opposite direction: its date picker reads what you type as wall-clock time in the zone you’ve selected, not in your browser’s zone. Batch mode helps with checking. Paste a column of next_send_at values from the database, choose the user’s zone, and every row should say 09:00. Any row that says 08:00 or 10:00 was calculated by adding seconds.
Two policies to decide before you need them
Two small decisions keep the dispatcher honest.
Use the local date as the idempotency key. last_sent_local_date is what prevents a double send. If two dispatcher runs overlap, or a retry fires after a timeout, the second one sees that today (in the user’s zone) has already been handled and skips. Cron won’t stop two copies running at once, and a date check in the data costs less than a distributed lock.
Decide what happens to times that don’t exist or happen twice. 9am is safe, but if users can choose their own send time, someone will choose 02:30. On the spring-forward night that time doesn’t exist, and on the autumn night it happens twice. Choose a rule (“first occurrence”, “shift forward an hour”) and write it down, because libraries differ on this. The Unix Timestamp Converter shows that they do. Type 1 November 2026, 01:30 in America/Chicago and it takes the first occurrence, 1793514600 (CDT). Add 3,600 for the second one. Type a nonexistent 02:30 on a spring-forward date and it resolves to a real time an hour away and moves the picker to show you which. In Chicago it lands on 01:30 and in Berlin on 03:30. Your library may choose differently, so test it the same way.
See where 9am falls before you deploy
Here’s a quick visual check that uses the Time Zone Overlap Calculator for something other than meetings. Set working hours from 9 to 10 and list your users’ zones. Each row then shows exactly one cell (two for a half-hour zone): where that zone’s 9am sits on the UTC axis today. The overlap row will be empty, which is expected, because you’re reading the positions, not the overlap. Run it again after each changeover date in the table above and you’ll see the cells move.
Write per-zone schedules in the Cron Expression Builder with the zone selected, check stored next_send_at values in the Unix Timestamp Converter against the user’s own zone, and use the overlap band to check where your users’ mornings fall in UTC.
Try the tools from this guide
-
Cron Expression Builder & Translator
Build cron expressions visually and translate any cron string to plain English with the next 5 fire times.
-
Unix Timestamp Converter (with Timezone)
Bidirectional Unix timestamp ↔ human date converter with timezones, milliseconds and ISO 8601.
-
Time Zone Overlap Calculator (Remote Teams)
Find overlapping productive hours across up to 8 time zones for distributed teams.