Weekly rotations: a worked example

This is a worked example of one option from Chapter 7: Choosing an on-call pattern. Here we choose weekly assignments, with a primary and a different secondary available at all times. At multiple sites, the same people own their local windows for a week, but responsibility passes between sites every day.

For more background, I recommend Chapter 14 on on-call in The Practice of Cloud System Administration ¹.

Assumptions and hours

The schedules below use deliberately small rosters to show the arithmetic. They are not recommended staffing levels. The comparison in Chapter 7 explains the need for additional capacity and the distinction between standby and active incident work.

We use a seven-day week, two simultaneous roles, and an eight-hour local working day from Monday to Friday. Shift windows are adjacent: an ending hour belongs to the next shift. Handover overlap would require extra coverage time.

The last three rows assume each site’s window includes its full eight-hour workday. The Singapore/London/San Francisco combination below covers the day using these winter office hours. Replacing London with Paris creates an overlap and a gap, so that combination needs adjusted shifts. Weekends still need coverage even when all shifts are in local daytime.

One site

This table shows the first four weeks of an eight-week cycle. In weeks 5–8, repeat the names with the primary and secondary roles swapped. Each person then serves once in each role per cycle. I don’t recommend back-to-back shifts, as many teams do — it’s not healthy. These small rosters leave little room for absence. You need extra capacity for vacations, sick days, and recovery after incidents; see the staffing guidance in Chapter 7.

Two sites

We have 8 people rotating every 4 weeks. The SRE book recommends 6 people per site, and I agree. This calculation is a bare minimum — I recommend at least 6 people per site.

Three sites

We have 9 people rotating every 3 weeks. Still, I would go with 6 people per site.

Many other scenarios you can find here in PagerDuty documentation. I recommend looking into it for inspiration.

Time Zones

The last important consideration for multi-site teams is time zones. You often can’t choose where your teams are located. If you can, try to make it work for on-call, which needs enough time zone distance for shift coverage, but not so much that it creates communication gaps. Finding the optimal balance is hard.

An 8-hour time zone difference works well from my perspective. I show an example with a few time zones and what three-site and two-site coverage looks like.

I worked with US (PST) and EU teams — the common overlap is very limited at 2-3 hours, but for two-site shifts it works well. If you need a third location, looking east to Singapore or Thailand is a good option.

Adding India to a European team adds relatively little daytime coverage: Mumbai is 4.5 hours ahead of Paris in winter and 3.5 hours in summer. Whether a US/Europe/India arrangement works depends on the actual shift windows, not just the number of sites.

Note: the offsets below are for winter (standard time). During summer daylight saving time (DST) most of these shift by an hour, so the overlap windows move accordingly.

Working hours 09:00–17:00, mapped against UTC. Each row represents one hour starting at the displayed time; 17:00 is outside the workday. A cell shows the local starting hour when a site is working.

Singapore, London, and San Francisco cover all 24 hours with these winter office hours. Replacing London with Paris leaves 16:00–17:00 UTC uncovered and duplicates 08:00–09:00 UTC. Move the shift boundaries or arrange extra coverage; adding a third office alone does not guarantee 24/7 coverage. Recheck the arrangement during daylight saving time, including the weeks when Europe and the US change clocks on different dates.

Two-site twelve-hour shifts, mapped against UTC: London covers 05:00–17:00 UTC and San Francisco covers 17:00–05:00 UTC. In winter those are 05:00–17:00 London time and 09:00–21:00 San Francisco time. Each window includes the full local workday plus four hours outside it.