- Run a primary and a backup. PagerDuty puts the backup shift directly after the primary shift.
- Google SRE caps on-call at 25% of an engineer's time, which means at least eight people for a single-site 24/7 rotation.
- Read the last handoff when your shift starts and write one when it ends.
- Whether on-call time counts as hours worked depends on how free the person is to use the time.
What is an on-call rotation?
Being on call means being reachable outside normal work to fix the systems you own. PagerDuty's guide puts it this way: "Being on-call means that you are able to be contacted at any time in order to investigate and fix issues that may arise for the system you are responsible for" (PagerDuty). The rotation is the schedule that shares that duty out, so one person is never on call all the time.
How fast the on-call engineer must respond depends on the service. Google's SRE book gives typical values of "5 minutes for user-facing or otherwise highly time-critical services, and 30 minutes for less time-sensitive systems" (Google SRE book). A page is the alert that reaches the on-call person's phone. Acknowledging it tells the paging tool, and the rest of the team, that someone is working on it.
Many teams have two people on call at once. The SRE book says "many teams have both a primary and a secondary on-call rotation" and that the split varies: the secondary may catch pages the primary misses, or take the non-urgent work while the primary takes pages (Google SRE book).
What on-call rotation schedules do teams use?
The right schedule depends on team size, time zones and how often the service pages. These are the common patterns:
| Schedule | How it works | Good for | Watch out for |
|---|---|---|---|
| Weekly 24/7 | One person covers a full week, day and night, then hands over | Small teams with a quiet service | Night pages wear people down by the end of the week |
| Daily | The duty moves every day | Teams that page often | More handoffs, so more chances to drop context |
| Follow the sun | Teams in different time zones each cover their own daytime | Teams with two or three sites | Handoffs between sites need to be written, not verbal |
| Business hours plus after hours | Different people for working hours and nights or weekends | Teams where nights are rare but serious | The after-hours person may lack context from the day |
| Primary plus backup | Two layers; the backup is paged if the primary does not answer | Almost every team | The backup needs the same access and runbooks |
Google prefers more than one site where the work justifies it, because "night shifts have detrimental effects on people's health" and "a multi-site "follow the sun" rotation allows teams to avoid night shifts altogether" (Google SRE book). PagerDuty adds that "a backup shift should generally come directly after a primary shift," so the person who just finished can pass on context (PagerDuty).
How many people do you need for a 24/7 on-call rotation?
Google's SRE teams cap on-call time: "no more than 25% can be spent on-call." With a primary and a secondary always on call and week-long shifts, "the minimum number of engineers needed for on-call duty from a single-site team is eight." For teams split across two sites, "a reasonable minimum size of each team is six" (Google SRE book).
The number of incidents matters as much as the number of hours. Google found that one incident, including the follow-up work and postmortem, "takes 6 hours" on average, so "the maximum number of incidents per day is 2 per 12-hour on-call shift" (Google SRE book). If a rotation regularly goes over that, the fix is fewer noisy alerts or more people, not longer shifts.
Smaller teams can still run a fair rotation. Cover only business hours with a pager, agree with a sister team to back each other up, or cut alerts down to the ones that truly need a person at night.
What should an on-call handoff note include?
The handoff is where context survives or gets lost. Google's SRE workbook describes a simple rule: "At the start of each shift, the on-call engineer reads the handoff from the previous shift," and "at the end of the shift, the on-call engineer sends a handoff email to the next engineer on-call" (Google SRE workbook). PagerDuty asks for "a quick summary to the next on-call about any issues that may come up during their shift" (PagerDuty).
The note below is an illustration written for this page. It lists open incidents with their IDs, things to watch, planned changes, silenced alerts and a count of pages. The IDs are what let anyone trace each item later, as shown in the incident timeline template.
On-call handoff: platform team, week 2026-W11 -> 2026-W12
From: on-call A (primary) To: on-call B (primary) Time: Mon 09:00 UTC
Open incidents
- INC-4471 (SEV-3) disk growth on log-store-2; volume grown, cause not found yet
Watch list
- api-gateway latency alert flapped 3 times on Thursday night; no action taken
- certificate for status.example.com expires in 9 days (TICKET-2210)
Changes planned this week
- Tue 14:00 UTC: database minor upgrade (CHG-1180), runbook linked in the change
Silenced alerts
- backup-job-late on build-03 until Wed 12:00 UTC (host being rebuilt, CHG-1176)
Pages last week: 6 (4 actionable, 2 noise). Noise alert ticket: TICKET-2214
How should on-call escalation work?
An escalation policy says who gets paged next when nobody answers. PagerDuty recommends that teams "set your escalation timeout to 5 minutes," and says the third level, after the backup, "should probably be your entire team" (PagerDuty). It also tells responders that there is no expectation to fix everything alone; its motto is "Never hesitate to escalate."
Write the escalation path into each runbook as well, so the on-call engineer knows when a problem is bigger than the steps in front of them. A runbook template with an escalation field makes this routine.
Is on-call time paid?
It depends on the employer and, in the US, on how restricted the person is while on call. This is general information, not legal advice. Under the federal rule at 29 CFR 785.17, an employee "required to remain on call on the employer's premises or so close thereto that he cannot use the time effectively for his own purposes is working while "on call"." An employee who only has to leave word where they can be reached "is not working while on call" (29 CFR 785.17). Time spent answering pages is a separate question from time spent waiting.
Many employers also pay for on-call duty by choice. Google's SRE book says it "offers time-off-in-lieu or straight cash compensation, capped at some proportion of overall salary," and that the cap limits how much on-call work any one person takes on (Google SRE book).
On-call rotation example: an eight-person platform team
This schedule is an illustration written for this page, not a real company's rotation. Eight engineers share a weekly 24/7 rotation with a primary and a backup. Each person is primary one week in eight and backup the week after, so each is on call two weeks in eight: 25% of their time.
| Week | Primary | Backup | Handoff | Notes |
|---|---|---|---|---|
| W10 | Engineer 1 | Engineer 8 | Mon 09:00 UTC | Engineer 8 was primary in W09 |
| W11 | Engineer 2 | Engineer 1 | Mon 09:00 UTC | New hire shadows Engineer 2 |
| W12 | Engineer 3 | Engineer 2 | Mon 09:00 UTC | Swap with Engineer 5 for travel, logged in the tool |
| W13 | Engineer 4 | Engineer 3 | Mon 09:00 UTC | Database upgrade week; extra review of runbooks |
New people shadow before they carry the pager. PagerDuty suggests new team members "shadow your on-call rotation during the first few weeks" (PagerDuty). Swaps go into the paging tool as overrides, so the record of who was on call stays correct.
Common mistakes with on-call rotations
Most rotations fail slowly, through tiredness and lost context rather than one bad night.
- No backup layer, so a missed page during a commute becomes an outage.
- Too few people, so each engineer is on call every other week.
- Verbal handoffs only, so open issues disappear at the change of shift.
- Swaps agreed in chat but not entered in the paging tool, so the wrong person gets paged and the schedule history is wrong.
- Noisy alerts left in place, so real pages are lost among false ones.
- Runbooks the on-call engineer cannot reach, or that need access they do not have.
Are on-call records also a data asset?
They can be, when they sit next to the incidents they cover. AI data buyers have asked Origin for the records of companies that run and defend their own infrastructure: incidents joined to the change that fixed them, configuration history, ordered logs of what happened, and the runbooks, SOC tickets, on-call records and postmortems from the same period. On-call records answer questions the incident ticket cannot: who was paged, when, how fast they acknowledged, whether the page escalated, and what the handoff note said they already knew. The records must cover the company's own infrastructure (a client's only where that client has agreed), and the company must own them. See the hub page for infrastructure and incident records and the incident timeline template.
Origin is an independent data partner. Your company decides what is shared and signs any license itself, and a buyer may pay Origin when a deal closes.
- Personal data comes out first: names, phone numbers and emails of on-call staff become consistent pseudonyms, such as Engineer 1.
- Pay, leave and travel reasons for swaps stay out; only the fact of the swap matters.
- Secrets pasted into handoff notes or pages are removed and rotated (remove secrets from git history).
- Customer names and client environments come out unless the client has agreed.
Does your company hold records like these? See what your data is worth.
Common questions
How long should an on-call shift be?
Most teams use one-week shifts, handing over on a weekday morning so the handoff happens during work hours. Teams that page often move to daily or 12-hour shifts so nobody carries a week of broken sleep. Google SRE also limits the load per shift to about two incidents in 12 hours.
How many engineers do you need for a 24/7 on-call rotation?
Google SRE sets the minimum at eight engineers for a single-site team that runs a primary and a secondary with week-long shifts, so each person is on call no more than 25% of the time. A team split across two sites needs about six people at each site.
What is a follow-the-sun on-call rotation?
It is a schedule where teams in different time zones each cover the hours that fall in their own working day, then hand over to the next site. It removes most night shifts, which Google SRE cites as a health reason to prefer it, but it needs clear written handoffs between sites.
What is a good escalation timeout for on-call?
Five minutes is a common setting, and it is what PagerDuty recommends for its own teams. If the primary has not acknowledged the page in that time, the paging tool moves to the backup, and after the backup to the whole team.
Is on-call time considered hours worked in the US?
Under 29 CFR 785.17, waiting time counts as work when the employee must stay at or so close to the employer's premises that the time cannot be used for their own purposes. If they only need to leave word where they can be reached, it does not. Time spent answering pages is work. This is general information, not legal advice.
Should managers be part of the on-call rotation?
Many teams include them. PagerDuty says team managers can and should be in the normal rotation because it gives them a better view of what is going on. It keeps managers close to the real alert load, which helps when deciding whether the team needs more people or fewer noisy alerts.
How do you stop on-call from burning out the team?
Keep enough people in the rotation that nobody is on call more than about a quarter of the time, cut noisy alerts, pay or give time off for the duty, and let people swap shifts freely. Consider giving the next morning off after a night of pages.
