More than about two incidents per 12-hour on-call shift is too many, according to Google's SRE book. The reasoning is simple: each incident takes hours to handle and follow up, so past two you cannot investigate properly or learn anything. The ideal median is zero pages a day.
The reasoning behind two
The SRE book says dealing with an incident takes about six hours on average once follow-up work is counted. That includes triage, mitigation, writing it up and filing the fixes. Two incidents in a shift already fills most of the working capacity of the person carrying the pager.
The chapter on service reliability makes the same point from the other side. Regularly exceeding two events per shift leaves engineers unable to investigate thoroughly or learn from what happened, and it causes pager fatigue. Fewer than one event per shift, it adds, wastes on-call time.
Measure before you argue
You cannot negotiate for relief without numbers. For a month, record for each shift:
- pages received, split by hours worked and hours off
- pages that needed human judgment versus pages that were routine
- time from page to first useful action
- follow-up tickets created
A median above one incident per day is the book's warning sign that a component is in trouble.
What to do when the number is high
The book lists several remedies for operational overload: temporarily loaning an experienced engineer to the noisy team, negotiating with the teams that own the noisy systems, and in extreme cases handing the pager back to the owners.
For a security team the same moves translate directly.
- Ask the owners of noisy detections to fix or retire them.
- Move low-urgency alerts to a queue reviewed in daylight.
- Bring in a second pair of hands for a few weeks while you fix the top three sources of pages.
- Put a threshold in writing, for example fewer than two paging events per shift, and review it every quarter.
Pages are not all equal
A page at 3 a.m. costs more than one at 2 p.m. If you want a richer measure, count night pages separately and track how often the same person receives them. A rotation that looks fine on average can still be hurting one person.
A note on tracking honestly
Pager data tells you how often you were interrupted, not how heavy each interruption was. Add a short comment field when closing a page: routine, needed thought, or major. After a month you will see whether most of your pages are noise.
Share the data with the owners of noisy systems, not as a complaint but as a request with evidence. A table of the top five sources of pages, with counts, gets a faster response than a general statement that the pager is too loud.
Quick check
Is zero pages in a shift a problem?
Not usually. The book calls a median of zero pages per day ideal. The risk of too few pages is skill decay, covered in a separate post on underload.
Sources
Last reviewed: 19 October 2026.
