There's an engineer on every team you can spot a mile off: the one who's been on-call too long, on a rotation that pages them three times a night for things that didn't matter. They're exhausted, a little jumpy, and quietly updating their CV.
On-call doesn't have to be like that. Done well, it's a fair, well-understood safety net that the whole team shares and trusts. Done badly, it's a slow-motion way to lose your best people. The difference is almost entirely in the practices — here are the ones that matter.
What "on-call" really is
On-call is simply whose job it is to respond when something breaks at any given moment — including nights and weekends. Someone has to be the person who picks up when the pager goes off. The goal isn't to make that heroic; it's to make it rare, fair, and humane.
The practices that make it humane
1. Make ownership crystal clear
At any moment, exactly one person should know "I'm on-call right now." Ambiguity — "I thought you had it" — is how incidents get missed. A clear schedule, visible to everyone, removes all doubt.
2. Build a real escalation path
The primary on-call shouldn't be a single point of failure. If they don't acknowledge in a few minutes (they're driving, asleep, in a tunnel), it should automatically escalate:
| Tier | Role | Kicks in |
|---|---|---|
| Primary | First responder | Immediately |
| Secondary | Backup | If primary doesn't ack (~5 min) |
| Manager | Last resort | If neither responds (~15 min) |
Nobody should ever be the only thing standing between an outage and a fix.
3. Keep the rotation fair
Spread the load. A rotation where the same person always covers weekends, or where one team of three is permanently on-call, burns people out fast. Rotate evenly, account for time zones, and make sure everyone gets real time off the rotation.
4. Only page for things worth waking up for
This is where on-call lives or dies. If the pager goes off for non-actionable noise, people stop trusting it — and miss the real thing. Every page should pass one test:
"If this wakes me at 3am, will I be glad it did?" If not, it shouldn't be a page. Route it to a daytime channel or a dashboard instead.
This is the same discipline that beats alert fatigue — and it's the single biggest lever for a humane on-call.
5. Give people runbooks
Being paged at 3am is stressful enough without having to reverse-engineer the system from scratch. A short runbook for each common alert — "what this means, how to check, how to fix or who to call" — turns panic into procedure.
6. Do humane handoffs
When the shift changes, hand over context: what's flaky, what's in progress, what to watch. A 10-minute handoff saves the next person from walking into a mess blind.
7. Stay blameless
When something goes wrong on someone's shift, the response is a blameless postmortem, not a telling-off. People who fear blame hide problems — exactly the opposite of what you want from the person holding the pager.
8. Compensate and recover
On-call is real work, often at antisocial hours. Recognise it — through pay, time off in lieu, or both — and make sure someone who had a brutal night isn't expected to ship features the next morning.
Common mistakes
| Mistake | Do instead |
|---|---|
| One or two people always on-call | Share the load across a fair rotation |
| No escalation — pages vanish into the void | Auto-escalate to a secondary and manager |
| Paging on everything | Page only on actionable, real impact |
| No runbooks | Write a short one per common alert |
| Treating on-call as invisible/free | Recognise and compensate it |
| Blaming whoever was on shift | Run a blameless postmortem |
The bottom line
| Principle | In one line |
|---|---|
| Clarity | One person always knows they're on-call. |
| Escalation | No one is the single point of failure. |
| Signal | Page only for what's worth waking up for. |
| Humanity | Fair rotation, runbooks, recovery, no blame. |
A good on-call rotation is quiet. The pager rarely goes off, and when it does, it means something — and the person holding it has the backup, the runbook, and the trust to handle it. Build that, and on-call stops being the thing people dread and starts being just another part of running a reliable service.
The fastest way to a calmer rotation is fewer, better alerts — start with Alert Fatigue and the alerting guide.