# A Beginner's Guide to SRE: reliability interview worksheet

Lesson: https://everythingsre.com/concepts/sre-interview-capstone
Updated: 2026-10-10

TicketDesk is a fictional ticket-booking service. Everything here is original practice material, not a question reported from an employer. Work through it at your own pace, and use the 14 lessons to look up any term you are unsure about. Writing your answers down is the point: the act of explaining is what shows you the gaps.

## Scenario

Customers browse events, reserve tickets, and receive a confirmation. Some try again when their confirmation is delayed. The optional "suggested events" feature can be switched off. Confirmed reservations must never be silently lost or duplicated.

- Forecast peak: 900 valid checkout requests per second. A request is one attempt asking the software to perform a booking.
- Tested capacity: 200 requests per second per serving instance. An instance is one running copy of the software. The test meets the chosen response-time and correctness limits.
- Failure assumption: one serving instance may be lost. Several copies failing together need separate analysis.
- Proposed success SLO (service level objective): 99.9% over the latest 30 days, recalculated as the window moves.
- Measured window: 30 million valid requests and 18,000 failures.
- Change: a new way to send confirmations is ready to launch.

## 1. Define the customer promise

What would you need to clarify about what counts as a completed booking and its confirmation?

Write your success SLI (the measurement), the rules for which attempts count, where you measure, and the SLO (the target):

Write a target for timely completion, and the rules that must always stay true for a booking to be correct:

## 2. Calculate failure allowance and capacity

How many failures does the SLO allow in this window? What fraction of that budget is used? How much remains?

How many running copies do you need to handle peak demand after losing one?

Which of your assumptions could turn out to be wrong? Could another part of the service limit how many bookings complete?

## 3. Decide whether to expand the new version

The canary (the first small group receiving the change) has 200 failures out of 20,000 valid requests. The control (the unchanged group) has 180 failures out of 180,000.

Confirmations are waiting longer than before. The installation tool reports success.

Should more customers receive the change? Explain your evidence, your stop conditions, who owns the decision, and how you would recover:

Could booking records written by the new version make returning to the old software unsafe?

## 4. Respond to an incident and overload

Confirmation work arrives at 1,000 items per second and completes at 800. Customers are retrying. Several teams propose moving requests elsewhere.

How quickly does the waiting list grow?

What can reduce immediate harm, limit the extra work, and protect essential bookings?

Who coordinates the response, who makes repairs, and who gives updates?

Write an update that states the known impact, the current action, what is still uncertain, and when the next update will arrive:

What must you check before you can say the service has recovered?

## 5. Account for dependencies and data recovery

Two independent services are both required for a booking. Each is 99.9% available. Calculate their joint availability and explain when this simplified model would not apply:

Choose an RPO (the acceptable lost-data interval) and an RTO (the acceptable recovery duration) for the important records:

How would you restore the saved records, check them, reconcile external payments, and reopen safely?

## 6. Plan improvements

Write two follow-up actions. Give each an owner, a priority, and a test that demonstrates the intended result:

Which comes first, and why?

## Self-review

Mark each area as missing, mentioned, or explained with reasoning.

- Customer promise and measurable targets:
- Calculations and assumptions:
- Failure cases and correct records:
- Measurements and response responsibilities:
- Limited changes, immediate harm reduction, and verified recovery:
- Specific improvements and acknowledged limits:

Now explain your answer out loud, as if to an interviewer. Wherever you cannot explain why you made a decision, go back to that lesson. If several areas are still "missing", that is normal on a first pass; the worksheet is for self-review, not a hiring score.