Site Reliability Objectives
Project workbook · Path 86

Define a request-based SLI and SLO, calculate an error budget and write an incident-review rule for a synthetic service.

The case

A fictional library API serves 100,000 eligible requests in a 30-day window. Its objective is 99.9% successful requests. The sample contains 80 failed requests. Success is defined by a documented response rule; excluded health checks are not in the denominator.

MeasureValue
Eligible requests100000
SLO percent99.9
Failed requests80
Window days30
STAGE 1

Choose a user-facing measure

An SLI needs a population, event and measurement rule. Choose something related to the user experience rather than a convenient internal count. Define eligible requests and success consistently. The case uses a request-based ratio, so translating it directly into minutes of downtime would change the measurement model.

Your task

Write the SLI definition and exclusions.

Required output

SLI specification

Record your result and evidence here or in a separate file.

Review gate

The denominator is eligible requests, not every log entry.

If the gate is not met

Explain every exclusion and keep it stable across the window.

STAGE 2

Calculate the objective and budget

The allowed bad-event fraction is one minus the objective. Multiply it by eligible requests to obtain the budget for this sample window. The actual success rate and budget consumption answer related but different questions. Keep the units visible and avoid rounding before the final calculation.

Your task

Calculate actual success, allowed failures and remaining budget.

Required output

Budget sheet

Record your result and evidence here or in a separate file.

Review gate

Success is 99.92%; budget is 100; 80 are consumed and 20 remain.

If the gate is not met

Use 0.001, not 0.1, as the allowed failure fraction.

STAGE 3

Define a review response

An error-budget policy connects reliability evidence with a decision process. It should describe who reviews risk, what work may be prioritized and how exceptions are recorded. A policy is not an automatic ban on every change. This fictional exercise does not authorize operations on any real service.

Your task

Draft a review rule for budget exhaustion and a documented exception route.

Required output

Budget policy

Record your result and evidence here or in a separate file.

Review gate

The policy identifies an owner and evidence, not just a red status.

If the gate is not met

Replace vague be more careful instructions with a specific review action.

STAGE 4

Avoid misleading alerts

An alert needs enough traffic and a useful time window. A single failure in a tiny sample can create a dramatic rate without justifying the same response as a sustained incident. This path does not derive a full burn-rate alert system; it asks for a proportionate evaluation plan and an explicit data-sufficiency rule.

Your task

Compare 1 failure in 10 requests with 1,000 in 10,000 and describe the context needed.

Required output

Alert design note

Record your result and evidence here or in a separate file.

Review gate

The note distinguishes equal 10% rates from very different event volumes.

If the gate is not met

Show counts beside rates and state the response context.

STAGE 5

Write the reliability review

A reliability review combines the measured objective, budget state, incidents and improvement actions. Avoid claiming the service is reliable in every dimension because one ratio passes. Latency, data correctness and availability may need separate measures. The supplied case contains no live monitoring evidence.

Your task

Write a one-page monthly review and two next measurement questions.

Required output

Reliability report

Record your result and evidence here or in a separate file.

Review gate

The report limits its conclusion to the supplied request-success measure.

If the gate is not met

Name at least one user-experience dimension not measured.

Final project

Deliver the SLI contract, budget calculation, policy, alert note and review. Recalculate with 120 failures and explain the policy response without inventing a live incident.

Critical condition

Request ratios are not silently converted to downtime, and the denominator remains explicit.

CriterionPointsEvidence
Evidence and correctness30Calculations, source use and results match the supplied case; limitations are honest.
Required artifacts30All five stage outputs and the capstone handoff are present and inspectable.
Reasoning and decisions20Choices, assumptions, alternatives and the case-specific critical condition are explained.
Reproducibility10A reviewer can follow the stated steps and distinguish expected from observed results.
Clear communication10Labels, units, structure and evidence references make the work easy to review.

Score at least 80/100, supply every required artifact, and satisfy the path-specific critical condition. Rework missing evidence before marking complete.

Worked reference

Allowed failures are 100,000 × (1 − 0.999) = 100. Actual success is 99,920 / 100,000 = 99.92%. The case consumes 80% of its error budget and leaves 20 failed requests of allowance for the same fixed window and population.

Boundaries

This is an SLO design and calculation exercise, not a monitored production service or contractual availability guarantee.