First, the key terms.

SLI
A measured indicator of service behavior.
SLO
A target for that indicator over a defined window.
Error budget
The amount of failure allowed by the chosen objective.

The idea, explained

Availability and latency are different dimensions. A service may answer almost every request while responding too slowly for users. Define what counts as success, which requests are included and the measurement period. Percentiles can show slow responses that an average hides. An error budget describes the permitted gap from perfect performance under the objective.

STEP 1

Define success concretely

For opening a lesson, success might require correct content within a useful latency, not just any HTTP response. Specify which requests count and how they are measured. A misleading denominator makes the objective hard to interpret.

STEP 2

Use the budget to guide trade-offs

An SLO acknowledges that perfect reliability is generally unattainable. Consuming the budget may justify slowing releases or addressing recurring failures. It does not mean every allowed failure is harmless to the affected user.

Follow a worked example.

An illustrative monthly objective is 99.9% successful eligible requests. That allows a 0.1% failure fraction in the measured request population. It does not by itself state the time allowance for an outage, because requests may arrive unevenly.

Walk it through, one step at a time

  1. For a simplified window with 100,000 eligible requests, a 99.9 percent success target permits 100 failures.
  2. If 80 have already failed, 20 remain within that request-based budget.
  3. A single outage can affect many requests, and a rare critical action may need its own objective. Do not confuse request-based percentages with a guaranteed outage duration.

Where people get stuck

A target without a measurement definition can look precise while remaining impossible to assess.

Now try it yourself.

Pause before opening each answer. Explain your reasoning in your own words.

1. Check the core idea

Why is average response time insufficient on its own?

Reveal the answer & reasoning

A good average can hide a group of users receiving very slow responses. Tail measures reveal that experience.

2. Apply what you learned

Why is “the server was up all day” insufficient?

Reveal the answer & reasoning

Users may still receive errors, stale content or failed actions. Measure the user-visible operation and its important dependencies.