SLOs and Error Budgets: Follow-Up Questions Interviewers Ask
When you drop SLOs and error budgets into a design interview, the recruiter has already decided you know the basics. The next step is to see whether you can turn those concepts into day-to-day operational decisions. Interviewers probe because SLOs sit at the intersection of product, reliability, and cost – a place where vague statements quickly turn into risky assumptions.
The follow-up questions are designed to surface three things: how you pick measurable targets, how you allocate the budget across teams, and how you act when the budget is exhausted. Below is a typical progression of probes, from the easiest to the most strategic. Answer each as if you were speaking to the interviewer, using concrete numbers or a short example to ground your reasoning.
How do you choose an appropriate SLO target for a new service?
I start by looking at the user-impact of the request. If a read operation is user-facing, I aim for a latency SLO of 99.9% of requests completing under 200 ms. I gather historical latency data from a prototype or a similar service, then add a safety margin of roughly 20% to account for traffic spikes. If the service is internal, such as a batch job, I might use availability (e.g., 99.5% of jobs finishing within the SLA window) because latency is less visible to end users.
What error budget does that SLO give you?
For a 99.9% latency SLO, the error budget is the 0.1% of requests that can exceed the latency threshold. If the service handles 1 million requests per day, that translates to 1,000 “slow” requests per day, or about 0.012 per second (1 000 ÷ 86 400). I track this budget in real time using a monitoring dashboard that aggregates latency percentiles.
How do you monitor whether you are staying within the error budget?
I instrument the service to emit latency histograms to a metrics system like CloudWatch or Prometheus. A rolling 24-hour window calculates the percentage of requests above the threshold. I set up an alert that fires when the budget consumption exceeds 80% of the daily allowance, giving the team time to investigate before the budget is fully spent.
What actions do you take when the error budget is exhausted?
When the budget is exhausted, I shift the team’s focus from feature development to reliability. First, I trigger a post-mortem to identify the root cause. Then I prioritize remediation work: scaling out the bottleneck tier, adding caching, or tightening circuit-breaker thresholds. I also communicate the status to product managers so they understand why new features are on hold until reliability is restored.
How do you decide which reliability improvements to prioritize?
I rank improvements by their impact on the error budget and their effort. A quick win, like increasing the read replica count, might reduce latency outliers by 30% with a small cost. A larger project, such as redesigning the data model, could cut tail latency by 70% but takes months. I use a simple ROI matrix: (budget saved per day) × (days until deadline) versus implementation effort, and pick the highest net gain that fits the current sprint.
How do you allocate error budget across multiple dependent services?
I treat the end-to-end request as the ultimate SLO. If Service A calls Service B, the budget for Service A includes the latency contributed by Service B. I calculate a split based on the proportion of total latency each service historically adds. For example, if Service A contributes 60 ms and Service B 40 ms on average, I give Service A a 60% share of the budget and Service B the remaining 40%. This shared ownership forces both teams to keep their part of the latency within limits.
What happens if two services compete for the same error budget?
When two services share a budget, I introduce a coordination layer—often a reliability champion or a runtime throttling mechanism. The champion reviews each service’s consumption and can re-allocate budget dynamically, for instance by temporarily lowering Service B’s latency target during a traffic surge. If the conflict persists, I negotiate a re-balancing of the SLOs: perhaps relaxing Service A’s latency target to give Service B more breathing room, always keeping the overall user experience in mind.
How do you handle SLOs for features that are still in development?
For unfinished features, I set a provisional SLO that is intentionally loose, such as 99.5% of requests under 500 ms. This gives the team room to iterate without immediately hitting the error budget. As the feature stabilizes, I tighten the SLO incrementally—say, moving to 99.9% under 200 ms—while monitoring the budget consumption. The gradual tightening signals to stakeholders that reliability is improving and that the feature is ready for production scale.
How does the error-budget concept change at a staff-level interview?
At the staff level, interviewers expect you to discuss organization-wide policies, not just a single service. I would describe a governance model where SLO owners report weekly budget consumption to a reliability steering committee. The committee decides on cross-team budget reallocations, invests in platform-level tooling (e.g., automated SLO dashboards), and defines escalation paths that span multiple services. I also talk about long-term strategies like chaos engineering to deliberately consume budget and test the system’s resilience.
How to practice this
- Write an SLO policy for a simple service you know (e.g., a user-profile API). State the latency target, calculate the daily error budget, and note the monitoring alerts.
- Simulate budget consumption by generating synthetic traffic and plotting the budget curve. Identify the point where you would switch to reliability work.
- Review the worked design in the System Design AI library, then draw your own version on the canvas, focusing on the SLO definition, budget tracking, and escalation flow.
Questions people ask
What is the difference between an SLA and an SLO?
An SLA is a formal contract with a customer that defines penalties for missed targets. An SLO is an internal reliability goal used to measure and improve service quality. SLAs often reference SLOs but add legal language and compensation terms.
Can you have multiple SLOs for the same service?
Yes. A service can have latency, availability, and error-rate SLOs, each with its own error budget. You track consumption separately and prioritize the metric that most affects user experience.
What if the error budget is never consumed?
If the budget remains largely unused, the SLO may be too loose. Tighten the target gradually to extract more reliability value, but monitor closely to avoid unexpected outages.
How often should SLOs be revisited?
Review SLOs at least quarterly, or after any major incident or product change, to keep them aligned with user expectations and system capabilities.