Skip to content
Tech Interview Prep
Technical interview guide

SLIs, SLOs & Error Budgets

The vocabulary reliability is measured in, and how an error budget turns 'be reliable' into a concrete number.

Read
30 min
Practice MCQs
25
Interview Q&A
25
Edition
v4
Editorial status
Reviewed

Scope: Google SRE guidance, Prometheus, and OpenSLO current 2026-08-31.

Curated: · Written: · Reviewed:

25 questions

MCQ-4A service's SLO is 99.9% good events, measured over eligible requests. The error-budget policy must state the fraction of eligible events that may fail. Which fraction is correct?
MCQ-25A team declares its checkout availability SLO as 99.9% of requests over a rolling 28-day window, measured at the load balancer, counting only requests to the /checkout path. The dashboard is green. Which claim does that green state establish?
MCQ-1Over a 28-day window a checkout API received 200,000 user requests: 199,300 returned 2xx, 100 returned 5xx, and 600 were rejected by the load balancer with 503 before reaching the API. The team's SLO is "99.9% of user requests to the checkout API succeed." What is the SLI value for this window?
MCQ-2A payments API carries an SLO of 99.95% successful requests over a rolling 28-day window, computed from the load balancer logs as successful user requests divided by user requests. The spec counts a request as failed only when the response is a 5xx or the request times out; 429 throttle responses count as successful because the client is told to retry. Over the last 28 days the load balancer logged 10,000,000 requests: 5,000,000 were Kubernetes liveness and readiness probes returning 200, and 5,000,000 were user requests. Of the user requests, 10,000 returned 5xx, 5,000 timed out and were logged as 504, and 5,000 were rejected with 429. What success rate should the SLO report for the window?
MCQ-8An availability SLO of 99.9% is measured over a rolling 30-day window by a synthetic probe that samples every 30 seconds. The team's documented policy counts missing probe data as bad time. During one window the probe's own infrastructure fails for 6 hours and records nothing; there are no other failures. How much error budget remains for that window?
MCQ-9A checkout service defines its availability SLI as good requests divided by valid requests. Last week it recorded 1,000,000 valid requests and 20,000 failures, an SLI of 98.0%. A team proposes excluding all 200,000 requests from client version 2.3.1, arguing the version is deprecated and emits malformed telemetry; those 200,000 requests contain 15,000 failures. What does the exclusion do to the reported SLI, and what should the team conclude?
MCQ-6A service's SLO is 99.5% of requests completing within 300 ms over a 5-minute window. In the latest window, endpoint A handled 18,000 requests with 99.0% within 300 ms, and endpoint B handled 2,000 requests with 80.0% within 300 ms; the mean latency across both was 210 ms. What is the SLI for this window, and does it meet the objective?
MCQ-7A mobile app calls an API through a CDN. Over one day the CDN handled 2,000,000 requests. It returned 502 for 0.3% of them without forwarding them to the origin. The app retries a 502 once, and half of those retries succeed. The origin received the other 1,994,000 requests and reported 99.95% success on them. Separately, 0.2% of the CDN's requests were aborted by the app after a 10-second timeout; the origin logged a 200 for each of those at around 12 seconds. Counting only the 2,000,000 original requests, what fraction did users see fail?
MCQ-5A service publishes a 99.9% availability SLO over a rolling 30-day window, and its SLI is good requests divided by total requests. A new engineer converts the target to '43 minutes of allowed downtime per month' for the dashboard. Which statement explains why that figure misrepresents the budget?
MCQ-16A payments API takes 12 million requests per 30-day window. Its availability SLO is 99.9%, and deploys and dependency blips caused an average of 28 minutes of unavailability per month last quarter. A director proposes raising the SLO to 100% for the next quarter. What does that target require of the team?
MCQ-10A service carries a 99.9% availability SLO over a 30-day window. In the last hour it served 100,000 requests, 500 of which failed. The on-call engineer needs the burn rate to judge how fast the error budget is being spent. Which figure is it?
MCQ-11A service has a 99.9% availability SLO over a 30-day window and serves 7.2 million requests per window, so its error budget is 7,200 failed requests. In the last hour it recorded 10 failed requests. What does that hour imply for the rest of the window?
MCQ-21A service has a 99.9% success SLO over a 30-day rolling window and handles 1,000,000 requests per day. Twelve days ago a bad deploy caused 40,000 failed requests in that one day; every other day in the window has 100 failures. Assuming no further failures, when will the service next be compliant with the SLO?
MCQ-3A service has an internal SLO of 99.95% availability over a 30-day window. Its customer SLA promises 99.9% monthly availability and service credits if missed. Last month the service measured 99.92% availability. What is the correct interpretation?
MCQ-18A service carries a 30-day availability SLO of 99.9%, and its error-budget policy states two tiers: at 50% of the budget consumed, feature releases require additional review; at 100%, only reliability work ships until the budget recovers. On day 12 of the window, the service has accumulated 38 minutes of unavailability. What does the policy require?
MCQ-19A service has a 30-day rolling availability SLO of 99.9%, giving a 43.2-minute error budget. Its policy freezes feature releases when the trailing 7-day unavailability exceeds 25% of the monthly error budget. The service recorded 14 minutes of unavailability in the last 7 days and 18 minutes over the 20 days of the current window. A feature release is scheduled tomorrow. What does the policy require?
MCQ-24A 30-day SLO has a multi-window, multi-burn-rate alert with a 1h long window, a 5m short window, and a burn-rate threshold of 14.4. The long window is at 20 and the short window is at 2. What does the alert do?
MCQ-14A service holds a 99.9% success SLO over 30 days and receives a steady 120 requests per hour. Its alert pages when the 5-minute burn rate exceeds 14.4. At 03:12 one request in the 5-minute window fails; the window holds 10 requests, and the page fires. Which change best stops this page without weakening the SLO or slowing detection of a real fast burn?
MCQ-13A checkout service has a 99.9% availability SLO over 30 days, which is 43.2 minutes of error budget. At 02:40 on a Tuesday two alerts fire. Alert A: 1-hour burn rate 14.4. 2% of the 30-day budget consumed in the last hour, error rate up from 0.05% to 1.44% since 02:15 and still climbing. Alert B: 3-day burn rate 1.0. 10% of the budget consumed over the last 3 days, error rate flat at 0.10% with no change in the last hour. The on-call engineer is asleep. Which routing is correct?
MCQ-12A service has a 99.9% availability SLO over 30 days, so the monthly error budget is 0.1% of requests. Its only alert is a single 5-minute window that pages when the error ratio exceeds 2%. Last month it fired 38 times, every one a 1–3 minute spike that self-resolved and burned under 0.5% of the budget. Separately, the error ratio sat at 0.6% for 9 hours straight: a 6x burn rate that consumed 7.5% of the monthly budget, and never paged. Which change to the alerting configuration addresses both failures?
MCQ-22A service has a 99.9% availability SLO evaluated over calendar months. At 10M requests per month the error budget is 10,000 failed requests, and traffic is steady at ~13,900 requests per hour. A bad deploy at 23:30 on January 31 fails 100% of requests until it is rolled back at 00:30 on February 1. The month-to-date budget gauge never falls below 30% remaining in either month, and no alert fires. Which explanation accounts for the missing alert?
MCQ-17A search API's 28-day metrics are mean latency 140 ms, p99 latency 1,850 ms, and availability 99.95%. The team proposes adopting 'mean latency ≤ 140 ms' and 'availability ≥ 99.95%' as the SLOs, since both are already met. The product requirement, from abandonment data, is that 99% of searches complete within 1 s. Which latency SLO should be adopted?
MCQ-15A service publishes a 99.9% availability SLO over 30 days. Last month it served 100M requests: 99M in us-east-1 with 49,500 failures, and 1M in ap-southeast-2 with 30,000 failures. A normal month in ap-southeast-2 is about 1,200 failures. The global SLO read 99.92% and stayed green. The regional team wants an objective that this incident breaches but a normal month does not. Which one should it add?
MCQ-20A checkout request fans out in parallel to three dependencies: A and B each have a 99.9% availability SLO, and C has a 99.95% availability SLO. A and B share a database cluster; when that cluster failed in the last two incidents, A and B failed together. C has a fallback that returns cached data and marks the order 'pending review'. The journey owner has ruled that pending-review orders are not successful checkouts. What end-to-end availability can the team defensibly claim from these SLOs and this failure model?
MCQ-23Checkout's SLO is 'availability ≥ 99.9% over a rolling 28-day window', measured by a query that counts every HTTP 5xx the service returns. Over the current window the service served 10,000,000 requests and the query counted 15,000 bad events, 6,000 of them 503s returned during the payment provider's maintenance window on day 4. On day 10 an engineer edits the SLO in place: the query now excludes payment-provider 503s, and the target and window are unchanged. On day 28 the dashboard reads 99.91% for the window. The day-4 incident review asks whether that outage stayed inside the error budget. What should the review conclude?