Debugging questions.

79 debugging questions from the bank, ready to practice out loud.

Learn the ideas first · free with an account

Debugging as hypothesis testing STAR and CARL: structuring your story Making a write safe to retry

All 79 questions

503s at peak trafficAn API endpoint started returning 503s only at peak traffic in one of three regions. Walk me through how you investigate.Symptom To Cause reasoningMid–senior

Batch job silently producing duplicatesA nightly batch job has been silently producing duplicate rows for the last three nights. Where do you look first?Hypothesis prioritizationMid–senior

Intermittent cart emptying issuesA user reports that their cart 'sometimes' empties between page loads. You can't reproduce it. Walk me through your approach.Reproducing intermittent issuesMid–senior

Latency spike after benign deployA service's p99 latency spiked from 80ms to 600ms after a deploy that 'changed nothing relevant.' What's your investigation path?Tools fluencySenior–staff+

Growing queue depth despite processingA queue worker is processing messages but the queue depth keeps growing. Walk me through what you check.Symptom To Cause reasoningMid–senior

Customer errors with healthy dashboardsYour service is healthy on every dashboard, but customers report errors. How do you bridge the gap?ObservabilitySenior–staff+

Report totals wrong percentageA weekly report has the wrong totals — they're off by exactly the same percentage every week. What do you investigate?Hypothesis prioritizationMid–senior

Test fails intermittently in CIA test in CI fails about 1 in 20 runs and passes when re-run. Walk me through how you'd track it down.Reproducing intermittent issuesMid–senior

Search results in wrong orderA search service is returning correct results, but they're in the wrong order. The ranking team says nothing's changed. What do you check?Symptom To Cause reasoningSenior–staff+

Memory spike in productionA function is consuming 4GB of memory in production but only 200MB in staging on identical inputs. How do you find what's different?Tools fluencySenior–staff+

Login broken with no detailsA bug ticket says 'login is broken' with no other detail. The user can't be reached. What do you do?ObservabilityEntry–mid

Healthy database, slow queriesPerformance metrics show a database is healthy, but every query taking it as a dependency is slow. What's most likely?Hypothesis prioritizationMid–senior

Payment failures by countryCustomers in one country report payment failures, but staging works fine and the same flow works for adjacent countries. Walk me through your investigation.Reproducing intermittent issuesSenior–staff+

Error spikes every six hoursA microservice's error rate spikes for exactly 90 seconds every six hours. Where do you start?Symptom To Cause reasoningSenior–staff+

NPS drops without error spikeA new release shows no error spike but customer NPS drops by 12 points. Where do you look?Tools fluencySenior–leadership

Dashboard shows wrong dataA customer reports that their data is wrong on the dashboard, but the underlying database shows the right values. Walk me through.ObservabilityMid–senior

Connection reset by peerA new error appears in logs with the text 'connection reset by peer' at low but rising frequency. What does that tell you and what do you check next?Hypothesis prioritizationMid–senior

Feature feels slowReports come in that a feature 'feels slow' but no latency metric is showing degradation. Walk me through the possibilities.Symptom To Cause reasoningSenior–staff+

Bug only with simultaneous deploymentsA bug only happens when two specific microservices are deployed at the same time. How do you debug it without rolling back?Reproducing intermittent issuesSenior–staff+

Stream processing job stops after four hoursA long-running stream-processing job stops emitting after 4 hours, every time. The process is still alive. Where do you look?Tools fluencySenior–staff+

Alert never firedA revenue-affecting bug has been live for a week and no alert fired. What's your first move after the bleeding stops?ObservabilitySenior–leadership

CORS error in browser onlyAn API client is failing on what looks like a CORS error in browser, but identical requests from server-side work fine. Walk me through.Hypothesis prioritizationMid–senior

Scheduled task not runningA scheduled task isn't running but the scheduler claims it is. What do you do to confirm and then locate the gap?Symptom To Cause reasoningMid–senior

Workflow fails one customerA workflow fails for exactly one of your enterprise customers and passes for everyone else. Where do you start?Reproducing intermittent issuesMid–senior

Memory leak on hot pathAn app's memory grows steadily over hours until OOM, on a hot path you've never touched. How do you find the leak?Tools fluencySenior–staff+

Inherited service with incomprehensible logsYour team owns a service whose owners just left the company. Production traffic is fine, but you can't make sense of the logs. What's your first week look like?ObservabilitySenior–staff+

Stable service started flakingA service that's been stable for a year started flaking. The last code change was two months ago. What's most likely?Hypothesis prioritizationSenior–staff+

Bug with two feature flagsA bug shows up only when two specific feature flags are both on. How would you confirm and isolate it?Symptom To Cause reasoningMid–senior

Test fails on CI onlyA test passes locally and on your colleague's machine but fails on CI consistently. Where do you look?Reproducing intermittent issuesEntry–mid

Symptoms persist after rollbackA change was rolled forward, broke production, and rolled back — but the symptoms didn't fully disappear. What does that tell you and what do you check?Tools fluencySenior–leadership

Form fails in SafariA form submission works fine in Chrome but fails silently in Safari. The network tab shows a 200 response. Walk me through how you'd debug this.Symptom To Cause reasoningEntry–mid

Production crash works locallyYour app crashes with 'undefined is not a function' in production but works locally. What are the top three things you check and in what order?Hypothesis prioritizationEntry–mid

Unreproducible double-trigger bugA button click sometimes triggers the action twice, but you've clicked it 50 times and can't reproduce it. How do you approach this?Reproducing intermittent issuesEntry–mid

Image not loading on webpageWalk me through how you'd use browser DevTools to figure out why an image isn't loading on a webpage.Tools fluencyEntry–mid

API succeeds but app shows no resultsYour REST API returns data, but the mobile app shows 'No results found.' The API logs show successful 200 responses. What do you check?ObservabilityEntry–mid

CSV export headers scrambled overnightA CSV export feature worked yesterday but today generates files with scrambled column headers. No code was deployed. What's your debugging process?Symptom To Cause reasoningEntry–mid

Prioritizing among three bug causesYou have three possible causes for a bug: a code logic error, bad input data, or a misconfigured environment variable. How do you decide what to check first?Hypothesis prioritizationEntry–mid

Race condition under loadA race condition happens only under load. You don't have access to a staging environment with production-level traffic. How would you reproduce it?Reproducing intermittent issuesEntry–mid

Debugging wrong return valueDescribe how you'd use console.log, breakpoints, and the call stack to debug a function that's returning the wrong value.Tools fluencyEntry–mid

Database timeout with normal metricsError logs show 'database timeout' but the database monitoring dashboard shows CPU and memory are normal. What else should you look at?ObservabilityEntry–mid

LLM feature returns malformed JSONAn LLM-backed feature returns malformed JSON for about 2% of production requests, but you've never seen it fail in testing. How do you track down what's different and harden the feature?Reproducing intermittent issuesMid–senior

Daily job runs wrong daysA daily job ran twice last Sunday and didn't run at all on Monday. The code hasn't changed in months. What's your leading hypothesis and how do you confirm it?Symptom To Cause reasoningMid–senior

Cross-user data leakTwo users report briefly seeing someone else's account data after a release. You can't reproduce it. What classes of bug produce cross-user data leaks, and which do you rule out first?Symptom To Cause reasoningSenior–staff+

Missing time in traceA distributed trace shows a request took 2 seconds end to end, but the spans only add up to 400ms. Where does the missing time hide, and how do you make it visible?Tools fluencySenior–staff+

Latency lag after dependency recoveryA dependency outage ended 30 minutes ago and its error rate is back to zero, but your service's latency is still elevated and recovering slowly. What mechanisms cause this lag, and what do you check?Symptom To Cause reasoningSenior–staff+

Canary passes, full rollout failsA change sailed through canary at 5% of traffic but broke when rolled to 100%. What categories of bug pass a canary and fail at full rollout, and how do you narrow down which one you have?Hypothesis prioritizationSenior–staff+

Test passes, customer bug remainsYour coding assistant proposed a fix that makes the failing test pass, but the customer bug reports keep coming. What discipline keeps you from shipping fixes that cure the test instead of the bug?Hypothesis prioritizationMid–senior

P99 alert fires with no slow requestsA p99 latency alert keeps firing, but when you search the logs you can't find a single request slower than the threshold. How can both be true, and how do you reconcile them?ObservabilitySenior–staff+

Bug after dependency version bumpsA bug appeared right after a deploy whose only diff is dependency version bumps — no application code changed. How do you isolate which dependency, and which change inside it, is responsible?Hypothesis prioritizationMid–senior

Wrong API URL in productionYour feature works in development, but in production the app calls the wrong API URL. How do you find out what configuration the deployed app is actually running with?ObservabilityEntry–mid

User still sees old behaviorYou fixed a bug and deployed, but a user on the call still sees the old behavior. What are the caching layers between your code and their screen, and in what order do you rule them out?Symptom To Cause reasoningEntry–mid

Stack trace mostly frameworkYou hit an exception with a 40-frame stack trace that's mostly framework code. How do you find the line that actually matters, and what do you read first?Tools fluencyEntry–mid

Hundreds of duplicate database queriesA page has gotten slower as the product grew, and the logs show hundreds of nearly identical database queries for a single page load. What's going on, and how do you confirm and fix it?Symptom To Cause reasoningEntry–mid

Synchronized global authentication failureAt 14:00 UTC, every service in every region started failing authentication simultaneously. No deploys happened anywhere. What failure classes produce a synchronized global break, and how do you confirm the culprit fast?Symptom To Cause reasoningSenior–staff+

Cache hit rate dropped overnightA cache hit rate dropped from 92% to 45% overnight with no code changes or traffic pattern shifts. Walk me through your debugging process.Symptom To Cause reasoningMid–senior

Stale data by timezoneUsers in a specific timezone report seeing stale data every morning between 6-7am, but logs show successful cache refreshes. What do you investigate?Hypothesis prioritizationMid–staff+

Error spike by OS versionA feature flag rollout at 10% shows a 3x error rate compared to control, but only for Android users on version 12. How do you narrow down the root cause?Symptom To Cause reasoningSenior–staff+

Service down with normal metricsYour observability dashboard shows CPU at 40%, memory at 60%, no errors logged, but the on-call page says the service is down. What's your first move?ObservabilityMid–senior

Direct call works, gateway failsA microservice returns correct data when called directly via curl, but returns errors when called through the API gateway. Walk me through your debugging steps.Hypothesis prioritizationEntry–mid

Replication lag vs write visibilityDatabase replication lag is under 100ms, but users report seeing their own writes disappear for 5-10 seconds. What could cause this discrepancy?Symptom To Cause reasoningSenior–staff+

Test passes locally, fails parallelA flaky integration test passes locally and in CI 95% of the time, but fails 100% of the time when run in parallel with test suite B. How do you isolate the issue?Reproducing intermittent issuesMid–senior

Spans fast, page loads slowYour distributed tracing shows every span completing successfully in under 50ms, but users report 10-second page loads. What's missing from your observability?ObservabilitySenior–staff+

Scheduled job suddenly timing outA scheduled job ran successfully for 18 months, then suddenly started timing out. The data volume hasn't changed. What non-obvious factors do you check?Hypothesis prioritizationMid–senior

Third-party API returning empty responsesError logs show a third-party API returning 200 OK, but the response body is empty 2% of the time. The vendor insists their service is healthy. How do you prove what's happening?Tools fluencyMid–senior

Memory grows in production onlyMemory usage grows steadily at 50MB/hour in production but stays flat in load testing with 10x the traffic. What could explain this?Symptom To Cause reasoningSenior–staff+

Insert succeeds, GET returns 404A POST endpoint returns 201 Created, the database shows the row inserted, but a GET immediately after returns 404. This happens randomly to 1 in 500 requests. Where do you start?Reproducing intermittent issuesSenior–staff+

False positive alerting incidentsYour alerting system fired 47 PagerDuty incidents last night for a service outage, but the service was actually healthy. How do you debug the monitoring itself?ObservabilityMid–staff+

Query slow with sixth fieldA GraphQL query returns in 80ms when requesting 5 fields but takes 12 seconds when requesting 6 fields. The 6th field is a simple string. What do you investigate?Symptom To Cause reasoningMid–senior

Pods running different app versionsAfter a Kubernetes deployment, 3 out of 20 pods are serving traffic but showing different application versions in logs. How do you identify what went wrong?Tools fluencySenior–staff+

Reconciliation off by consistent amountA financial reconciliation job reports totals that differ from the source system by $0.03 every single day. Manual spot checks show individual transactions are correct. What's your approach?Hypothesis prioritizationSenior–staff+

WebSocket drops after exact timeoutWebSocket connections drop after exactly 60 seconds in production but stay open indefinitely in staging. Both environments use identical application code. What do you check?Hypothesis prioritizationMid–senior

Data corruption nowhere to findA user reports data corruption but you can't find the corrupted record in the database, backups, or audit logs. Walk me through how you verify their claim and investigate.Symptom To Cause reasoningMid–senior

High cache hit rate, high origin trafficYour CDN reports 99.9% cache hit rate, but origin server traffic hasn't decreased after enabling caching. What could explain this discrepancy?Symptom To Cause reasoningMid–senior

Race condition during business hoursA race condition occurs only when deploying during business hours, never during off-hours deployments. The deployment process is identical. How do you reproduce and fix it?Reproducing intermittent issuesSenior–staff+

Events ingested but not indexedYour logging pipeline shows 1M events/minute ingested but only 400K/minute indexed in your search backend. There are no errors. What's happening to the missing events?ObservabilitySenior–staff+

Same request different response timesAn API consistently returns in 45ms for user A and 850ms for user B requesting identical data. Both are hitting the same server instance. What do you investigate?Hypothesis prioritizationMid–senior

Traffic routes to old environmentA blue-green deployment switches traffic successfully, but 0.1% of requests still route to the old environment for 6 hours afterward. How do you track down where they're coming from?Tools fluencySenior–staff+

Bimodal response time distributionMetrics show average response time is 120ms, but the histogram reveals a strange bimodal distribution with peaks at 50ms and 900ms. What patterns would you look for?ObservabilitySenior–staff+

App works WiFi, fails cellularA mobile app works perfectly on WiFi but fails to load data on cellular networks. The API logs show successful 200 responses for all requests. How do you debug this?Symptom To Cause reasoningMid–senior