Performance Engineer Interview Prep Guide

The fact that p50 is unchanged but p99 spiked suggests a tail latency issue affecting a subset of requests. Investigate code paths that differ for edge cases, check for lock contention under specific conditions, look for garbage collection pauses, examine database query plans for queries that fan out differently based on data, and check if certain user cohorts hit a different code path. Use distributed tracing to find the specific service and operation causing the tail latency.

Cover workload modeling (realistic user journeys, not just endpoint hammering), ramp-up strategy, identifying the target requests per second, test environment requirements (production-like or production with traffic mirroring), key metrics to monitor during the test, acceptance criteria, and how to test graceful degradation. Discuss testing individual services in isolation versus full system load tests, and how you handle test data at scale.

CPU profiling identifies hot functions consuming processing time (use when CPU utilization is high). Memory profiling tracks allocations, heap usage, and garbage collection (use for memory leaks or excessive GC pauses). Distributed tracing follows requests across services to identify latency bottlenecks (use for end-to-end latency issues). Discuss specific tools: async-profiler or perf for CPU, heap dumps and allocation tracking for memory, and Jaeger or Zipkin for distributed tracing.

Quantify the before and after metrics: latency reduction, throughput improvement, resource cost savings, and user experience impact. Walk through your analysis methodology, how you identified the bottleneck among many possibilities, the optimization you chose, how you validated it did not introduce regressions, and how you measured the user experience improvement through core web vitals or business metrics.

Define budgets for JavaScript bundle size, total page weight, time to interactive, largest contentful paint, and cumulative layout shift. Integrate budget checks into CI/CD pipelines to fail builds that exceed thresholds. Use Lighthouse CI or custom tooling for measurement. Discuss how you set initial budgets based on competitor analysis and user device profiles, and how you handle exceptions and budget adjustments as features grow.

Start by profiling the current bottleneck: is it CPU-bound, memory-bound, I/O-bound, or network-bound? For CPU: optimize hot paths, reduce serialization overhead, use more efficient algorithms. For I/O: add caching layers, optimize database queries, use connection pooling. For network: reduce payload sizes, use compression, implement batching. Discuss when to optimize code versus when to scale horizontally, and how to validate improvements with controlled load tests.

Collect historical resource utilization data correlated with traffic metrics. Build regression models that predict resource needs based on traffic projections. Account for headroom margins for traffic spikes, seasonal patterns, and new feature launches. Discuss how you validate your models against actual growth and adjust them. Mention lead time considerations for infrastructure procurement and the cost implications of over-provisioning versus under-provisioning.

Walk through the testing scenario, what failure mode was discovered (connection pool exhaustion, memory leak under load, cascade failure), the severity assessment, the fix implemented, and how you incorporated the scenario into ongoing regression testing. Show that your testing methodology was realistic enough to catch issues that unit and integration tests would miss.