Platform & InfrastructureYear 3 · Q11 (Previous SaaS)

Engineered an automated performance benchmarking harness and chaos testing framework using k6 and Chaos Mesh to enforce SLA budgets and validate microservice fault tolerance before production deployment.

Automated Catching of 95%+ Performance Regressions in CI
100+ Chaos Fault Scenarios Automated
Zero Outages from Unhandled Cascading Failures
k6Chaos MeshGoPrometheusGrafanaKubernetes

Situation

Performance testing was previously treated as an infrequent, manual pre-release activity. As a result, memory leaks, latency regressions, and fragile circuit-breaker configurations slipped into production, triggering cascading outages during peak load periods.

Task

Architect an automated performance and resilience testing platform that enforces p95/p99 latency budgets in continuous integration pipelines and continuously injects network and infrastructure faults to verify cluster resilience.

Action

  • Automated Load Gates in CI: Built a lightweight k6 load testing execution runner integrated into GitHub Actions, evaluating custom latency SLOs ($p_{95} < 150\text{ms}$) against target microservices and failing pull requests that violated performance budgets.
  • Chaos Mesh Kubernetes Fault Injection: Deployed Chaos Mesh custom resources to automatically simulate network partitions, packet loss (20–50%), and random pod evictions in staging environments while active synthetic load runs executed.
  • Prometheus Alert-Driven Assertion: Wrote custom validation runners in Go that queried Prometheus metrics during chaos runs to confirm that Envoy circuit breakers tripped as expected and downstream retries gracefully shed load.

Result

  • Shifted performance testing left, catching over 95% of latency regressions in pull requests before staging deployment.
  • Verified microservice resilience across 100+ automated chaos experiments, completely eliminating cascading failures in production.