Tran Manh Tuan

Production Troubleshooting

Finding out what is actually wrong when a production failure does not stay inside one layer.

The failures that take longest are rarely the ones with a clear stack trace. They are the ones where the application looks fine, the infrastructure looks fine, and the problem only appears for certain clients, certain regions, or certain request paths — so each team can reasonably conclude it is not theirs.

What that needs is not a guess at the fix but a method: reproduce the failure precisely, isolate which layer it originates in by testing each one directly, and keep narrowing until the actual cause is identified rather than a plausible-sounding one. The fix is usually the short part.

Every engagement ends with something that prevents a silent repeat — a monitor, a smoke test, or a log that would have caught it.

Typical engagements

  • Investigating an incident that crosses application, infrastructure and CDN boundaries
  • Failures that only affect some clients — crawlers, specific regions, specific routes
  • Deciding whether a problem belongs to the application or the platform it runs on
  • Adding the monitoring that would have caught the failure earlier

Related work

Technologies used in this work

  • Apollo Client
  • AWS AppSync
  • AWS Cognito
  • AWS Lambda
  • Docker
  • Elasticsearch
  • GraphQL
  • Next.js
  • PostgreSQL
  • Prisma
  • React
  • Terraform
  • TypeScript

Is this a fit?

  • A good fit: business systems, backends and APIs, GCP and AWS infrastructure, production incidents.
  • Not the right fit: work that is mainly pixel-perfect UI or heavy animation.

If this sounds like the problem you have, Send a project brief.

Your email app opens with a short brief to fill in.