Removing bottlenecks before a live product launch
How a focused database and delivery review reduced technical risk ahead of a scheduled live-stream event.
- Client
- Tilt
- Date
- Service
- Focused delivery
What changed
- Used load tests to locate the database bottlenecks.
- Moved the production database to a managed PostgreSQL service.
- Reduced avoidable Docker rebuild work through multi-stage builds and layer caching.
Context
Tilt was preparing a live-commerce product for a scheduled seller event. Load tests showed that the database path needed attention before the team could rely on the planned traffic profile.
My remit
I was brought in for a focused piece of diagnosis and delivery across the database, infrastructure and deployment path, with the launch constraint setting the order of work.
Decisions and delivery
- Built a repeatable load-test setup and traced the queries creating the most pressure.
- Moved the PostgreSQL workload to AWS RDS Aurora and described the infrastructure with Terraform.
- Reviewed indexes, expensive queries, connection handling and caching against the measured workload.
- Reordered Docker build inputs and used multi-stage builds so dependency layers were reused when their inputs had not changed.
- Re-ran the load tests against the changed path before handing the system back.
Outcome
The team had a tested managed-database path and a less wasteful build process ahead of the event. No percentage improvement, traffic number or event-success claim is published without the missing baseline evidence.