Back
2024CompletedFeatured
Atlas Ingest
A back-pressure aware ingestion service that survives its own bad days.
RoleDesign and implementation
- Go
- Kafka
- ClickHouse
- Kubernetes
The problem
The original pipeline dropped events silently under load. Nobody found out until a customer asked why a dashboard disagreed with their own numbers.
The approach
Atlas Ingest makes back-pressure a first-class concept. Producers get an explicit signal instead of a timeout, and every batch carries an idempotency key so a retry is always safe.
Result
180M events/day at a steady p99 of 40ms, with zero silent drops in eighteen months of operation.