Aug 03, 2024 · 1 min read
Scaling Microservices on AWS with Event-Driven Concurrency
Design notes from rewiring AffCollect into a message-driven mesh that processed affiliate statements 10x faster.
AffCollect ingests thousands of affiliate dashboards every hour. We shifted the pipeline from cron-based polling to an event-driven graph built on Kafka + AWS Lambda.
Ingredients
- Orchestrator topics manage fetch → normalize → enrich jobs.
- Lambda shards keep third-party API quotas respected via token buckets.
- DynamoDB stream fans out incremental updates to downstream billing services.
Outcomes
- 10x faster ingestion with concurrency controls.
- 99%+ accuracy because retries are isolated per merchant stream.
- Observability by default using OpenTelemetry into Honeycomb.
This playbook now powers every data workload I touch.