- architecture
- redis
- dynamodb
- caching
- performance
- aws
- engineering
•
•
•
•
•
•
-
Scaling Our Microservice Made Latency Worse - Here's How We Fixed It
We prescale for EC2 headroom and regional failover during business-critical times. Low traffic left DynamoDB connections idle and the HTTP client we'd chosen for fast pod startup closed them in ~5s. Here's how we fixed p99 with configurable idle timeout and TCP keepalive.
-
Stop Writing Data You Never Read
At 2,600 TPS, we were executing 85M writes daily for data we rarely read. Here's how understanding our access pattern reduced writes by 75% using caching.