High-throughput telemetry ingestion at 1,000+ RPS.
A decoupled, cloud-native streaming pipeline that ingests 1,000+ RPS of client telemetry without turning the database into a bottleneck.
Naive high-concurrency ingestion causes DB write-storms, connection starvation, and runaway query latency during live-user spikes.
The database stopped being the bottleneck — throughput scales with gateway pods, and tail latency stays flat under 1,000+ RPS.
Split the write path: async FastAPI gateway → thread-safe TCP socket queue → stateful compute worker with 10-second fixed tumbling windows.
Compressed 6,800+ micro-transactions into batched ON CONFLICT DO UPDATE upserts, cutting index lock contention and query latency to near-zero.
Architected for horizontal scale on GKE + Cloud Pub/Sub, so gateway and worker scale independently.
Validated with Locust distributed load testing to triage bottlenecks, analyze backpressure, and confirm fault tolerance.