Serverless Prometheus
summary
Architecture implements Prometheus queries using thousands of Lambda functions that each perform at most two S3 GETs and scan small portions of object files, with storage moved to S3 so compute and storage scale independently. Default Lambda result buffer hits a 6 MB payload limit, streaming raises to 20 MB but still transfers after 6 MB, so payloads are compressed to stay under 6 MB or large blobs are left in S3. Result: 2× query speedup and P99 latency around 3 seconds.
AI-written summary · about
related talks
54mServerless beyond Functions
49mUsing serverless functions for real-time observability
32mUsing Serverless Functions for Real-time Observability
39mServerless in production, an experience report
41mUsing Serverless & ARM64 for Real-Time Observability
47mServerless in production, an experience report
01h 00mServerless Architecture - Tales from a world without servers
44mServerless Architecture on Production: Good, Bad and Ugly Parts