Monitoring in a Serverless World
summary
Explains why Prometheus’ pull model fails for short-lived, non-networked serverless functions and describes building an open source gravel-gateway (Rust) that accepts pushes, performs aggregation, and exposes Prometheus-compatible metrics. Implements a clear_mode label to express aggregation semantics (sum, family/info, etc.), uses circular weighted buckets to keep bounded-memory statistical aggregates (mean, median, p95) across many ephemeral observations, and interoperates with existing Prometheus clients and Cloudflare Workers. Code is public and the gateway strips internal labels before Prometheus ingestion.
AI-written summary · about
chapters (6)
- 0:00Introduction to serverless monitoring
- 2:09Prometheus limitations for serverless
- 7:04Engineering teams' ad-hoc solutions
- 13:52Building the gravel Gateway
- 21:13Future of serverless observability
- 23:20Q&A
related talks
30mOperating Prometheus in a Serverless World
34mFleeting Metrics: Monitoring Short-lived or Serverless Jobs with Prometheus
32mDemocratizing Application Monitoring with Prometheus
34mMonitor My Socks: Using Prometheus in a Polyglot Open Source Microservices
21mMonitoring Serverless Workloads with OpenTelemetry and Prometheus
13mMonitoring Kubernetes VS Serverless-based applications
03mUsing OpenTelemetry and Prometheus to monitor serverless
24mObservability in the MLOps Lifecycle with Prometheus