How Workers powers our internal maintenance scheduling pipeline

Cloudflare details how it built an automated maintenance scheduler on Workers to safely coordinate disruptive data center operations, using a graph-based interface for targeted data retrieval, a fetch pipeline with request deduplication, LRU and CDN caching, and backoff retries, plus focused Prometheus/Thanos queries and Parquet-transformed historical metrics in R2 for fast analytics. The approach slashes memory and payload sizes, resolves subrequest bottlenecks, achieves ~99% cache hit rates, and delivers up to 15x P90 performance gains for historical queries—enabling reliable, real-time, large-scale maintenance planning without impacting customer traffic.