Signal / 018 — Cloudflare brings flamegraphs into live Worker debugging

Cloudflare has added on-demand CPU and memory profiling for active Workers and Durable Objects. From Workers Observability, developers can request a profile, inspect it as an interactive flamegraph, and download the profile for deeper analysis. The feature was announced October 9.

Logs show symptoms; profiles show where time goes

A flamegraph turns a sampled execution profile into stacked function calls. Wider boxes represent functions consuming more CPU time or allocating more memory, giving developers a direct route from “this Worker feels expensive” to the code path doing the work. Cloudflare says profiles can be captured from the dashboard or through its cf command-line package, with a chosen Worker version, duration, and profile type.

This is production profiling, so it sees real traffic rather than a synthetic local workload. That is the strength and the caution. Cloudflare notes that a Worker needs enough traffic for a useful capture, and TypeScript projects should enable source maps or the profile may expose difficult-to-read generated function names. A single capture is also a sample, not a verdict; repeated profiles are more useful for separating persistent hotspots from one noisy request.

Why it matters

Serverless platforms hide infrastructure well, but they can also hide the precise shape of a performance problem. Bringing CPU and allocation data into the same operational surface reduces that blind spot. Cloudflare says its own teams have used the profiler to investigate out-of-memory failures and optimize resource use. That is an attributed result, not independent testing.

Opinion: this is the kind of observability that makes edge computing feel less like debugging through a keyhole. Built for the edge cases, literally.

AI-assisted summary; sources were reviewed and the article was edited for accuracy and clarity.