Workers profiling exposes CPU samples and memory allocations

Cloudflare now lets developers capture on-demand CPU and memory profiles from active Workers and Durable Objects. Results appear as flamegraphs, where wide stacks identify code responsible for more sampled work or allocations. Profiles can be requested from the dashboard or command line and downloaded for further analysis, giving more detail than aggregate resource metrics.

Cloudflare · Oct 9, 2026

Why it matters

Useful when a deployed Worker consumes unexpected CPU or memory and logs do not identify the responsible code.

More context

Try a capture from the Worker’s Observability page or the documented cf CLI. Choose a version receiving enough traffic and enable source maps for understandable TypeScript names. Durable Objects can be targeted by name; ordinary Worker profiling selects a running isolate.

Keep in mind

This is on-demand sampling, not a historical recording of every request. Memory profiles include allocations during the selected window and can miss startup activity. Continuous profiling is future work in the announcement. We have not measured overhead or captured a production profile.

Original read