Cloudflare launches Traces in public beta: site operators can follow every step a request takes through the platform on one timeline

What Cloudflare announced

On October 2, 2026, Cloudflare announced Cloudflare Traces on its official blog, offered as a public beta. "Tracing" means recording, in chronological order, every step a network request goes through from arrival to completion, so you can later check where something went wrong. According to Cloudflare, the feature extends automatic tracing, previously limited to Workers (Cloudflare's service for running code on its platform), to the other stages a request passes through on the Cloudflare platform. Cloudflare says a single trace can show supported security rules, transformations, cache decisions, routing, Worker execution and origin server (the website's own server) processing, and can continue into services running on Cloudflare, on the origin, or elsewhere. Cloudflare also describes this as a long-term investment in OpenTelemetry (an open standard for monitoring data).

Cloudflare says the traces its internal teams use when investigating issues often contain thousands of spans (segments of a trace that each record a single step, including its duration and result), generated by dozens of services and features; Cloudflare Traces aims to give all users similar visibility. Workers Tracing, launched last year, was the first step. As Cloudflare describes it, it automatically records Worker calls, including outbound fetches and calls to KV, R2, D1, Durable Objects and other Workers.

Key features at a glance

• Automatic tracing: Cloudflare says that once tracing is enabled for a domain, spans are generated automatically, with no special instrumentation (extra code added to record what an application does), configuration or plugins required.
• Sampling control: the sampling rate is the share of requests that get recorded. You can set a baseline sampling rate and then override it for specific traffic with Trace Rules, which use the same rule language as Cloudflare Rules.
• Context propagation: the W3C traceparent header is a standard request header used to pass the identity of a single trace between systems. Cloudflare can accept an incoming traceparent header (governed by an incoming propagation policy) and can also forward a new traceparent header to the origin server.
• Viewing and export: you can view the request timeline and span details in the dashboard, or export via OTLP (OpenTelemetry's data transport protocol) to a compatible observability platform.
• Agent assistance: Cloudflare says that through the Cloudflare Observability MCP server (a connector that lets AI assistants use a service's tools and data), AI coding agents can query trace data using the SQL API.

Cloudflare gives an example of how sampling can be used: trace only 1% of requests normally; when a customer reports a problem, trace 100% of traffic for their hostname, source IP or an identifying request header, while other traffic stays at the baseline. Cloudflare also shows in the post a request that was a cache miss (the content was not already stored in Cloudflare's cache), so it was sent on to the origin server. Of the 539ms total, 527ms was spent waiting for a response, which shows how tracing pinpoints where time goes.

Pricing plans compared

Cloudflare says Cloudflare Traces will be part of the unified Cloudflare Observability pricing model. It is billed by ingestion (the amount of observability data sent in) and by how long that data is kept, rather than by number of spans or events. Storage is measured in GB-months, meaning one gigabyte kept for one month. The new pricing applies to Cloudflare Tracing and Workers Tracing from December 1, 2026. The table below summarizes the plans Cloudflare published.

What it means for everyday users and site operators

For everyday internet users, this is a diagnostic tool for website and service operators, and browsing websites usually requires no changes. For website operators and developers using Cloudflare, according to Cloudflare's description, some questions that previously had to be pieced together from scattered logs and settings, such as why a request was blocked, whether a URL was rewritten, whether a response came from cache, and which stage consumed the time, can now be viewed on a single timeline. Because trace data volume is tied to cost, the sampling rate setting directly affects the trade-off between visibility and cost.

What comes next

Cloudflare says it plans to deliver the following after the public beta: broader automatic instrumentation (for example DDoS rules, Access, and Workflows, Queues and Pipelines), authenticated context propagation, ad hoc tracing, OpenTelemetry API support in Workers, and retention of up to 365 days. It is currently available via the dashboard, API or Terraform, with support for export to OTLP destinations. These are Cloudflare's stated plans; the announcement does not give actual release timelines.

Browse the latest news in this topic

• Tech news