World · Cloud Technologies
Cloudflare adds automatic retries to Durable Objects calls
Source announcement date: 2026-11-05 · News/event date: 2026-11-05
On November 5, 2026, Cloudflare announced runtime retries for certain transient failures affecting Durable Objects calls. The change concerns fetch and RPC calls from Workers or other Durable Objects; it does not mean unlimited retries for every failed operation.

Which failures are covered?
The default permits up to four retries after the initial call; no new retry starts more than 10 seconds after that call begins. Errors from the application's own code and errors marked as overloaded are not retried automatically. The technical documentation also distinguishes HTTP error responses, such as a 500 response, from the errors this mechanism retries.
When receipt of the request is uncertain, the same request identity is used. A Durable Object that received the first request rejects the retry unless the method is additionally marked with retryable(). Automatic retrying is therefore not presented as unconditional duplicate execution of every method. The announcement specifies Wrangler 4.158.0 or later for the new settings and marking utility.
Quad Bilişim comment
In business cloud applications, restoring a connection and performing a job the correct number of times are separate reliability concerns. Reading the same record again and adding an item to the same order a second time do not have the same effect. For businesses, this announcement matters because it separates brief infrastructure disruptions from application-logic errors; it does not by itself guarantee that an entire workflow completes without error.
Source
Cloudflare — Automatic retries for Durable Object calls, November 5, 2026