Set Timeout Budgets across the Whole Request

A practical approach

Step 1

Set a realistic total time budget for the user journey, then allocate smaller limits to required downstream operations.

Step 2

Pass cancellation or deadline information where supported. Include retries and backoff in the same overall budget.

Step 3

Define the response when the deadline expires. For state-changing work, check whether the action completed before telling the client to repeat it.

Worked example

An assistant retrieves documents, calls a model and writes a result. Giving each stage a long independent timeout can make the total wait unacceptable. A shared deadline lets the workflow stop with a useful partial result or convert the task into an asynchronous job.

A mistake to avoid

A client timeout does not necessarily cancel server work. If the client retries a write without reconciliation or idempotency, the original and repeated operations may both complete.

Try it yourself

Draw the request chain for one feature and assign a budget to each stage. Inject a slow dependency and verify that the total response stays within the intended limit.

Your implementation checklist

• The user journey has one overall budget.
• Retries consume that budget.
• Cancellation is propagated where possible.
• Timed-out writes are reconciled safely.