NodeFox

Lesson 21 · Advanced

Handle failures without duplicate actions

Separate retryable reads from uncertain writes and human recovery.

Written guide · 3 min read · video not yet produced

Before you start

A signed-in NodeFox workspace. Use synthetic data and prepare outputs before enabling external writes.

  1. 01Classify the failure

    Distinguish invalid input, unavailable dependency, rejected authorization and uncertain external action. A timeout after sending a write is ambiguous: the destination may have accepted it.

  2. 02Preserve context

    Record the input identifier, proposed action and available response. Use stable business IDs. Do not discard the evidence needed to determine whether an order, ticket or payment was already created.

  3. 03Design explicit recovery

    Use the supported branching and human review mechanisms to decide the next step. Automatic retry, backoff and dead-letter handling are not implied by the graph. Implement and test the specific policy you require.

  4. 04Reconcile before repeating

    Check the destination using its supported lookup or idempotency mechanism. Retry only when the consequences are understood. For learning exercises, use a local proposal file so recovery does not affect real records.

If the result is wrong

The absence of a success response is not proof that a remote side effect did not occur.