> ## Documentation Index
> Fetch the complete documentation index at: https://docs.e6data.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Request identity and revisions

> Identify one intended operation with the identity field that API expects. Keep that identity and request content when recovering a lost response.

How-to guideApplies to: Unified engine deployments

Identify one intended operation with the identity field that API expects. Keep that identity and request content when recovering a lost response.

## Know the identity field

| Operation family | Common identity field |
| - | - |
| SQL execution | `operation_id`. |
| Workload creation, runs, and many document actions | `submission_id`. |
| Catalog mutations | `mutation_id`. |
| Resource updates | A revision guard as well as any required request identity. |

The exact request schema decides which fields are supported. There is no universal Idempotency-Key request header for all engine APIs. Notification delivery has its own event identity.

## Recover the same request

If a mutation returns a server error or its response is interrupted, the example keeps the exact prepared URL, method, headers, and body. Its example values are locked. For requests with a supported identity, **Retry same request** resends that exact request.

A fresh ID can create new work. Read status using the original identity when possible. Open-session creation does not have a caller-supplied idempotency identity; inspect server state before creating another session after an uncertain response.

A successful write is marked **Sent** to avoid starting the same work again by mistake. Start new work from your own client with a new identity. Leaving the page does not cancel or undo earlier work.

## Handle revision conflicts

`expected_revision` protects a workload or document against a stale edit. Catalog mutations use their catalog revision contract. A conflict means you should read and review the current resource; it is not a signal to replace the revision silently and repeat a write.

A selective rerun preserves an earlier invocation. A normal new run can use the current definition. See [Workloads](https://docs.e6data.com/transformation/workloads.md) for execution semantics.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.