Canva Reconciliation and Recovery

Overview

Reliability begins with knowing whether an operation happened. Reconcile existing resources and jobs before replaying a mutation, and keep credential rotation outside generic request retries.

Prerequisites

• Operation types, retry classifications, and duplicate impact
• Durable identity/job ledger and terminal-state contract
• Endpoint/user queues, time budgets, alert owner, and rollback

Instructions

Step 1: Classify operations

Use Read and Grep to label safe read, token exchange, mutating submission, job poll, webhook delivery, and cleanup. Define retriable errors per operation.

Step 2: Persist before dispatch