To ensure the reliability and correctness of Structured Streaming jobs in production, it's recommended to configure them with Unlimited Retries and 1 Maximum Concurrent Run. This setup allows the query to restart on failure without overlapping runs, ensuring data consistency. Additionally, always use a new job cluster with the latest Spark version (or at least version 2.1) for recoverability after upgrades. Avoid setting a schedule or timeout, as streaming queries are designed to run indefinitely. Notifications can be set for failure alerts. Reference