You must deploy an incompatible update to a running Dataflow streaming pipeline using a Pub/Sub subscription, without losing data. What should you do?

When deploying an incompatible update to a Dataflow streaming pipeline without losing data, the drain flag is the correct approach. Draining a pipeline allows it to finish processing all buffered data and gracefully shut down, ensuring no data loss. After draining, you can update the pipeline code and restart it. Providing a transform-mapping JSON is used for stateful transformations during a job update, but not for an incompatible update that requires a full restart. Creating a new pipeline and reusing the same Pub/Sub subscription would lead to duplicate processing of messages already acknowledged by the old pipeline. Creating a new pipeline with a new subscription would lose all messages processed by the old pipeline and not yet acknowledged, or those still in transit.