You have a running streaming Cloud Dataflow job; engineers produced a new version with different windowing and triggers. To update without data loss, what should you do?

Stopping a Dataflow job with "Drain" ensures that all buffered data is processed before the job terminates, preventing data loss. After the old job finishes processing its remaining data, you can then launch a new Dataflow job with the updated code (including new windowing and triggers). Using --update (even with the same job name) is suitable for minor code changes or dependency updates, but it does not support changes to core pipeline logic like windowing or triggers, which require a new job. Canceling the job would immediately stop processing, leading to data loss for in-flight elements.