Google Professional Data Engineer

Explanation

When you need to make code changes that make a new Dataflow pipeline incompatible with the current version while ensuring no data loss, the recommended approach is:

Option B is correct because:

• Creating a new Pub/Sub subscription allows both pipelines to run independently
• Running pipelines in parallel ensures continuous data processing
• The old pipeline can continue processing existing messages without interruption
• Once the old pipeline processes all existing messages, it can be safely stopped
• This approach maintains data integrity and prevents data loss

Why other options are incorrect:

• A: Dataflow cannot automatically handle state migration for incompatible pipeline versions
• C: Pausing and resuming doesn't work for incompatible code changes
• D: Stopping and restarting would cause data loss in streaming pipelines
• E: Templates help with deployment but don't solve the compatibility issue
• F: State export/import is complex and not recommended for incompatible changes

This approach is part of Google Cloud's recommended practice for zero-downtime pipeline updates with incompatible changes.