Before you begin:
1. In the Cloud Console, on the project selector page, select or create a Cloud project.
2. Make sure that billing is enabled for your Google Cloud project. Learn how to confirm billing is enabled for your project.
To use the gcloud command-line tool for this quickstart, you must first install and initialize the Cloud SDK:
1. Download and install the Cloud SDK using the instructions given on Installing Google Cloud SDK.
2. Initialize the SDK using the instructions given on Initializing Cloud SDK.
To use gcloud in Cloud Shell for this quickstart, first activate Cloud Shell using the instructions given on Starting Cloud Shell.
https://cloud.google.com/ai-platform/deep-learning-vm/docs/quickstart-cli#before-you-begin

Cloud Run is a fully managed platform that allows you to deploy and run containerized applications without worrying about the underlying infrastructure. Cloud Run automatically scales your application based on the incoming requests, and you only pay for the resources you use. Cloud Run also supports deploying multiple revisions of your service, and splitting the customer traffic between them. This allows you to gradually roll out new features, test them in production, and monitor their performance and reliability. If you encounter any problems with the new revision, you can easily roll back to a previous revision by adjusting the traffic split. This is a Google-recommended practice for managing revisions to a service, as it minimizes the risk associated with a new revision, reduces the number of customers who might be affected by an outage, and does not introduce any development or operational costs to your customers.
The other options are not correct because they either do not follow the Google-recommended practice, or they do not reduce the risk or cost associated with a new revision. Option B is not correct because deploying your application to a second Cloud Run service, and asking your customers to use the second Cloud Run service, is not a gradual roll out, but a complete switch. This may cause confusion and inconvenience for your customers, and it does not allow you to monitor the performance and reliability of the new revision in comparison with the old one. Option C is not correct because asking your customers to retry access to your service with exponential backoff to mitigate any potential problems after the new revision is deployed, is not a good user experience, and it does not address the root cause of the problems. Option D is not correct because sending all customer traffic to the new revision, and rolling back to a previous revision if you witness any problems in production, is a risky and costly strategy, as it exposes all your customers to the potential problems, and it does not allow you to test the new revision before deploying it to all customers.
Reference:
Cloud Run documentation
Managing revisions
Splitting traffic