Implement back-out/rollback for website with hundreds of VMs. Site has frequent critical updates. Create a nearline copy of static data in Cloud Storage. Create a snapshot of each VM prior to update in case of failure. Use managed instance groups with the "rolling-action start update" command when starting a rolling update. Only deploy changes using Deployment Manager templates. The correct answer is C. Use managed instance groups with the "rolling-action start update" command when starting a rolling update. Allows compute engine to handle updates, easy management of VMs. Website with hundreds of VMs, load-balanced likely already using a managed instance group. Now here's a tip. Did you know about this command? This is an example of the level of detail you should be familiar with. If you had studied managed instance groups features, you would have at least seen the "rolling-action start update" option and recognized it in the question. Nearline copy would have been unreliable because once the copy has overwritten, you can't roll it back. Creating VM snapshots could work but it's not an efficient way to backup big data. Also the bigger the data the longer the backup takes which could impact production. Using Deployment Manager templates runs the risk of version conflicts. So, managed instance group features are most efficient for this situation. A car reservation system has long-running transactions, which one of the following deployment methods should be avoided? Execute canary releases, perform AB testing prior to release, introduce a blue-green deployment model, introduce a pipeline deployment model. The answer is introduce a blue-green deployment model. Switching the load balancer from pointing at the green, good environment to the blue new environment is a fast way to roll back if there's a problem during a release. However, long-running transactions would be disrupted by that switch. This question requires you to know a little about AB testing and a little about blue-green deployments and a little about canary releases. They're covered lightly in our courses but it would be advisable to study these separately since they are not Google specific methods. The second link discusses long-running connections and how to support them.