* Given that the new instance wants to access the application on the existing compute engine instance, these applications seem to be related so they should be within the same VPC. It is possible to have them in different VPCs and peer the VPCs but this is a lot of additional work and we can simplify this by choosing the option below (which is the answer)
1. Create a subnet in the same VPC, in europe-west1.
2. Create the new instance in the new subnet and use the first instance subnets private address as the endpoint.
is the right answer.
* We can create another subnet in the same VPC and this subnet is located in europe-west1. We can then spin up a new instance in this subnet. We also have to set up a firewall rule to allow communication between the two subnets. All instances in the two subnets with the same VPC can communicate through the internal IP Address Ref: https://cloud.google.com/vpc

The best option is to attach a single service account to the compute instances and add minimal rights to the service account. Then, allow the service account to impersonate a Cloud Identity user with elevated permissions to create, update, or delete resources. This way, the service account can use short-lived access tokens to authenticate to Google Cloud APIs without needing to manage service account keys. This option follows the principle of least privilege and reduces the risk of credential leakage and misuse.
Option A is not recommended because it requires human intervention, which can slow down the CI/CD pipeline and introduce human errors. Option C is not secure because it grants all required IAM permissions to a single service account, which can increase the impact of a compromised key. Option D is not cost-effective because it requires creating and managing multiple service accounts and keys, as well as using a secret manager service.
1: https://cloud.google.com/iam/docs/impersonating-service-accounts
2: https://cloud.google.com/iam/docs/best-practices-for-managing-service-account-keys
3: https://cloud.google.com/iam/docs/understanding-service-accounts

Ref: https://cloud.google.com/compute/docs/machine-types#n1_machine_type

auto-provisioning = Attaches and deletes node pools to cluster based on the requirements. Hence creating a GPU node pool, and auto-scaling would be betterhttps://cloud.google.com/kubernetes-engine/docs/how-to
/node-auto-provisioning