What we build on Google Cloud Platform
The platform choice follows the process. A request-sized container, a Kubernetes cluster, and a virtual machine are different operating models, and we pick one.

Each environment gets a project. Folders group those projects when the organization needs it. Service accounts own deploy and runtime access. Human users do not share a key.
IAM bindings are written down: who can deploy, who can read production data, and which APIs are enabled. A new service starts from that list.

Cloud Run is the default for a stateless container with HTTP or a job. GKE is for a workload that already needs Kubernetes: several services, a mesh, or a platform team that will run the cluster. Compute Engine is for the process that has to be a VM.
We do not put a single container on GKE to make the diagram look complete. The runtime matches who will operate it.

Cloud SQL for relational data the application queries. Cloud Storage for objects and uploads. BigQuery when the work is analysis over large tables, not a row the user is editing. Memorystore when a cache is actually required.
Pub/Sub carries events between services. A service publishes a message it can name. Consumers are separate deploys with their own service accounts.

Cloud Build, or the pipeline you already run, builds the artifact and deploys it into the right project. Environments are not two folders in the same project with a comment that says prod.
Logs and error reporting are on from the first deploy. A service without a way to tell a failed revision from a healthy one is not ready for traffic.
Typical use cases
These are the platform shapes we set up. They are not a claim about a particular customer's bill or traffic.

An API or a worker with a container, a concurrency setting, and a service account that can reach Cloud SQL or Cloud Storage. Revisions are the rollback.

GKE when several services share a cluster and someone will own node pools, upgrades, and namespaces. We set the namespace, the workload identity, and the deploy. We do not treat the cluster as a substitute for IAM.

Events or exports land in BigQuery. The application database stays for transactions. Queries for reporting run against BigQuery, with a dataset and a group that is allowed to read it.

We start with what the process needs: a VM, a container, or a managed SQL instance. The move is a runtime and a data store, plus the IAM that replaces the old shared login.
How we work
Google Cloud work starts with identity and projects. Product selection comes after we know who deploys and where data lives.

Development, staging, and production are separate projects unless there is a written reason not to. Service accounts are created for deploy and for runtime. Keys are a last resort.

HTTP container, Kubernetes workload, scheduled job, or VM. We name the Google Cloud product and the operational task it creates, including who upgrades it.

Transactions in Cloud SQL. Objects in Cloud Storage. Analytical queries in BigQuery. The application gets credentials for the one it needs, not for the whole project.

The pipeline identity can deploy to the target project and cannot rummage through another environment. A rollback is a previous revision or a previous image, tried before the production project depends on it.
Frequently asked questions
Questions about how we use Google Cloud Platform.
OLIANT is a registered partner in the Google Cloud Partner Network. On this page that means we design and run workloads on Google Cloud Platform: projects, IAM, and the managed services the application uses.
Cloud Run when you have a container and you want revisions, scale, and a small operating surface. GKE when you already need Kubernetes and someone will run the cluster. A single service is rarely a reason to start a cluster.
Yes. We use the folder, project, and organization policies you have. New services get projects and service accounts inside that structure. We do not stand up a shadow organization beside it.
Yes. AWS and Microsoft Azure have their own pages. A system can span more than one cloud, and we still keep each environment's identity and data access explicit. The application language is a separate choice, covered under custom development.