Skip to main content
Now that you know how people and services sign in to your workspace, this page looks at the infrastructure that the workspace runs on. Every workspace has its own compute, database, secrets, and encryption keys, and your plan decides where that infrastructure lives:
  • Baseline and Growth workspaces run on Google Cloud Platform (GCP), in one of our managed regions
  • Scale workspaces run in a Provisionr-managed AWS account, in the United States or the European Union
  • Sovereign workspaces run in your own AWS account, so you can choose any region around the world
Wherever you deploy, your data stays in that region, including backups, which use a secondary region in the same jurisdiction where possible. This gives you direct control over data residency for compliance with regional regulations.

Google Cloud Platform (GCP)

Workspaces on the Baseline and Growth plans run on Google Cloud Platform (GCP), using Cloud Run for compute and Postgres on Cloud SQL for the database. Each region has an independent control plane for data residency and data sovereignty compliance, and each workspace has its own isolated compute, database, secrets management, and encryption keys. We selected GCP for its global presence, its strong security controls, and its ability to isolate the compute and data of each tenant, along with scale-to-zero compute that allows us to scale cost-efficiently.

GCP Regions

You can deploy your workspace in any of our eight GCP regions around the world. Each region has its own independent control plane and infrastructure, which meets data residency and sovereignty requirements and allows us to optimize latency and availability for customers around the world.

GCP Data Isolation

Each region’s workspaces share a GCP project, but every workspace on this shared infrastructure gets its own resources, as the diagram shows.
GCP Isolation DiagramGCP Isolation Diagram

Each tenant runs in its own isolated environment, with isolated compute, database, and secrets. Each service account is granted roles only for its specific tenant's resources.

The word “shared” refers to the GCP project for that region’s workspaces, and to the Cloud SQL instances (compute) that are shared across tenants. Each tenant’s workload is fully isolated, with its own app container, Postgres database, secret management, and encryption key. All of a workspace’s services run with a GCP service account that’s dedicated to that tenant, which provides Workload Identity isolation for each workspace.

Cloudflare Edge Routing

All tenant traffic passes through Cloudflare before it reaches Cloud Run, and Cloudflare handles DDoS mitigation and TLS termination. Along the way, the Cloudflare Worker handles these tasks:
  • Extracting the subdomain from each request
  • Validating the JWT session at the edge
  • Looking up the tenant’s Cloud Run service URL in Workers KV
  • Handling cold starts with a loading page

Cloud Run Serverless Compute

Each tenant’s Cloud Run service runs in a gVisor sandbox, which provides container-level isolation. Each tenant’s service also scales independently, so one tenant under heavy load doesn’t affect the capacity available to another tenant. Because a service scales to zero when it’s idle, the first request after a quiet period can arrive before an instance is running. That’s the cold start that the Cloudflare Worker covers with its loading page.

Database Connections

Each GCP project contains one or more Cloud SQL PostgreSQL instances, and every tenant has its own database within an instance. Cloud Run services connect to their databases through the Cloud SQL Auth Proxy sidecar, which is built into Cloud Run’s native Cloud SQL integration, and each tenant’s service connects only to its own database. The Cloud SQL Auth Proxy provides secure, passwordless authentication with Workload Identity service accounts and encrypted connections to the database, which ensures that one tenant can’t access another tenant’s database.

GCP Secret Management

Each tenant’s application encryption key (APP_KEY) and other secrets are stored in GCP Secret Manager, within the region’s GCP project. The secrets are handled in the same way for every tenant:
  • Each tenant has one secret, with one active version
  • The secret is mounted into the tenant’s Cloud Run service as an environment variable at deploy time
  • An IAM policy restricts access to the tenant’s own Cloud Run service identity
  • The secret isn’t accessible to other tenants or to other services within the project

Managed AWS Account

Workspaces on the Scale plan run on Amazon Web Services (AWS), using ECS Fargate for compute and RDS for the PostgreSQL database. Scale workspaces are managed by Provisionr’s internal orchestration system, which provides isolation and multi-tenancy controls similar to those of the GCP architecture. Each Scale workspace is provisioned in an AWS account that’s shared with other tenants, but each workspace has its own dedicated ECS compute, RDS database server, secrets management, and encryption keys. As with GCP, each region has an independent control plane for data residency and data sovereignty compliance.

Managed AWS Regions

You can deploy your workspace in a Provisionr-managed AWS account in the United States or the European Union.

Self-Hosted AWS Account

The Sovereign plan is available on the AWS Marketplace, and you deploy it in your own AWS account with a CloudFormation template, which gives you complete control over the infrastructure, data, and security. The Provisionr team doesn’t have access to your workspace, and we provide updates through Amazon Elastic Container Registry (ECR). The workload that you deploy has an architecture similar to the Scale plan, with dedicated ECS Fargate compute, RDS database server, secrets management, and encryption keys.

Self-Hosted AWS Regions

You can deploy a Sovereign workspace in any of the AWS regions in this table. In a few regions, only the availability zones listed beside the region identifier are supported.
Within each of these environments, every record that Provisionr stores has an ID that starts with a short prefix for its type, and the next page explains how those prefixes are built. Continue reading: Naming Conventions →