GCP Setup Tool (setup-gcp)

This tool automates the provisioning and configuration of Google Cloud Platform (GCP) resources required to run Agent Substrate. It is designed to be idempotent, meaning it can be run multiple times safely to ensure the environment is correctly configured.

Overview

The setup-gcp tool provisions and configures GCP resources for Agent Substrate. It uses a hierarchical command structure:

• setup-gcp (root)
  • enable apis - Enable required GCP APIs.
  • create - Parent command for resource creation.
    • cluster - Create GKE cluster.
    • bucket - Create GCS bucket.
    • iam - Create IAM policy bindings and grant permissions.
    • dashboards - Create Cloud Monitoring dashboards.
    • cloudsql - Create a Cloud SQL PostgreSQL instance for the ateapi store, with IAM database authentication (see cloud-sql.md).
  • bootstrap - Run all setup steps in order.

Prerequisites

1. Go: Ensure Go is installed (version compatible with the project, see root go.mod).
2. Google Cloud SDK (gcloud): Installed and authenticated.
3. Application Default Credentials (ADC): The tool uses Google Cloud client libraries. You must set up ADC:

   gcloud auth application-default login
   

4. Target Project: You must have a GCP project created and have sufficient permissions (typically Owner or Editor) to create resources like GKE clusters, GCS buckets, and IAM bindings.

Configuration & Defaults

All CLI flags can be configured via environment variables. If an environment variable is set, it will be used as the default value for the corresponding flag. Command-line flags always take precedence over environment variables.

A convenient way to manage these is to copy the example environment file, customize it, and source it before running the tool:

cp hack/ate-dev-env.sh.example .ate-dev-env.sh
# Edit .ate-dev-env.sh to match your project and preferences
source .ate-dev-env.sh

Global Flags

These flags can be passed to the root command and apply to all subcommands:

Subcommands

1. Enable APIs

Enables the required GCP APIs for the project.

go run ./tools/setup-gcp enable apis [flags]

Flags:

• (Uses global --project-id)

2. Create Cluster

Creates a GKE cluster configured for Agent Substrate (with Workload Identity, required Kubernetes beta APIs, and Managed OpenTelemetry enabled, and the Filestore CSI driver disabled).

gcloud container clusters create "${CLUSTER_NAME}" ... \
  --enable-kubernetes-unstable-apis=certificates.k8s.io/v1beta1/podcertificaterequests,certificates.k8s.io/v1beta1/clustertrustbundles

gcloud container node-pools update "${NODE_POOL}" \
  --cluster "${CLUSTER_NAME}" --location "${CLUSTER_LOCATION}" \
  --no-enable-autoupgrade

go run ./tools/setup-gcp create cluster [flags]

Flags:

NODE_MACHINE_TYPE was named GVISOR_NODE_MACHINE_TYPE. The old name still works and warns; rename it in your .ate-dev-env.sh.

Node version labels: pool labels are the birth default for every node GKE creates later (autoscaling, auto-repair, node upgrades), and setup-gcp does not set ate.dev/substrate-version on the pool, so those nodes arrive unlabeled and run no dataplane pods (see the README's note on node version labels). Stamp the pool with gcloud container node-pools update ... --node-labels=... (list the existing labels first and carry them all over; the flag replaces the full set), and create additional pools with --node-labels=ate.dev/substrate-version=<build version>.

3. Create Bucket

Creates a GCS bucket for storing snapshots.

go run ./tools/setup-gcp create bucket [flags]

Flags:

*Note: Required unless the BUCKET_NAME environment variable is set.

4. Create IAM

Configures IAM permissions and Workload Identity bindings.

go run ./tools/setup-gcp create iam [flags]

Flags:

*Note: Required for bucket bindings unless the BUCKET_NAME environment variable is set.

What create iam actually grants

On a fresh GCP project these bindings do not exist, and nothing else creates them — without them atelet cannot read or write external snapshots (403 from GCS on the first suspend/resume), ate-api cannot copy or collect them, and nodes cannot pull images. bootstrap and create iam apply them for you; the table below is the reference for auditing them, or for applying them by hand in projects where you cannot run the tool with project-level IAM permissions.

atelet and ate-api authenticate via GKE Workload Identity Federation: their Kubernetes ServiceAccounts (ate-system/atelet and ate-system/ate-api-server, created by install-ate.sh) are addressed directly as IAM principals — no Google service account is created or impersonated. This only works on a cluster with Workload Identity enabled (create cluster enables the PROJECT_ID.svc.id.goog pool; a pre-existing cluster must have it enabled too, or their GCS calls fail with 401).

¹ principal://iam.googleapis.com/projects/PROJECT_NUMBER/locations/global/workloadIdentityPools/PROJECT_ID.svc.id.goog/subject/ns/ate-system/sa/SERVICE_ACCOUNT — note it is keyed to the project number, not the ID. ² Project-level today; scoping these down is tracked in the TODOs in cmd/iam.go. ³ PROJECT_NUMBER-compute@developer.gserviceaccount.com (least-privileged node SA is #76).

Manual equivalent:

WI="principal://iam.googleapis.com/projects/${PROJECT_NUMBER}/locations/global/workloadIdentityPools/${PROJECT_ID}.svc.id.goog/subject/ns/ate-system/sa"
ATELET="${WI}/atelet"
ATE_API="${WI}/ate-api-server"

gcloud projects add-iam-policy-binding "${PROJECT_ID}" \
  --member="${ATELET}" --role=roles/storage.objectAdmin
gcloud projects add-iam-policy-binding "${PROJECT_ID}" \
  --member="${ATELET}" --role=roles/artifactregistry.reader
for MEMBER in "${ATELET}" "${ATE_API}"; do
  gcloud storage buckets add-iam-policy-binding "gs://${BUCKET_NAME}" \
    --member="${MEMBER}" --role=roles/storage.objectAdmin
  gcloud storage buckets add-iam-policy-binding "gs://${BUCKET_NAME}" \
    --member="${MEMBER}" --role=roles/storage.bucketViewer
done

5. Create Dashboards

Creates or updates Cloud Monitoring dashboards.

go run ./tools/setup-gcp create dashboards [flags]

Flags:

6. Bootstrap (All Steps)

Runs all the setup steps in the correct order to fully bootstrap the environment.

go run ./tools/setup-gcp bootstrap [flags]

Flags:

*Note: Required unless the BUCKET_NAME environment variable is set.

--enable-nested-virtualization is on by default and needs a --machine-type that supports the feature; pass --enable-nested-virtualization=false to turn it off. It applies only to a cluster this command creates.

Examples

Run the tool from the repository root to ensure relative paths to dashboard configurations are resolved correctly.

Bootstrap everything using environment variables

If you have sourced your .ate-dev-env.sh:

go run ./tools/setup-gcp bootstrap

Bootstrap everything overriding some values

go run ./tools/setup-gcp bootstrap \
  --cluster-name="custom-cluster" \
  --machine-type="n2-standard-8"

Bootstrap a cluster whose nodes cannot run micro-VM workers

Nested virtualization is on by default, so turning it off is the case that needs a flag:

go run ./tools/setup-gcp bootstrap \
  --enable-nested-virtualization=false

Only create the cluster (using env vars for defaults)

go run ./tools/setup-gcp create cluster