Connect Google Cloud to Prunio
About 10 minutes, one prepared script in Cloud Shell. No keys or passwords change hands, and you can start read-only.
- No service-account keys
- No Owner or Editor role for Prunio
- No agent in your GKE clusters
- One connection per project
Before you start
- The detailed billing export to BigQuery is on. In the Google Cloud console: Billing → Billing export → BigQuery export → Detailed usage cost. Google does not backfill this export, so costs only exist from the day it is turned on. If it is off, turn it on today.
- The person running the setup can change IAM on the project. Project Owner is simplest. Without Owner: Project IAM Admin, Workload Identity Pool Admin, Service Account Admin, Role Administrator and Service Usage Admin on the project, Billing Account Viewer on the billing account, and BigQuery Data Owner on the billing export dataset. The script checks this first and names anything missing. These rights are only needed during setup.
- You know which project to connect. Each project is one connection. Start with the one that has your GKE clusters and the most spend.
- Pop-ups are allowed for app.prunio.ai. Cloud Shell opens in a new window.
Connect, step by step
- Sign in. Go to app.prunio.ai and sign up. A new workspace starts with a 30-day trial.
- Pick your cloud: GCP.
- Choose how much Prunio can do. Read-only reads inventory, metrics and billing and changes nothing. Scoped Optimize also starts and stops plain VMs on schedules you set. You can switch later.
- Get the setup code and open Cloud Shell. Choose Guided Cloud Shell, copy the one-time setup code, then press Open Cloud Shell. The code never goes in a URL. If your company blocks Cloud Shell, the same page shows a command for your own terminal (it needs gcloud, bq, git, jq and curl).
- Run the script in Cloud Shell. Sign in with
gcloud auth loginif no account is active, then run./setup.sh. The script first checks it is the exact reviewed release; nothing in Google Cloud changes before that check passes. Paste the setup code, pick the project, accept the billing export dataset it finds, and type the project ID to confirm. Keep the cleanup commands it prints. - Back in Prunio. Prunio checks access, runs a safety check and starts the first scan by itself. First results appear within a few minutes.
What the setup creates in your project
Created: one Workload Identity Federation pool and provider (how Prunio signs in without keys), one connector service account with exact custom roles, and a billing view filtered to this project only.
Never created: service-account keys, Owner or Editor grants, organization or folder IAM changes, or anything inside your GKE clusters.
Prunio's read access covers listing VMs, disks, reservations and commitments, GKE clusters, Cloud Monitoring metrics, Google's machine-type recommendations, asset search, and queries on the project's billing view. It cannot read data inside your project: no storage objects, database contents or secrets.
What you will see
- Within minutes: VMs, disks, Cloud SQL, and GKE clusters with each node pool's billed cost per month. VM findings: idle or oversized VMs, unattached disks, idle Cloud SQL.
- After 7 days of data: node-pool findings for fixed-size GKE pools that use far less than they have, priced from your own bill.
- Every 6 hours: everything refreshes automatically.
Prunio never treats GKE nodes as ordinary VMs: they get no VM findings and can never be put on a stop schedule. Node-pool findings are advice only; you make the change in Google Cloud or your IaC.
If something stops
- "Detailed billing export not found": turn it on, wait a few hours for Google to fill the table, then run
./setup.shagain. It reuses what it already created. - Missing permissions: the script names the exact role and prints the grant command.
- Cloud Shell did not open: allow pop-ups for app.prunio.ai and press Open Cloud Shell again.
- Setup code expired or already used: reissue it on the same Prunio page.
- More than one project: repeat the steps for each project.
Questions? Write to [email protected].