Idle VM Instance (GCP)
CostLens surfaces Compute Engine's idle-VM recommendations so you can stop or right-size instances with low utilisation.
CostLens surfaces idle Compute Engine VMs flagged by Google's Recommender — instances with low CPU/network utilisation over the observation window. Idle VMs are the largest source of GCP compute waste; Google validates the idle signal from actual usage, and stopping the instance captures the full compute cost.
Reversible fix
Stopping a VM is reversible — CostLens can start it again within the 24-hour rollback window. A stopped VM stops compute charges (you still pay for attached persistent disks).
How it works
Recommender analyses usage
Google's Idle Resource Recommender evaluates each VM's utilisation and publishes idle recommendations per zone.
CostLens reads them read-only
Using the recommender.viewer + Compute read scopes from GCP Accounts, CostLens ingests each recommendation and its projected saving.
Severity and savings are assigned
Savings come from the Recommender's costProjection, normalised to a 30-day month.
What to do
- Confirm the VM is genuinely unused (not a warm standby or an occasional batch host).
- Stop it to halt compute charges, or resize it if lightly used (see VM Right-sizing).
- Apply via CostLens (Apply Fix stops the instance; reversible for 24 h) once write access is granted, or in the Cloud Console.
Severity levels
| Severity | Estimated monthly savings |
|---|---|
| critical | >$500 |
| high | $100–$500 |
| medium | $20–$100 |
| low | <$20 |
Required permissions
- Detect:
roles/recommender.viewer+ Compute read — see GCP Accounts. - Apply Fix (stop/start): a write role with
compute.instances.stopandcompute.instances.start(e.g. Compute Instance Admin).
Limitations
- Stopping a VM halts compute billing but not attached persistent disk or reserved IP charges.
- Recommendations depend on Google's observation window; very new instances may not yet have a recommendation.