Google Cloud Recommender vs ZopNight: Idle and Rightsizing

What Google’s recommenders do

Google groups its recommendations under Active Assist and says it generates “most recommendations and insights for free to all customers”. Firewall Insights is the one premium recommender. Everyone gets console and API access, but exporting recommendations to BigQuery needs a Standard, Enhanced or Premium support package (Recommender pricing).

Three recommenders matter most for compute cost.

Idle VM recommendations (overview). Compute Engine looks at CPU and network use over an observation period of 14 days by default. For new VMs it starts one day after creation. If both stay under Google’s thresholds, the VM is classed as idle. You can set the period anywhere from 1 to 14 days (a Preview feature). Google does not publish the default thresholds and says they “might change in the future”. Standalone VMs with local SSDs, GPUs or TPUs are excluded, as are App Engine flex, Dataflow and GKE resources.

Machine type (rightsizing) recommendations (docs). These use “the last 8 days” of CPU and memory data, averaged over 60-second intervals. Google says this suits weekly patterns but is “less suited” to workloads that spike monthly or in very brief bursts. There are no recommendations for GKE, Dataflow or App Engine flex VMs, VMs with ephemeral disks, GPUs or TPUs, or the memory-optimized family. Google also skips any recommendation that would save less than $10 a month.

Idle resource recommendations (docs). Persistent Disks, custom images and external IP addresses are flagged once they have gone unused for 15 days. A disk detached for 15 days gets a snapshot-then-delete suggestion. One that was never attached and is blank gets a plain delete. These recommendations are updated once every 24 hours.

Where ZopNight differs

Google’s recommenders tell you what is idle or oversized. Everything after that is left to you.

• Off-hours, not just idle. A development VM used every working day will not be classed as idle over a 14-day window. ZopNight’s schedule recommendations find the hours it could be off, and a schedule, set in your own timezone, stops and starts it. See Compute Engine scheduling.
• GKE is covered. Google’s idle and rightsizing recommenders exclude GKE VMs. ZopNight can stop a GKE cluster on a schedule, or scale the Deployments and StatefulSets in a single namespace to zero and suspend its CronJobs. See GKE scheduling.
• Cloud SQL on a schedule. Cloud SQL instances can go on the same schedules and groups as VMs. See Cloud SQL scheduling.
• Credits counted. With billing sync on, ZopNight reads your BigQuery billing export net of credits, so committed use and sustained use discounts show up the way the invoice bills them.
• A longer metric history. ZopNight’s metrics sync reads 42 days of Cloud Monitoring history for GCP resources, which its rules can draw on beyond Google’s 8-day rightsizing window.

Choose Google’s recommenders if…

• You run only on Google Cloud and want a free list of idle VMs, oversized machines and forgotten disks and IPs.
• Your waste is mostly abandoned resources that can simply be deleted.
• You are happy to apply changes yourself in the console or with gcloud.

Choose ZopNight if…

• Your GCP spend includes GKE clusters and Cloud SQL instances that sit idle every night.
• You also run AWS or Azure and want one list of recommendations with the metric evidence for each.
• You want showback by team and label, budgets per project or resource group, and anomaly alerts alongside the recommendations. See showback vs chargeback.