GCP Rightsizing: Machine Types, Node Pools, Disks, Buckets

The GCP rightsizing rules are easiest to read as questions. Each one names the resource, the metric it reads, and the smaller target it would move to, and each prices the move from live GCP catalog rates for both the current and the target SKU.

Is this VM bigger than its load?

GCP VM Oversized, Downsize to Smaller Predefined Machine Type steps a VM down one size to the next predefined machine type and prices the difference from the two hourly rates. The rule was renamed after an audit: it does not build custom machine types, and its title no longer suggests it does. See Compute Engine.

Are the GKE nodes oversized?

• GKE Node Pool Underutilized needs both peak node CPU below 20% and peak memory below 30% over 30 days before it suggests the next-smaller machine type in the same family.
• GKE Node Pool VM-Size Rightsizing downsizes only when measured average node CPU is under 50%, and abstains without the metric.
• GKE Cluster Resource Quotas Overprovisioned looks at quota headroom at the cluster level.

See GKE node pools and GCP GKE.

Is the database on the right tier?

GCP Cloud SQL Right-sizing reads 30-day CPU and picks the next-smaller tier. Its coverage is limited to the standard and high-memory db-n1 families today; custom and Enterprise Plus shapes have no ladder yet, so the rule stays silent on them. GCP Cloud SQL Excessive Backup Retention for Dev/Test trims retention beyond seven days on dev/test instances.

Is the disk faster than it needs to be?

GCP Disk SSD with Low IOPS reads read and write operations over 32 days and suggests moving a pd-ssd disk to pd-balanced when combined average operations stay under 100 per second.

Is the bucket in the right class?

GCS Bucket Standard Class with Low Access requires a measured access rate below 0.05 operations per day on a Standard bucket before suggesting a colder class, and abstains on buckets with no access metric. GCP GCS Bucket Missing Lifecycle Policy currently abstains: pricing an age-based policy needs a per-bucket split of data by object age, which ZopNight does not yet collect, and the rule will not guess.

Is Cloud Run holding capacity it does not use?

Cloud Run Service Low Concurrency fires when a service’s concurrency setting is low and its instance count peaks high, and Cloud Run Dev/Test Service Has Min Instances Greater Than 0 applies to non-production services only, with a production label always taking precedence.

Applying

The VM downsize runs through remediation as a resize with a stop, which is demoted to advisory on stateful workloads. Rightsizing findings record the target they asked for, so adoption can be verified after you apply it. Full list: GCP recommendations; cross-cloud view: rightsizing.