Cloud SQL db-g1-small: CPU allocation, minimum share, and relation to Compute Engine g1-small

Hello,

We run Cloud SQL for PostgreSQL (Enterprise edition, asia-northeast1) on the db-g1-small shared-core machine type, and we would like to confirm how CPU is allocated to it.

  1. The Cloud SQL documentation lists db-g1-small as 1 vCPU with a shared core (“Partial for shared vCPU”), but it does not say how much CPU time is allocated. Is the CPU allocation of db-g1-small the same as Compute Engine g1-small (“sustains a single vCPU for up to 50% of CPU time”, fractional vCPU 0.5)?
  2. Is there a minimum (guaranteed) CPU share for db-g1-small? Or can the available CPU fall below 50% when other workloads on the same host are busy?
  3. Under what conditions, and for how long, can db-g1-small burst above its sustained share? How does the burst allowance recover?
  4. In our tests, database/cpu/usage_time leveled off at about 0.50 CPU-seconds per second after a short burst (up to about 0.92). database/cpu/utilization matched the 2-minute average of usage_time. Is utilization defined as a 2-minute average? Is there any metric that shows CPU contention (steal) for shared-core instances?

We understand that shared-core instances are not covered by the SLA and are intended for testing and development. We want to describe the behavior accurately in our internal documents.

Thank you.

Cloud SQL db-g1-small should not be assumed to have the same 50% CPU limit as Compute Engine g1-small. The test shows around 50% sustained usage and a temporary 92% burst, but Google does not guarantee these limits or explain its burst recovery.

Google provides CPU usage, reserved cores, and utilization metrics, sampled every 60 seconds. However, the exact CPU guarantee, burst behavior, and host-level contention rules are not documented. Contact Google Cloud Support for confirmation.

References: