Google Cloud Quotas & Spend Capping Guide

keywords plugin keywords, from plugin.json: gcp
google-cloud
firebase
audit
billing
bigquery-billing-export
quota-governance
cost-optimization
cloud-management
plugin derived: GCP Managementgcp-management
conformance derived from plugin.json $schema: Targets Agent Plugins 1.0.0.
source derived: plugins/gcp-management/skills/gcp-quota-governance/references/quota-capping-guide.md

A plugin-level reference of gcp-management. This location is not defined by Agent Plugins §6.1; the title below is this file's own H1, extracted, not composed.

This reference provides technical guidance on how to enforce hard caps, rate limits, and auto-scaling constraints on Google Cloud Platform and Vertex AI.

1. Cloud Quotas API (`gcloud beta quotas`)

Google Cloud Quotas API allows declaring downward quota preferences. Unlike quota increases (which require manual approval), quota decreases are fulfilled immediately.

Key Quota IDs for Vertex AI (`aiplatform.googleapis.com`)

Quota ID	Metric Description	Dimensions
`OnlinePredictionRequestsPerMinutePerProjectPerRegion`	Online prediction requests/min in a region	`region`
`GenerateContentRequestsPerMinutePerProjectPerRegionPerBaseModel`	Gemini/Multimodal request rate per model	`base_model`, `region`
`GlobalOnlinePredictionTokensPerMinutePerBaseModel`	Global token throughput per minute	`base_model`
`OnlinePredictionInputTokensPerMinutePerRegionPerBaseModel`	Input token rate per minute	`base_model`, `region`
`AnthropicMaasConcurrentBatchPredictionJobs`	Concurrent Claude batch prediction jobs	None (Project-wide)

2. Command Syntax for Setting Quota Caps

A. Cap Regional Online Prediction Requests (Vertex AI / Claude)

gcloud beta quotas preferences create \
  --service=aiplatform.googleapis.com \
  --project=<PROJECT_ID> \
  --quota-id=OnlinePredictionRequestsPerMinutePerProjectPerRegion \
  --dimensions=region=us-central1 \
  --preferred-value=50 \
  --allow-high-percentage-quota-decrease

Effect: Any requests exceeding 50 requests/min in us-central1 return HTTP 429 RESOURCE_EXHAUSTED instead of incurring billing fees.

B. Cap Specific Model Request Rates (e.g. Gemini 1.5 Pro)

gcloud beta quotas preferences create \
  --service=aiplatform.googleapis.com \
  --project=<PROJECT_ID> \
  --quota-id=GenerateContentRequestsPerMinutePerProjectPerRegionPerBaseModel \
  --dimensions=base_model=gemini-1.5-pro,region=us-central1 \
  --preferred-value=10 \
  --allow-high-percentage-quota-decrease

3. Cloud Run Service Scaling Constraints

Cloud Run defaults to a maximum of 100 container instances. To prevent runaway costs during traffic spikes, enforce explicit instance caps:

# Cap service to 2 maximum instances
gcloud run services update <SERVICE_NAME> \
  --project=<PROJECT_ID> \
  --region=<REGION> \
  --max-instances=2 \
  --concurrency=80

4. Quota Reversion & Resetting

To remove a custom quota preference and restore default provider quotas:

gcloud beta quotas preferences create \
  --service=aiplatform.googleapis.com \
  --project=<PROJECT_ID> \
  --quota-id=<QUOTA_ID> \
  --dimensions=<DIMENSIONS> \
  --preferred-value=-1

(-1 indicates reset to default/unlimited).