GCP Databases (stub)

Status: stub. Production depth pending.

Decision tree

Quick starts

# Cloud SQL for Postgres
gcloud sql instances create my-pg \
  --database-version POSTGRES_16 \
  --region africa-south1 \
  --tier db-custom-2-7680 \
  --storage-type SSD --storage-size 20 --storage-auto-increase \
  --availability-type ZONAL \
  --backup --enable-point-in-time-recovery

# Firestore (Native mode)
gcloud firestore databases create --location=africa-south1 \
  --type=firestore-native

# Memorystore Redis
gcloud redis instances create my-cache \
  --size 1 --region africa-south1 --tier basic

Connecting from Cloud Run to Cloud SQL

# Option 1: Cloud SQL Auth Proxy built in (recommended)
gcloud run services update my-service --region africa-south1 \
  --set-cloudsql-instances=my-app-prod:africa-south1:my-pg \
  --set-env-vars INSTANCE_UNIX_SOCKET=/cloudsql/my-app-prod:africa-south1:my-pg
# Connect via unix socket — no IP, no network config

Gotchas (high-level)

• Cloud SQL private IP requires a VPC + allocated service range — plan at project creation.
• AlloyDB and Spanner have high floor costs (~$200+/mo for the cheapest valid config). Cloud SQL starts much lower.
• Firestore Native vs Datastore mode are different products — Native is what modern apps use; don't create in Datastore mode by accident.
• Spanner needs at least 100 processing units (~$65/mo) to exist. Below 1000 PUs, query concurrency is limited.
• Memorystore Basic has no HA; Standard has failover. Choose based on "can we tolerate cache-flushing restart?"

See Also

• GCP parent
• AWS database sibling
• Cloudflare D1 (SQLite at edge)
• Cloudflare Hyperdrive (pool Postgres from Workers)