Google Cloud SQL
by Google Cloud
Fully managed relational database service for MySQL, PostgreSQL and SQL Server.
What it is
Each instance runs on a Google Cloud VM with the database engine and service agents; data lives on scalable network storage and the instance keeps a static IP. Cloud SQL automates backups and recovery, high availability with failover, networking, import/export, maintenance, monitoring and logging.
Features
- Editions: Enterprise and Enterprise Plus (Plus offers sub-second-downtime maintenance, per docs).
- Updates: hardware updates by live migration and online agent updates need no downtime; maintenance updates (OS/database) restart the instance.
- Regional HA with automatic failover, read replicas (including cross-region), point-in-time recovery, backups (including enhanced backups and backup vault options per the PITR docs).
- Security: private IP, Cloud SQL Auth Proxy, TLS, IAM, CMEK via Cloud KMS.
- Database Migration Service for moves from on-prem or other clouds; Datastream to BigQuery for analytics.
- For globally consistent writes or beyond Cloud SQL scale see Cloud Spanner.
AI features (release notes, fetched 2026-10-05)
- Feb 2026: calling Vertex AI models for online prediction and embeddings from SQL, GA for MySQL and PostgreSQL; registering third-party model endpoints (e.g. OpenAI, Hugging Face).
- Feb 2026: the Cloud SQL remote MCP server moved from Preview to GA for all three engines; from Mar 2026 it is enabled automatically when the Cloud SQL Admin API is enabled.
- Mar 2026: conversational analytics (natural-language querying) for MySQL and PostgreSQL via the Conversational Analytics API.
- Aug 2026: MySQL 9.7 GA with standard vector storage and ANN vector indexing; Sep 2026: PostgreSQL
pg_textsearch(BM25) extension; Apr 2026: SQL Server 2025 GA.
Billing model
Billed on CPU and memory, storage, location, network egress and IP addresses, with extras for HA and replicas; new customers get a free trial credit. Use the calculator and https://cloud.google.com/sql/pricing; no amounts here.
Practices (author’s suggestions, opinion)
- Use Auth Proxy or private IP; add a pooler (e.g. PgBouncer) in front for Cloud Run to avoid connection exhaustion.
- Enable HA and maintenance windows in production; test restores regularly.
Sources
- Overview: https://docs.cloud.google.com/sql/docs/introduction (fetched 2026-10-05)
- Release notes: https://docs.cloud.google.com/sql/docs/release-notes (fetched 2026-10-05)
- PITR: https://docs.cloud.google.com/sql/docs/postgres/backup-recovery/pitr (fetched 2026-10-05)
Open items
- Current maximum storage, memory and PITR retention window not confirmed (older draft figures of ~64 TB, ~624 GB and 7-35 days were removed).
- Compliance certification claims not re-verified.