GCP Cloud SQL Database Deployment Guide

Databases are the backbone of application architecture. GCP Cloud SQL provides fully managed relational database services supporting MySQL, PostgreSQL, and SQL Server. This guide walks you through deploying a production-grade Cloud SQL instance on GCP.

Cloud SQL Overview

Engine Support

| Engine | Supported Versions | Use Cases | |--------|-------------------|-----------| | MySQL | 5.7, 8.0, 8.4 | Web apps, e-commerce, CMS | | PostgreSQL | 13, 14, 15, 16 | GIS, complex queries, JSON processing | | SQL Server | 2019, 2022 | Enterprise ERP, .NET migrations |

Comparison with Other Providers

| Feature | GCP Cloud SQL | Alibaba Cloud RDS | AWS RDS | |---------|--------------|-----------|------| | MySQL version | 8.4 | 8.0 | 8.0 | | Max storage | 10 TB | 6 TB | 64 TB | | Backup retention | 365 days | 730 days | 35 days | | Private connectivity | Private Service Connect | VPC Endpoint | PrivateLink | | Cross-region read replicas | Yes | Yes | Yes | | Free tier | Yes (limited) | No | Yes (12 months) |

Creating a Cloud SQL Instance

Console Creation

1. Navigate to SQL → Create Instance
2. Select the database engine (MySQL 8.0 in this example)
3. Configure instance parameters:

| Setting | Recommendation (Production) | Notes | |---------|----------------------------|-------| | Instance ID | prod-mysql-primary | Lowercase + hyphens | | Password | Strong + auto-rotation | Enable password policy | | DB version | MySQL 8.0 | Latest stable | | Region | asia-east1 (Taiwan) | Closest to users | | Machine type | db-custom-4-16384 | 4 vCPU, 16 GB | | Storage | 100 GB SSD | Enable auto-increase | | HA | Regional HA | Cross-zone failover | | Auto backup | On, 02:00 UTC | Retain 30 days | | Private IP | Enabled | Access via VPC |

gcloud CLI Creation

gcloud sql instances create prod-mysql-primary \
  --database-version=MYSQL_8_0 \
  --tier=db-custom-4-16384 \
  --region=asia-east1 \
  --storage-type=SSD \
  --storage-size=100GB \
  --storage-auto-increase \
  --availability-type=REGIONAL \
  --backup-start-time=02:00 \
  --enable-point-in-time-recovery \
  --network=projects/my-project/global/networks/my-vpc \
  --no-assign-ip

Machine Types and Pricing

Shared-Core Machines

| Type | vCPU | Memory | Monthly Cost (asia-east1) | Use Case | |------|------|--------|-------------------------|----------| | db-f1-micro | 1 (shared) | 0.6 GB | ~$7 | Dev/test | | db-g1-small | 1 (shared) | 1.7 GB | ~$18 | Lightweight apps |

Custom Machine Types (Recommended)

Custom machine types let you independently select vCPU and memory for precise resource matching:

| Config | vCPU | Memory | Monthly (Single Zone) | Monthly (HA) | |--------|------|--------|----------------------|-------------| | db-custom-2-8192 | 2 | 8 GB | ~$110 | ~$220 | | db-custom-4-16384 | 4 | 16 GB | ~$210 | ~$420 | | db-custom-8-32768 | 8 | 32 GB | ~$420 | ~$840 | | db-custom-16-65536 | 16 | 64 GB | ~$840 | ~$1,680 |

Compared to Tencent Cloud TDSQL, Cloud SQL custom machines offer more flexible memory ratios (1:2 to 1:8), ideal for workloads with diverse requirements.

Storage Pricing

| Storage Type | Price (/GB/month) | Performance | Use Case | |-------------|------------------|-------------|----------| | SSD | $0.17 | High IOPS | OLTP production | | HDD | $0.11 | Low IOPS | Logs, archiving |

High Availability Architecture

Regional HA

Cloud SQL HA uses a cross-zone primary-standby hot architecture:

Zone a: Primary ←sync replication→ Zone b: Standby
                ↑
         Automatic failover

• RPO ≈ 0 (synchronous replication)
• RTO ≈ 60–120 seconds
• Additional cost: Compute doubles; storage does not

Cross-Region Disaster Recovery

Implement DR through cross-region read replicas:

gcloud sql instances create prod-mysql-dr \
  --master-instance-name=prod-mysql-primary \
  --region=asia-northeast1 \
  --tier=db-custom-4-16384

Cross-region replication is asynchronous, typically with sub-second lag. Cost includes cross-region network egress ($0.01/GB).

Read Replicas

Add read replicas for read-heavy applications:

gcloud sql instances create prod-mysql-replica1 \
  --master-instance-name=prod-mysql-primary \
  --region=asia-east1

| Metric | Details | |--------|---------| | Max replicas | MySQL: 10, PostgreSQL: 20 | | Replication mode | Asynchronous | | Promotable to primary | Yes | | Connection | Use replica IP |

Networking and Connectivity

Private IP Connection (Recommended)

Compute Engine and GKE access Cloud SQL via VPC peering without public exposure:

1. Specify VPC network during instance creation
2. Allocate private IP range for Cloud SQL
3. Use Private Service Connect or VPC peering

Cloud SQL Auth Proxy

The most secure connection method—no SSL certificate or IP allowlist management needed:

# Install Auth Proxy
wget https://dl.google.com/cloudsql/linux/amd64/cloud_sql_proxy
chmod +x cloud_sql_proxy

# Start proxy
./cloud_sql_proxy -instances=my-project:asia-east1:prod-mysql-primary=tcp:3306

Applications connect to localhost:3306 for secure database access.

Connection Pool Configuration

| Parameter | Recommended Value | Notes | |-----------|------------------|-------| | max_connections | 4000 (4 vCPU instance) | Default: vCPU × 1000 | | connect_timeout | 10 seconds | Prevent connection pile-up | | wait_timeout | 300 seconds | Release idle connections promptly | | Pool size | CPU cores × 2 + disk spindles | HikariCP recommendation |

Backup and Recovery

Automatic Backups

• Point-in-Time Recovery (PITR): Enable binlog retention to recover to any second
• Retention: 7–365 days configurable
• Cost: Backup storage $0.08/GB/month

Manual Backups and Exports

# Create on-demand backup
gcloud sql backups create --instance=prod-mysql-primary

# Export SQL to Cloud Storage
gcloud sql export sql prod-mysql-primary gs://my-bucket/backup.sql \
  --database=mydb

Recovery Operations

# Restore from backup (creates new instance)
gcloud sql backups restore BACKUP_ID --restore-instance=prod-mysql-restored

# Point-in-time recovery
gcloud sql instances clone prod-mysql-primary prod-mysql-pitr \
  --binlog-file-position=mysql-bin.000123,45678

Performance Optimization

Key Parameter Tuning

| Parameter | Default | Recommended | Notes | |-----------|---------|-------------|-------| | innodb_buffer_pool_size | 128 MB | 60–70% of physical memory | Most critical parameter | | innodb_log_file_size | 48 MB | 1–4 GB | Reduce checkpoint frequency | | innodb_flush_method | fdatasync | O_DIRECT | Avoid double buffering | | max_connections | 4000 | Adjust as needed | Use with connection pool |

Monitoring Metrics

Monitor these key metrics in Cloud Monitoring:

• database/cpu/utilization — CPU utilization
• database/disk/utilization — Disk usage
• database/memory/utilization — Memory usage
• database/network/connections — Connection count
• database/mysql/innodb_buffer_pool_hit_ratio — Buffer pool hit ratio

Security Hardening

1. Private IP only: Disable public access to the database
2. Cloud IAM Authentication: MySQL 8.0+ supports IAM DB authentication
3. SSL/TLS Encryption: Enforce SSL on all connections
4. Data Encryption: Default Google-managed keys; CMEK also available
5. Audit Logging: Enable Cloud SQL Admin API audit logs
6. Maintenance Window: Schedule during low-traffic periods

Conclusion

GCP Cloud SQL delivers a reliable managed relational database solution with built-in HA and deep VPC integration. With proper instance sizing, Auth Proxy, and security configuration, you can build a production-grade database architecture.

Duoyun Cloud is an authorized GCP partner offering exclusive channel discounts on Cloud SQL and other GCP resources. Purchase through Duoyun for better-than-list pricing, Chinese-language technical consultants, and a unified multi-cloud management platform. Visit duoyun.io for discount details.