# For Drizzle ORM
npx drizzle-kit generate  # Generate migrations from schema
npx drizzle-kit migrate   # Apply migrations

This is Day 8 of Building SaaS Solo - Design, Implementation, and Operation Advent Calendar 2025.

Yesterday I wrote about "Database ID Design." Today I’ll explain how to manage database migrations.

::message The approach described in this article is something I arrived at through trial and error. If you know a better way, please let me know in the comments.

::

🎯 Common Migration Management Approaches

There are several ways to manage database migrations.

Using ORM Migration Features

ORMs like Drizzle ORM, Prisma, and TypeORM have built-in migration features.

# For Drizzle ORM
npx drizzle-kit generate  # Generate migrations from schema
npx drizzle-kit migrate   # Apply migrations

For example, when you write table definitions in TypeScript, it detects changes and automatically generates SQL like ALTER TABLE.

Benefits:

• Automatically generates migration SQL from code changes
• Manages application history in database tables
• Can be applied with a single command

Challenges:

• Difficult to handle complex data migrations (like transforming existing data)
• Sometimes hard to understand what will be executed

Challenges When Managing Multiple Environments

ORM migration features are convenient for single environment management, but when operating separate development and production environments, there are challenges:

• Schema can get out of sync between dev and production
• Hard to track "applied to dev but not yet to production" state
• Difficult to know what was applied when

I faced these same challenges while developing Memoreru, my indie project, and through trial and error arrived at my current operational rules.

📂 Migration Management for Indie Projects

Sequential File Management

Migration files are created by Claude Code and managed with sequential numbers.

database/migrations/
├── sql/
│   ├── 001_create_users_table.sql
│   ├── 002_create_posts_table.sql
│   ├── 003_add_user_profile.sql
│   ├── 004_add_status_column.sql
│   └── ...
├── scripts/
│   └── migrate.sh
├── status.json
└── README.md

Benefits of sequential management:

• Application order is clear at a glance
• File names show which point in time the schema represents
• Easy to identify differences between production and development

Why Manage SQL Files Directly

Instead of using Drizzle ORM’s migration generation feature (drizzle-kit generate), I create SQL files directly. However, schema definitions themselves are managed with Drizzle ORM, so type safety is maintained.

Reasons for managing SQL files directly:

• Easier to handle complex changes (those involving data migration)
• Complete understanding of what will be executed
• Easier to identify causes when troubleshooting

🔄 Shared Migration Script for Dev and Production

In Memoreru, I use the same script to apply migrations in both development and production environments.

Why a Shared Script

# Development environment
./database/migrations/scripts/migrate.sh dev 004_add_status_column.sql

# Production environment
./database/migrations/scripts/migrate.sh pro 004_add_status_column.sql

Benefits of a shared script:

1. Rehearsal effect: By following the same procedure in dev as production, you can discover problems before applying to production
2. Unified procedure: Having different approaches (Claude Code directly in dev, script in production) leads to accidents
3. Centralized logging: Execution logs from both environments remain in the same format

Environment-Specific Differences

Item	Development	Production
Connection info	Auto-loaded from .env.local	Manual input each time
Backup recommendation	None	Warning displayed

Entering the production connection string each time is tedious, but it functions as a safety measure. It prevents accidents where you think you’re in dev but accidentally operate on production.

Additionally, I don’t give Claude Code the production DB connection information. This eliminates the risk of AI accidentally operating on the production database.

Also, I take backups using pgAdmin’s backup feature before applying migrations in both environments. It’s important to prepare for potential rollbacks.

🛡️ Safety Mechanisms

The script includes the following safety measures:

• Confirmation flow: Displays confirmation prompts before applying to prevent mistakes
• Connection test: Verifies DB connection before applying
• Automatic log saving: Saves all execution logs to logs/migrations/ for later review

The specific script implementation was created by Claude Code. Just tell it your requirements, and it will generate a script suited to your environment.

📊 Centralized Status Management with status.json

I manage the application status of both dev and production environments in a single file.

{
"lastUpdated": "2025-12-04",
"environments": {
"dev": {
"name": "Development",
"lastApplied": "004_add_status_column",
"appliedAt": "2025-12-04"
},
"pro": {
"name": "Production",
"lastApplied": "003_add_user_profile",
"appliedAt": "2025-11-30"
}
},
"pending": {
"pro": ["004_add_status_column"]
}
}

Checking Pending Production Migrations

# Display pending list
jq '.pending.pro' database/migrations/status.json
# => ["004_add_status_column"]

You can see at a glance which migrations have been applied to dev but not yet to production.

Automatic Updates

After applying migrations, the script automatically updates status.json. No need to update manually, preventing forgotten updates.

💡 Practical Tips

Tip 1: Make Destructive Changes Gradually

Execute column name changes or table structure changes in stages, not all at once.

-- Step 1: Add new column
ALTER TABLE contents ADD COLUMN new_name TEXT;

-- Step 2: Migrate data
UPDATE contents SET new_name = old_name;

-- Step 3: Drop old column (in a separate migration)
ALTER TABLE contents DROP COLUMN old_name;

By checking application behavior between steps 2 and 3, you can minimize impact even if problems occur.

Tip 2: Prepare Rollback SQL

For important migrations, leave rollback SQL as comments.

-- Migration
ALTER TABLE contents ADD COLUMN status TEXT DEFAULT 'draft';

-- Rollback (execute only when needed)
-- ALTER TABLE contents DROP COLUMN status;

Tip 3: Collaboration Rules with Claude Code

I document migration operation rules in CLAUDE.md.

## Migration Operations

- Don't execute SQL directly with psql
- Always apply via migrate.sh script
- Rehearse in dev environment before production
- Commit status.json after applying

This prevents AI agents from accidentally executing SQL directly.

✅ Summary

Here’s what I’ve learned from database migration operations.

What’s working well:

• Sequential file management for chronological tracking
• Rehearsals with shared dev/production script
• Centralized status management with status.json
• Confirmation flow to prevent mistakes

Things to be careful about:

• Manual SQL management can become difficult as change volume increases
• Test complex data migrations with test data beforehand
• Think through rollback procedures in advance

Even for indie development, establishing rules from the start prevents problems later.

Tomorrow I’ll explain "Migrating from NextAuth.js to Better Auth: A Practical Guide."

---

Other articles in this series

• 12/7: Database ID Design: Choosing ID Methods and Primary Key Strategies
• 12/9: Migrating from NextAuth.js to Better Auth: A Practical Guide