api-database-drizzle（API 数据库 Drizzle）

Database with Drizzle ORM + Neon

Quick Guide: Use Drizzle ORM for type-safe queries, Neon serverless Postgres for edge-compatible connections. Schema-first design with automatic TypeScript types. Use RQB v2 with defineRelations() and object-based where syntax. Relational queries with .with() avoid N+1 problems. Use transactions for atomic operations.

<critical_requirements>

CRITICAL: Before Using This Skill

All code must follow project conventions in CLAUDE.md (kebab-case, named exports, import ordering, import type, named constants)

(You MUST set casing: 'snake_case' in Drizzle config to map camelCase JS to snake_case SQL)

(You MUST use tx parameter (NOT db) inside transaction callbacks to ensure atomicity)

(You MUST use .with() for relational queries to avoid N+1 problems - fetches all data in single SQL query)

(You MUST use defineRelations() for RQB v2 - the old relations() per-table syntax is deprecated)

</critical_requirements>

Detailed Resources:

• For code examples, see examples/ folder:

- core.md - Connection setup and schema definition (always loaded) - queries.md - Relational queries and query builder - relations-v2.md - RQB v2 with defineRelations() (NEW) - transactions.md - Atomic operations - migrations.md - Drizzle Kit workflow - seeding.md - Development data population (includes drizzle-seed)

• For decision frameworks and anti-patterns, see reference.md

Auto-detection: drizzle-orm, @neondatabase/serverless, neon-http, db.query, db.transaction, drizzle-kit, pgTable, defineRelations, drizzle-seed

When to use:

• Serverless functions needing type-safe database queries
• Schema-first development with migrations
• Building server-rendered apps with API routes

When NOT to use:

• Simple apps using framework server actions directly (overhead not justified)
• Apps needing traditional TCP connection pooling only (use standard Postgres clients)
• Non-TypeScript projects (lose primary benefit of type safety)
• Edge functions requiring WebSocket connections (not supported in edge runtime)

Core Patterns

Pattern 1: Database Connection (Neon HTTP)

Configure Drizzle with Neon for serverless/edge compatibility. Key setup requirements:

export const db = drizzle(sql, {
  schema,
  casing: "snake_case", // Maps camelCase JS to snake_case SQL
});

• Validate DATABASE_URL before use (throw on missing)
• Always set casing: "snake_case" to prevent field name mismatches
• Use neon() for HTTP (edge-compatible) or Pool for WebSocket (long queries)

Full connection setup, WebSocket config, and Drizzle Kit config in examples/core.md.

Pattern 2: Schema Definition

Define tables with TypeScript types using Drizzle's schema builder:

export const companies = pgTable("companies", {
  id: uuid("id").primaryKey().defaultRandom(),
  name: varchar("name", { length: 255 }).notNull(),
  slug: varchar("slug", { length: 255 }).unique(),
  deletedAt: timestamp("deleted_at"), // Soft delete
  createdAt: timestamp("created_at").defaultNow(),
});

• Use pgEnum() for constrained values instead of varchar
• Always include createdAt/updatedAt timestamps
• Add deletedAt for soft deletes
• Set onDelete: "cascade" on foreign keys to prevent orphaned records
• Use uuid().defaultRandom() or integer().generatedAlwaysAsIdentity() for primary keys

Full schema examples (enums, relations, junction tables, identity columns) in examples/core.md.

Pattern 3: Relational Queries with .with()

Fetch related data efficiently in a single SQL query using .with():

const job = await db.query.jobs.findFirst({
  where: and(eq(jobs.id, jobId), isNull(jobs.deletedAt)),
  with: {
    company: { with: { locations: true } },
    jobSkills: { with: { skill: true } },
  },
});
// Result is fully typed: job.company.name, job.jobSkills[0].skill.name

• Use db.query with .with() when fetching related data -- single SQL query, no N+1
• Use query builder (db.select()) for custom column selection, complex JOINs, aggregations
• Always include isNull(deletedAt) in WHERE conditions for soft-deleted tables

Full relational query examples, N+1 anti-patterns, and dynamic filtering in examples/queries.md.

Additional Patterns

The following patterns are documented with full examples in examples/:

• Query Builder - Complex filters, dynamic conditions, custom JOINs - see queries.md
• Transactions - Atomic operations, error handling, rollback - see transactions.md
• Database Migrations - Drizzle Kit workflow, generate vs push - see migrations.md
• Database Seeding - Development data, safe cleanup - see seeding.md

Performance optimization (indexes, prepared statements, pagination) is documented in reference.md.

<red_flags>

RED FLAGS

• ❌ Using db instead of tx inside transactions - Bypasses transaction context, breaking atomicity
• ❌ N+1 queries with relations - Use .with() to fetch in one query
• ❌ Not setting casing: 'snake_case' - Field name mismatches between JS and SQL
• ❌ Using v1 relations() per-table syntax - Deprecated, use defineRelations()
• ❌ Using callback-based where/orderBy - v1 syntax deprecated, use object-based syntax
• ⚠️ Queries without soft delete checks (isNull(deletedAt))
• ⚠️ No pagination limits on list queries

Gotchas & Edge Cases:

• Neon HTTP has 30-second query timeout - long queries need WebSocket
• Prepared statements created outside transactions cannot be used inside transactions
• enableRLS() deprecated in v1.0.0-beta.1 - use pgTable.withRLS() instead
• Validator packages consolidated: drizzle-zod is now drizzle-orm/zod (since v1 beta)

For the complete list of anti-patterns and gotchas, see reference.md.

</red_flags>

<critical_reminders>

CRITICAL REMINDERS

All code must follow project conventions in CLAUDE.md

(You MUST set casing: 'snake_case' in Drizzle config to map camelCase JS to snake_case SQL)

(You MUST use tx parameter (NOT db) inside transaction callbacks to ensure atomicity)

(You MUST use .with() for relational queries to avoid N+1 problems - fetches all data in single SQL query)

(You MUST use defineRelations() for RQB v2 - the old relations() per-table syntax is deprecated)

Failure to follow these rules will cause field name mismatches, break transaction atomicity, create N+1 performance issues, and use deprecated APIs.

</critical_reminders>