Drizzle Orm

Drizzle ORM patterns

Assume Drizzle 1.0.0-beta.1+ and Relational Query Builder v2 unless the installed package or codebase explicitly proves otherwise. In this model, db.query.* is the v2 API and relations are defined with defineRelations.

For formatting of relational queries and SQL builder chains, especially when a query mixes columns, where, and with or builder helpers like .select() and .where(), read references/formatting.md. Use that layout consistently in generated examples and code review suggestions because deep Drizzle configs get hard to scan when everything is packed onto a few lines.

Version cues

Treat these as current Drizzle v1 beta patterns:

• Use defineRelations(schema, (relationBuilder) => ...) from drizzle-orm.
• Pass the resulting relations object to drizzle(client, { relations }).
• Use db.query.table.findMany() and db.query.table.findFirst() for RQB v2.
• Use object-shaped where and orderBy in relational queries.

Do not add compatibility branches unless the current project has a reproduced need for them. Prefer keeping new code in the v2 shape.

Driver caveats

Drizzle query advice depends not only on query shape, but also on the database driver in use. Before suggesting transactions or multi-step write fixes, check which client the code is actually using.

Neon HTTP vs WebSocket

If the code uses drizzle-orm/neon-http, do not suggest a normal db.transaction(async (tx) => ...) fix. The Neon HTTP driver does not support Drizzle transaction callbacks, so advice that assumes transactional writes is incorrect for that client.

If the code uses drizzle-orm/neon-serverless with a WebSocket or Pool client, transactions are supported and db.transaction(...) is a valid option.

Review guidance for multi-step writes

When a handler performs multiple writes that must succeed or fail together, first identify the client:

• If it is an HTTP Neon client, call out the atomicity issue, but do not recommend db.transaction(...) on that same client without verifying a supported transactional path.
• If it is a WebSocket/serverless client with transaction support, recommending a transaction is appropriate.

This is especially important during code review: avoid suggesting fixes that the current driver cannot execute.

Relations v2 setup

Define relations in one place with defineRelations. The relationBuilder callback parameter exposes schema tables plus relation helpers such as one, many, and through.

import { defineRelations } from 'drizzle-orm'
import { drizzle } from 'drizzle-orm/node-postgres'
import * as schema from './schema'

export const relations = defineRelations(schema, (relationBuilder) => ({
  posts: {
    author: relationBuilder.one.users({
      from: relationBuilder.posts.authorId,
      to: relationBuilder.users.id
    })
  },

  users: {
    posts: relationBuilder.many.posts()
  }
}))

export const db = drizzle(client, { relations })

Use these relation options only when they solve a real schema need:

• from and to connect local relation columns to target relation columns.
• alias is useful when two tables have multiple relations between them.
• optional: false makes the relation required at the type level; use it only when the related row is guaranteed to exist.
• where in relation definitions creates predefined relation filters. It can filter only columns on the target to table.

Many-to-many through junction tables

Use through to expose the target table directly instead of querying a junction table and mapping it out manually.

export const relations = defineRelations(schema, (relationBuilder) => ({
  users: {
    groups: relationBuilder.many.groups({
      from: relationBuilder.users.id.through(
        relationBuilder.usersToGroups.userId
      ),

      to: relationBuilder.groups.id.through(
        relationBuilder.usersToGroups.groupId
      )
    })
  },

  groups: {
    participants: relationBuilder.many.users()
  }
}))

const users = await db.query.users.findMany({
  with: {
    groups: true
  }
})

Predefined relation filters

Use predefined filters when a named relation always needs the same target-table filter.

export const relations = defineRelations(schema, (relationBuilder) => ({
  groups: {
    verifiedUsers: relationBuilder.many.users({
      from: relationBuilder.groups.id.through(
        relationBuilder.usersToGroups.groupId
      ),

      to: relationBuilder.users.id.through(
        relationBuilder.usersToGroups.userId
      ),

      where: {
        verified: true
      }
    })
  }
}))

const groups = await db.query.groups.findMany({
  with: {
    verifiedUsers: true
  }
})

Relational API vs SQL builder

Drizzle has two distinct query APIs. Choosing the wrong one causes TypeScript errors or forces awkward query shapes.

Use the relational API (db.query.table.findFirst/findMany) when:

• Fetching a single record or a simple list.
• Loading nested relations in one query.
• Filtering with object-shaped RQB v2 filters.
• Using relation filters, predefined relation filters, nested limit, or nested offset.

Use the SQL builder (db.select().from().where()) when:

• Building arbitrary WHERE conditions dynamically at runtime.
• Running aggregations (count(), sum(), etc.).
• The query needs explicit joins or selected fields from unrelated tables.

Relational query builder v2

where takes an object

Use object-shaped filters in RQB v2. A plain property value means equality; use operator objects for non-equality conditions.

const item = await db.query.items.findFirst({
  where: {
    id
  }
})

const approved = await db.query.items.findMany({
  where: {
    status: 'approved',

    price: {
      gt: 0
    }
  }
})

Do not pass SQL helper expressions directly to relational where.

// Wrong in RQB v2: use this shape in the SQL builder instead.
const item = await db.query.items.findFirst({
  where: eq(items.id, id)
})

Complex filters

RQB v2 filters support AND, OR, NOT, column operators, RAW, and relation filters.

import { sql } from 'drizzle-orm'

const users = await db.query.users.findMany({
  where: {
    AND: [
      {
        id: {
          gt: 10
        }
      },
      {
        OR: [
          {
            email: {
              ilike: '%@example.com'
            }
          },
          {
            RAW: (table) => sql`${table.lastLogin} is not null`
          }
        ]
      },
      {
        posts: {
          title: {
            like: 'Release%'
          }
        }
      }
    ]
  }
})

Use RAW only for cases the object operators cannot express, and still route user input through parameterized sql expressions.

Select columns and load nested relations

Use columns to restrict selected fields and with to load related rows.

const brand = await db.query.brands.findFirst({
  columns: {
    id: true,
    name: true,
    slug: true
  },

  where: {
    slug
  },

  with: {
    items: {
      columns: {
        id: true,
        name: true
      },

      where: {
        status: 'approved'
      },

      orderBy: {
        id: 'asc'
      },

      limit: 10,
      offset: 0,

      with: {
        category: {
          columns: {
            name: true,
            slug: true
          }
        }
      }
    }
  }
})

When columns contains both true and false, the false entries are redundant because the included true fields already define the result shape.

Order related data

Prefer object-shaped orderBy for normal ordering. Use callback/custom SQL ordering only when the object form cannot express the requirement.

const posts = await db.query.posts.findMany({
  orderBy: {
    createdAt: 'desc'
  },

  with: {
    comments: {
      orderBy: {
        id: 'asc'
      }
    }
  }
})

Add computed fields with extras

Use extras for per-row computed fields. Do not put aggregations in extras; use core queries or an explicit supported subquery pattern such as db.$count when the project already uses it.

const posts = await db.query.posts.findMany({
  extras: {
    contentLength: (table, { sql }) => sql<number>`length(${table.content})`
  },

  with: {
    comments: {
      extras: {
        commentLength: (table, { sql }) => sql<number>`length(${table.content})`
      }
    }
  }
})

SQL builder

Dynamic WHERE conditions

Build conditions into an array, then spread into and():

import { and, eq, ilike } from 'drizzle-orm'

const conditions = [
  eq(items.status, 'approved')
]

if (categoryId) {
  conditions.push(eq(items.categoryId, categoryId))
}

if (search) {
  const escaped = search
    .replaceAll('%', String.raw`\%`)
    .replaceAll('_', String.raw`\_`)

  conditions.push(
    ilike(items.name, `%${escaped}%`)
  )
}

const results = await db
  .select({
    id: items.id,
    name: items.name
  })
  .from(items)
  .where(and(...conditions))
  .limit(limit)
  .offset((page - 1) * limit)

Always escape user input before passing to ilike() because % and _ are wildcards in SQL LIKE patterns.

Count query

count() returns exactly one row at runtime, but with noUncheckedIndexedAccess TypeScript still treats indexed access as possibly undefined. Avoid array destructuring here and read the first row safely.

import { count } from 'drizzle-orm'

const countRows = await db
  .select({ total: count() })
  .from(items)
  .where(
    and(...conditions)
  )

const total = countRows[0]?.total ?? 0

Parallel data and count queries

Run independent queries with Promise.all to avoid sequential round-trips.

const [rows, countRows] = await Promise.all([
  db
    .select()
    .from(items)
    .where(
      and(...conditions)
    )
    .limit(limit)
    .offset(offset),

  db
    .select({ total: count() })
    .from(items)
    .where(
      and(...conditions)
    )
])

const total = countRows[0]?.total ?? 0

return {
  items: rows,
  total,
  page,
  limit
}

Joins in the SQL builder

import { eq } from 'drizzle-orm'

const results = await db
  .select({
    brandName: brands.name,
    id: items.id,
    name: items.name
  })
  .from(items)
  .innerJoin(brands, eq(items.brandId, brands.id))
  .where(
    eq(items.status, 'approved')
  )