ORM (Sequelize) — Basics

What is an ORM

Maps relational rows ↔ object instances. Generates SQL from method calls (User.findAll, user.save). Handles connection pooling, type coercion, lifecycle hooks, transactions.

Sequelize at a glance

• Supports Postgres, MySQL/MariaDB, MSSQL, SQLite, Snowflake, DB2, Oracle.
• v6 = current stable; v7 alpha (full TS rewrite).
• Models defined via Model.init(attrs, options) or @Table decorator (legacy decorators).

Core concepts

• Model = JS class bound to a table.
• Attribute = column definition (type, allowNull, defaultValue, validate).
• Association = belongsTo, hasOne, hasMany, belongsToMany.
• Instance = single row in memory; has save(), destroy(), reload(), changed(), previous().
• Finder = findOne, findAll, findByPk, findOrCreate, findAndCountAll.
• Scope = named pre-canned where/include fragments.

Migration vs sync

• sync() reads model files, emits CREATE TABLE IF NOT EXISTS. Not versioned, not reversible. Dev only.
• sequelize-cli migrations = ordered files in migrations/, tracked in SequelizeMeta. Use in CI/prod.
• sync({ alter: true }) silently drops columns missing from models. Never run in prod.

Lifecycle hooks (quick reference)

Hook family	Fires
beforeValidate / afterValidate	Before / after model validation
beforeCreate / afterCreate	Single-row insert
beforeUpdate / afterUpdate	Single-row update
beforeDestroy / afterDestroy	Single-row delete
beforeSave / afterSave	Both create + update
beforeBulkCreate / beforeBulkUpdate / beforeBulkDestroy	Bulk ops (instance hooks skip!)

Pass individualHooks: true to force per-row hooks on bulk ops (expensive — loads all rows).

Quick gotchas

• bulkCreate skips beforeCreate by default. Password hashing in beforeCreate → plaintext in CSV imports.
• Raw queries (sequelize.query) bypass hooks, paranoid filters, default scopes.
• Forgetting { transaction: t } runs the query on a separate connection — silent atomicity break.
• paranoid: true doesn’t cascade. Orphan FKs after soft delete.
• unique index sees soft-deleted rows. Need partial unique WHERE deleted_at IS NULL (Postgres) or generated-column trick (MySQL).
• findAndCountAll with includes inflates count to joined row count. Use distinct: true.