Skip to main content
Version: 6.2

Upgrading from v3 to v4

Following sections describe (hopefully) all breaking changes, most of them might be not valid for you, like if you do not use custom NamingStrategy implementation, you do not care about the interface being changed.

Node 10.13.0+ required

Support for older node versions was dropped.

TypeScript 3.7+ required

Support for older TypeScript versions was dropped.


The ORM has been split into several packages. In v4 one needs to require @mikro-orm/core and a driver package, e.g. @mikro-orm/mysql. This driver package already contains the mysql2 dependency, so you can remove that from your package.json.

  • @mikro-orm/core
  • @mikro-orm/reflection - TsMorphMetadataProvider
  • @mikro-orm/cli - CLI support, requires entity-generator, migrator and knex
  • @mikro-orm/knex - SQL support
  • @mikro-orm/entity-generator
  • @mikro-orm/migrations
  • @mikro-orm/mysql
  • @mikro-orm/mariadb
  • @mikro-orm/mysql-base - Common implementation for mysql and mariadb (internal)
  • @mikro-orm/sqlite
  • @mikro-orm/postgresql
  • @mikro-orm/mongodb

For easier transition, meta package mikro-orm is still present, reexporting core, reflection, migrations, entity-generator and cli packages. You should not install both mikro-orm and @mikro-orm/core packages together.

You should prefer the @mikro-orm/core over mikro-orm package, there were weird dependency issues reported with the mikro-orm meta-package.

Default metadata provider is now ReflectMetadataProvider

If you want to use ts-morph, you need to install @mikro-orm/reflection package and enable the provider explicitly in the ORM config, as described here.

import { TsMorphMetadataProvider } from '@mikro-orm/reflection';

await MikroORM.init({
metadataProvider: TsMorphMetadataProvider,
// ...

Using ReflectMetadataProvider has some limitations, so be sure to read about them here.

One common gotcha with reflect-metadata is that you need to explicitly state the property type when using property initializer:

createdAt: Date = new Date();

Without the explicit type, we would infer Object instead of Date, which would be most probably mapped to JSON column type (depends on driver).

SqlEntityManager and MongoEntityManager

In v4 the core package, where EntityManager and EntityRepository are defined, is not dependent on knex, and therefore it cannot have a method returning a QueryBuilder. You need to import the SQL flavour of the EM from the driver package to access the createQueryBuilder() method.

The SQL flavour of EM is actually called SqlEntityManager, it is exported both under this name and under EntityManager alias, so you can just change the location from where you import.

import { EntityManager } from '@mikro-orm/mysql'; // or any other SQL driver package

const em: EntityManager;
const qb = await em.createQueryBuilder(...);

Same applies for the aggregate() method in mongo driver:

import { EntityManager } from '@mikro-orm/mongodb';

const em: EntityManager;
const ret = await em.aggregate(...);

The mongo flavour of EM is actually called MongoEntityManager, it is exported both under this name and under EntityManager alias, so you can just change the location from where you import.

Different default pivotTable

Implementation of UnderscoreNamingStrategy and EntityCaseNamingStrategy joinTableName() method has changed. You can use pivotTable on the owning side of M:N relation to specify the table name manually.

Previously the table name did not respect property name, if one defined multiple M:N relations between same entities, there were conflicts and one would have to specify pivotTable name manually at least on one of them. With the new way, we can be sure that the table name won't conflict with other pivot tables.

Previously the name was constructed from 2 entity names as entity_a_to_entity_b, ignoring the actual property name. In v4 the name will be entity_a_coll_name in case of the collection property on the owning side being named collName.

Changes in folder-based discovery (entitiesDirs removed)

entitiesDirs and entitiesDirsTs were removed in favour of entities and entitiesTs, entities will be used as a default for entitiesTs (that is used when we detect ts-node).

entities can now contain mixture of paths to directories, globs pointing to entities, or references to the entities or instances of EntitySchema.

This basically means that all you need to change is renaming entitiesDirs to entities.

entities: ['dist/**/entities', 'dist/**/*.entity.js', FooBar, FooBaz],
entitiesTs: ['src/**/entities', 'src/**/*.entity.ts', FooBar, FooBaz],

Changes in wrap() helper, WrappedEntity interface and Reference wrapper

Previously all the methods and properties of WrappedEntity interface were added to the entity prototype during discovery. In v4 there is only one property added: __helper: WrappedEntity. WrappedEntity has been converted to actual class.

wrap(entity) no longer returns the entity, now the WrappedEntity instance is being returned. It contains only public methods (init, assign, isInitialized, ...), if you want to access internal properties like __meta or __em, you need to explicitly ask for the helper via wrap(entity, true).

Internal methods (with __ prefix) were also removed from the Reference class, use wrap(ref, true) to access them.

Instead of interface merging with WrappedEntity, one can now use classic inheritance, by extending BaseEntity exported from @mikro-orm/core. If you do so, wrap(entity) will return your entity.

Removed flush parameter from persist() and remove() methods

flush param is removed, both persist and remove methods are synchronous and require explicit flushing, possibly via fluent interface call.

// before
await em.persist(jon, true);
await em.remove(Author, jon, true);

// after
await em.persist(jon).flush();
await em.remove(jon).flush();

remove() method requires entity instances

The em.remove() method originally allowed to pass either entity instance, or a condition. When one passed a condition, it was firing a native delete query, without handling transactions or hooks.

In v4, the method is now simplified and works only with entity instances. Use em.nativeDelete() explicitly if you want to fire a delete query instead of letting the UnitOfWork doing its job.

// before
await em.remove(Author, 1); // fires query directly

// after
await em.nativeDelete(Author, 1);

em.removeEntity() has been removed in favour of em.remove() (that now has almost the same signature).

Type safe references

EM now returns Loaded<T, P> instead of the entity (T). This type automatically adds synchronous method get() that returns the entity (for references) or array of entities (for collections).

Reference.get() is now available only with correct Loaded type hint and is used as a sync getter for the entity, just like unwrap(). You can use Reference.load(prop) for the original get() method functionality.

em.find() and similar methods now have two type arguments, due to TypeScript not supporting partial type inference, when you specify the T explicitly (without also explicitly specifying the load hint), the inference will not work. This use case was mainly for usage without classes (interfaces + EntitySchema) - in that case it is now supported to pass actual instance of EntitySchema as the first parameter to these methods, that will allow correct type inference:

const author = await em.findOne(AuthorSchema, { ... }, ['books']);
console.log(author.books.get()); // `get()` is now inferred correctly

Custom types are now type safe

Generic Type class has now two type arguments - the input and output types. Input type defaults to string, output type defaults to the input type.

You might need to explicitly provide the types if your methods are strictly typed.

Custom type serialization

Custom types used to be serialized to the database value. In v4, the runtime value is used by default. Implement custom toJSON() method if you need to customize this.

Property default and defaultRaw

Previously the default option of properties was used as is, so we had to wrap strings in quotes (e.g. @Property({ default: "'foo bar'" })).

In v4 the default is typed as string | number | boolean | null and when used with string value, it will be automatically quoted.

To use SQL functions we now need to use defaultRaw: @Property({ defaultRaw: 'now()' }).

autoFlush option has been removed

Also persistLater() and removeLater() methods are deprecated. Use persist() or remove respectively.

IdEntity, UuidEntity and MongoEntity interfaces are removed

They were actually never needed.

MongoDB driver is no longer the default

You need to specify the platform type either via type option or provide the driver implementation via driver option.

Available platforms types: [ 'mongo', 'mysql', 'mariadb', 'postgresql', 'sqlite' ]

Removed configuration discovery.tsConfigPath

Removed as it is no longer needed, it was used only for TsMorphMetadataProvider, when the entitiesDirsTs were not explicitly provided. In v4, this is no longer needed, as ts-morph discovery will use d.ts files instead, that should be located next to the compiled entities.

Changes in query highlighting

Previously Highlight.js was used to highlight various things in the CLI, like SQL and mongo queries, or migrations or entities generated via CLI. While the library worked fine, it was causing performance issues mainly for those bundling via webpack and using lambdas, as the library was huge.

In v4 highlighting is disabled by default, and there are 2 highlighters you can optionally use (you need to install them first).

import { SqlHighlighter } from '@mikro-orm/sql-highlighter';

highlighter: new SqlHighlighter(),
// ...

For MongoDB you can use @mikro-orm/mongo-highlighter.