Skip to main content
Version: Next

Unit of Work and Transactions

MikroORM uses the Identity Map pattern to track objects. Whenever you fetch an object from the database, MikroORM will keep a reference to this object inside its UnitOfWork.

This allows MikroORM room for optimizations. If you call the EntityManager and ask for an entity with a specific ID twice, it will return the same instance:

const authorRepository = orm.em.getRepository(Author);
const jon1 = await authorRepository.findOne(1);
const jon2 = await authorRepository.findOne(1);

// identity map in action
console.log(jon1 === jon2); // true

Only one SELECT query will be fired against the database here. In the second findOne() call MikroORM will check the identity map first and will skip the database round trip as it will find the entity already loaded.

The identity map being indexed by primary keys only allows shortcuts when you ask for objects by primary key. When you query by other properties, you will still get the same reference, but two separate database calls will be made:

const authorRepository = orm.em.getRepository(Author);
const jon1 = await authorRepository.findOne({ name: 'Jon Snow' });
const jon2 = await authorRepository.findOne({ name: 'Jon Snow' });

// identity map in action
console.log(jon1 === jon2); // true

MikroORM only knows objects by id, so a query for different criteria has to go to the database, even if it was executed just before. But instead of creating a second Author object MikroORM first gets the primary key from the row and checks if it already has an object inside the UnitOfWork with that primary key.

Persisting Managed Entities

The identity map has a second use-case. When you call em.flush(), MikroORM will ask the identity map for all objects that are currently managed. This means you don't have to call em.persist() over and over again to pass known objects to the EntityManager. This is a NO-OP for known entities, but leads to much code written that is confusing to other developers.

The following code WILL update your database with the changes made to the Author object, even if you did not call em.persist():

const authorRepository = orm.em.getRepository(Author);
const jon = await authorRepository.findOne(1);
jon.email = 'foo@bar.com';
await authorRepository.flush(); // calling orm.em.flush() has same effect

How MikroORM Detects Changes

MikroORM is a data-mapper that tries to achieve persistence-ignorance (PI). This means you map JS objects into a relational database that do not necessarily know about the database at all. A natural question would now be, "how does MikroORM even detect objects have changed?".

For this MikroORM keeps a second map inside the UnitOfWork. Whenever you fetch an object from the database MikroORM will keep a copy of all the properties and associations inside the UnitOfWork.

Now whenever you call em.flush() MikroORM will iterate over all entities you previously marked for persisting via em.persist(). For each object it will compare the original property and association values with the values that are currently set on the object. If changes are detected then the object is queued for a UPDATE operation. Only the fields that actually changed are updated.

Implicit Transactions

First and most important implication of having Unit of Work is that it allows handling transactions automatically.

When you call em.flush(), all computed changes are queried inside a database transaction (if supported by given driver). This means that you can control the boundaries of transactions simply by calling em.persist() and once all your changes are ready, simply calling flush() will run them inside a transaction.

You can also control the transaction boundaries manually via em.transactional(cb).

const user = await em.findOne(User, 1);
user.email = 'foo@bar.com';
const car = new Car();
user.cars.add(car);

// thanks to bi-directional cascading we only need to persist user entity
// flushing will create a transaction, insert new car and update user with new email
await em.persistAndFlush(user);

You can find more information about transactions in Transactions and concurrency page.

Flush Modes

The flushing strategy is given by the flushMode of the current running EntityManager.

  • FlushMode.COMMIT - The EntityManager tries to delay the flush until the current Transaction is committed, although it might flush prematurely too.
  • FlushMode.AUTO - This is the default mode, and it flushes the EntityManager only if necessary.
  • FlushMode.ALWAYS - Flushes the EntityManager before every query.

FlushMode.AUTO will try to detect changes on the entity we are querying, and flush if there is an overlap:

// querying for author will trigger auto-flush if we have new author persisted
const a1 = new Author(...);
orm.em.persist(a1);
const r1 = await orm.em.find(Author, {});

// querying author won't trigger auto-flush if we have new book, but no changes on author
const b4 = new Book(...);
orm.em.persist(b4);
const r2 = await orm.em.find(Author, {});

// but querying for book will trigger auto-flush
const r3 = await orm.em.find(Book, {});

We can set the flush mode on different places:

  • in the ORM config via Options.flushMode
  • for given EntityManager instance (and its forks) via em.setFlushMode()
  • for given EntityManager fork via em.fork({ flushMode })
  • for given QueryBuilder instance via qb.setFlushMode()
  • for given transaction scope via em.transactional(..., { flushMode })

This part of documentation is highly inspired by doctrine internals docs as the behaviour here is pretty much the same.