Transactions
Creating and using transactions
Transactions are created using DataSource or EntityManager.
Examples:
await myDataSource.transaction(async (transactionalEntityManager) => {
// execute queries using transactionalEntityManager
})
or
await myDataSource.manager.transaction(async (transactionalEntityManager) => {
// execute queries using transactionalEntityManager
})
Everything you want to run in a transaction must be executed in a callback:
await myDataSource.manager.transaction(async (transactionalEntityManager) => {
await transactionalEntityManager.save(users)
await transactionalEntityManager.save(photos)
// ...
})
The most important restriction when working in a transaction is to ALWAYS use the provided instance of entity manager -
transactionalEntityManager in this example. DO NOT USE GLOBAL ENTITY MANAGER.
All operations MUST be executed using the provided transactional entity manager.
Specifying isolation level
Specifying the isolation level for the transaction can be done by supplying it as the first parameter:
await myDataSource.manager.transaction(
"SERIALIZABLE",
(transactionalEntityManager) => {},
)
Supported isolation levels
Isolation level implementations are not agnostic across all databases. Each driver declares which levels it supports, and TypeORM will throw an error if you request an unsupported level.
CockroachDB
READ COMMITTED— requires the cluster settingsql.txn.read_committed_isolation.enabled, which is enabled by default in recent versionsSERIALIZABLE(default)
CockroachDB does not implement weaker isolation levels at all. When you request READ UNCOMMITTED or REPEATABLE READ via SQL (e.g. BEGIN TRANSACTION ISOLATION LEVEL READ UNCOMMITTED), CockroachDB accepts the syntax, silently ignores the requested weakness, and runs the transaction at a stricter level instead — no warning, no error:
READ UNCOMMITTED→ runs asREAD COMMITTEDREPEATABLE READ→ runs asSERIALIZABLE
Google Spanner
REPEATABLE READ*SERIALIZABLE
* REPEATABLE READ on Spanner is currently in Preview and not yet generally available. It provides snapshot isolation and permits write-skew anomalies — use SELECT ... FOR UPDATE when that matters.
MariaDB
READ UNCOMMITTEDREAD COMMITTEDREPEATABLE READSERIALIZABLE
MS SQL Server
READ UNCOMMITTEDREAD COMMITTEDREPEATABLE READSERIALIZABLESNAPSHOT
SQL Server also supports driver-specific options.isolationLevel and options.connectionIsolationLevel settings, but these are subject to an upstream pool limitation. The top-level isolationLevel option covered on this page is not affected, because it is applied explicitly on each transaction.
MySQL
READ UNCOMMITTEDREAD COMMITTEDREPEATABLE READSERIALIZABLE
Isolation levels are not supported on the aurora-mysql driver. Requesting any level throws a validation error.
This is a transport limitation, not an engine limitation. Aurora MySQL the database engine fully supports the standard set, but the aurora-mysql driver talks to the cluster over the stateless RDS Data API instead of a persistent MySQL protocol connection, and the Data API provides no way to attach an isolation level to a transaction:
BeginTransactionaccepts onlyresourceArn,secretArn,database, andschema— there is no isolation parameter.- The Data API pools backend connections opaquely. A
SET TRANSACTION ISOLATION LEVEL ...sent as a separateExecuteStatementbeforeBeginTransactionhas no guaranteed affinity to the backend session that the transaction will run on, so the setting is silently dropped. - Multi-statement SQL is not supported, so
SET TRANSACTION ...; START TRANSACTION;cannot be sent as a single call either. - MySQL rejects
SET TRANSACTION ISOLATION LEVELinside an already-started transaction with error 1568, so the approach used for Aurora PostgreSQL (issuingSETas the first statement inside the started transaction) is not available on MySQL.
mysql driverIf per-transaction isolation levels are needed against an Aurora MySQL cluster, use the standard mysql driver pointed at the cluster writer endpoint instead of aurora-mysql. That path uses a regular MySQL protocol connection (via mysql2) and supports the full set of isolation levels listed above.
PostgreSQL
READ UNCOMMITTEDREAD COMMITTEDREPEATABLE READSERIALIZABLE
The same set of isolation levels is supported by the aurora-postgres driver, which talks to Aurora PostgreSQL clusters through the RDS Data API.
SAP HANA
READ COMMITTEDREPEATABLE READSERIALIZABLE
SQLite
READ UNCOMMITTED— only takes effect when shared-cache mode is enabled; in the default mode SQLite always usesSERIALIZABLEregardlessSERIALIZABLE
Default isolation level
You can configure a default isolation level for all transactions by setting isolationLevel in the DataSource options:
const dataSource = new DataSource({
type: "postgres",
isolationLevel: "SERIALIZABLE",
// ...
})
When set, all transactions started without an explicit isolation level will use this default. An explicit isolation level passed to transaction() or startTransaction() will override the default.
Using QueryRunner to create and control state of single database connection
QueryRunner provides a single database connection.
Transactions are organized using query runners.
Single transactions can only be established on a single query runner.
You can manually create a query runner instance and use it to manually control transaction state.
Example:
// create a new query runner
const queryRunner = dataSource.createQueryRunner()
// establish real database connection using our new query runner
await queryRunner.connect()
// now we can execute any queries on a query runner, for example:
await queryRunner.query("SELECT * FROM users")
// we can also access entity manager that works with connection created by a query runner:
const users = await queryRunner.manager.find(User)
// lets now open a new transaction:
await queryRunner.startTransaction()
try {
// execute some operations on this transaction:
await queryRunner.manager.save(user1)
await queryRunner.manager.save(user2)
await queryRunner.manager.save(photos)
// commit transaction now:
await queryRunner.commitTransaction()
} catch (err) {
// since we have errors let's rollback changes we made
await queryRunner.rollbackTransaction()
} finally {
// you need to release query runner which is manually created:
await queryRunner.release()
}
There are 3 methods to control transactions in QueryRunner:
startTransaction- starts a new transaction inside the query runner instance.commitTransaction- commits all changes made using the query runner instance.rollbackTransaction- rolls all changes made using the query runner instance back.
Learn more about Query Runner.