Configuration Keys

Configuration is loaded from the Flask app.config when SQLAlchemy.init_app() is called. The configuration is not read again after that. Therefore, all configuration must happen before initializing the application.


The database connection URI used for the default engine. It can be either a string or a SQLAlchemy URL instance. See below and Engine Configuration for examples.

At least one of this and SQLALCHEMY_BINDS must be set.


Changed in version 3.0: No longer defaults to an in-memory SQLite database if not set.


A dict of arguments to pass to sqlalchemy.create_engine() for the default engine.

This takes precedence over the engine_options argument to SQLAlchemy, which can be used to set default options for all engines.


Changed in version 3.0: Only applies to the default bind.

Added in version 2.4.


A dict mapping bind keys to engine options. The value can be a string or a SQLAlchemy URL instance. Or it can be a dict of arguments, including the url key, that will be passed to sqlalchemy.create_engine(). The None key can be used to configure the default bind, but SQLALCHEMY_ENGINE_OPTIONS and SQLALCHEMY_DATABASE_URI take precedence.

At least one of this and SQLALCHEMY_DATABASE_URI must be set.


Added in version 0.12.


The default value for echo and echo_pool for every engine. This is useful to quickly debug the connections and queries issued from SQLAlchemy.


Changed in version 3.0: Sets echo_pool in addition to echo.


If enabled, information about each query during a request will be recorded. Use get_recorded_queries() to get a list of queries that were issued during the request.


Changed in version 3.0: Not enabled automatically in debug or testing mode.


If enabled, all insert, update, and delete operations on models are recorded, then sent in models_committed and before_models_committed signals when session.commit() is called.

This adds a significant amount of overhead to every session. Prefer using SQLAlchemy’s ORM Events directly for the exact information you need.


Changed in version 3.0: Disabled by default.

Added in version 2.0.

Changed in version 3.1: Removed SQLALCHEMY_COMMIT_ON_TEARDOWN.



Connection URL Format

See SQLAlchemy’s documentation on Engine Configuration for a complete description of syntax, dialects, and options.

A basic database connection URL uses the following format. Username, password, host, and port are optional depending on the database type and configuration.


Here are some example connection strings:

# SQLite, relative to Flask instance path

# PostgreSQL

# MySQL / MariaDB

SQLite does not use a user or host, so its URLs always start with _three_ slashes instead of two. The dbname value is a file path. Absolute paths start with a _fourth_ slash (on Linux or Mac). Relative paths are relative to the Flask application’s instance_path.

Default Driver Options

Some default options are set for SQLite and MySQL engines to make them more usable by default in web applications.

SQLite relative file paths are relative to the Flask instance path instead of the current working directory. In-memory databases use a static pool and check_same_thread to work across requests.

MySQL (and MariaDB) servers are configured to drop connections that have been idle for 8 hours, which can result in an error like 2013: Lost connection to MySQL server during query. A default pool_recycle value of 2 hours (7200 seconds) is used to recreate connections before that timeout.

Engine Configuration Precedence

Because Flask-SQLAlchemy has support for multiple engines, there are rules for which config overrides other config. Most applications will only have a single database and only need to use SQLALCHEMY_DATABASE_URI and SQLALCHEMY_ENGINE_OPTIONS.


Certain databases may be configured to close inactive connections after a period of time. MySQL and MariaDB are configured for this by default, but database services may also configure this type of limit. This can result in an error like 2013: Lost connection to MySQL server during query.

If you encounter this error, try setting pool_recycle in the engine options to a value less than the database’s timeout.

Alternatively, you can try setting pool_pre_ping if you expect the database to close connections often, such as if it’s running in a container that may restart.

See SQAlchemy’s docs on dealing with disconnects for more information.