Session Basics

What does the Session do ?

In the most general sense, the Session establishes all conversations with the database and represents a “holding zone” for all the objects which you’ve loaded or associated with it during its lifespan. It provides the interface where SELECT and other queries are made that will return and modify ORM-mapped objects. The ORM objects themselves are maintained inside the Session, inside a structure called the identity map - a data structure that maintains unique copies of each object, where “unique” means “only one object with a particular primary key”.

The Session begins in a mostly stateless form. Once queries are issued or other objects are persisted with it, it requests a connection resource from an Engine that is associated with the Session, and then establishes a transaction on that connection. This transaction remains in effect until the Session is instructed to commit or roll back the transaction.

The ORM objects maintained by a Session are instrumented such that whenever an attribute or a collection is modified in the Python program, a change event is generated which is recorded by the Session. Whenever the database is about to be queried, or when the transaction is about to be committed, the Session first flushes all pending changes stored in memory to the database. This is known as the unit of work pattern.

When using a Session, it’s useful to consider the ORM mapped objects that it maintains as proxy objects to database rows, which are local to the transaction being held by the Session. In order to maintain the state on the objects as matching what’s actually in the database, there are a variety of events that will cause objects to re-access the database in order to keep synchronized. It is possible to “detach” objects from a Session, and to continue using them, though this practice has its caveats. It’s intended that usually, you’d re-associate detached objects with another Session when you want to work with them again, so that they can resume their normal task of representing database state.

Basics of Using a Session

The most basic Session use patterns are presented here.

Opening and Closing a Session

The Session may be constructed on its own or by using the sessionmaker class. It typically is passed a single Engine as a source of connectivity up front. A typical use may look like:

from sqlalchemy import create_engine
from sqlalchemy.orm import Session

# an Engine, which the Session will use for connection
# resources
engine = create_engine('postgresql://scott:tiger@localhost/')

# create session and add objects
with Session(engine) as session:

Above, the Session is instantiated with an Engine associated with a particular database URL. It is then used in a Python context manager (i.e. with: statement) so that it is automatically closed at the end of the block; this is equivalent to calling the Session.close() method.

The call to Session.commit() is optional, and is only needed if the work we’ve done with the Session includes new data to be persisted to the database. If we were only issuing SELECT calls and did not need to write any changes, then the call to Session.commit() would be unnecessary.


Note that after Session.commit() is called, either explicitly or when using a context manager, all objects associated with the Session are expired, meaning their contents are erased to be re-loaded within the next transaction. If these objects are instead detached, they will be non-functional until re-associated with a new Session, unless the Session.expire_on_commit parameter is used to disable this behavior. See the section Committing for more detail.

Framing out a begin / commit / rollback block

We may also enclose the Session.commit() call and the overall “framing” of the transaction within a context manager for those cases where we will be committing data to the database. By “framing” we mean that if all operations succeed, the Session.commit() method will be called, but if any exceptions are raised, the Session.rollback() method will be called so that the transaction is rolled back immediately, before propagating the exception outward. In Python this is most fundamentally expressed using a try: / except: / else: block such as:

# verbose version of what a context manager will do
with Session(engine) as session:

The long-form sequence of operations illustrated above can be achieved more succinctly by making use of the SessionTransaction object returned by the Session.begin() method, which provides a context manager interface for the same sequence of operations:

# create session and add objects
with Session(engine) as session:
    with session.begin():
    # inner context calls session.commit(), if there were no exceptions
# outer context calls session.close()

More succinctly, the two contexts may be combined:

# create session and add objects
with Session(engine) as session, session.begin():
# inner context calls session.commit(), if there were no exceptions
# outer context calls session.close()

Using a sessionmaker

The purpose of sessionmaker is to provide a factory for Session objects with a fixed configuration. As it is typical that an application will have an Engine object in module scope, the sessionmaker can provide a factory for Session objects that are against this engine:

from sqlalchemy import create_engine
from sqlalchemy.orm import sessionmaker

# an Engine, which the Session will use for connection
# resources, typically in module scope
engine = create_engine('postgresql://scott:tiger@localhost/')

# a sessionmaker(), also in the same scope as the engine
Session = sessionmaker(engine)

# we can now construct a Session() without needing to pass the
# engine each time
with Session() as session:
# closes the session

The sessionmaker is analogous to the Engine as a module-level factory for function-level sessions / connections. As such it also has its own sessionmaker.begin() method, analogous to Engine.begin(), which returns a Session object and also maintains a begin/commit/rollback block:

from sqlalchemy import create_engine
from sqlalchemy.orm import sessionmaker

# an Engine, which the Session will use for connection
# resources
engine = create_engine('postgresql://scott:tiger@localhost/')

# a sessionmaker(), also in the same scope as the engine
Session = sessionmaker(engine)

# we can now construct a Session() and include begin()/commit()/rollback()
# at once
with Session.begin() as session:
# commits the transaction, closes the session

Where above, the Session will both have its transaction committed as well as that the Session will be closed, when the above with: block ends.

When you write your application, the sessionmaker factory should be scoped the same as the Engine object created by create_engine(), which is typically at module-level or global scope. As these objects are both factories, they can be used by any number of functions and threads simultaneously.

Querying (1.x Style)

The Session.query() function takes one or more entities and returns a new Query object which will issue mapper queries within the context of this Session. By “entity” we refer to a mapped class, an attribute of a mapped class, or other ORM constructs such as an aliased() construct:

# query from a class
results = session.query(User).filter_by(name='ed').all()

# query with multiple classes, returns tuples
results = session.query(User, Address).join('addresses').filter_by(name='ed').all()

# query using orm-columns, also returns tuples
results = session.query(, User.fullname).all()

When ORM objects are returned in results, they are also stored in the identity map. When an incoming database row has a primary key that matches an object which is already present, the same object is returned, and those attributes of the object which already have a value are not re-populated.

The Session automatically expires all instances along transaction boundaries (i.e. when the current transaction is committed or rolled back) so that with a normally isolated transaction, data will refresh itself when a new transaction begins.

The Query object is introduced in great detail in Object Relational Tutorial (1.x API), and further documented in Query API.

Querying (2.0 style)

New in version 1.4.

SQLAlchemy 2.0 will standardize the production of SELECT statements across both Core and ORM by making direct use of the Select object within the ORM, removing the need for there to be a separate Query object. This mode of operation is available in SQLAlchemy 1.4 right now to support applications that will be migrating to 2.0. The Session must be instantiated with the Session.future flag set to True; from that point on the Session.execute() method will return ORM results via the standard Result object when invoking select() statements that use ORM entities:

from sqlalchemy import select
from sqlalchemy.orm import Session

session = Session(engine, future=True)

# query from a class
statement = select(User).filter_by(name="ed")

# list of first element of each row (i.e. User objects)
result = session.execute(statement).scalars().all()

# query with multiple classes
statement = select(User, Address).join('addresses').filter_by(name='ed')

# list of tuples
result = session.execute(statement).all()

# query with ORM columns
statement = select(, User.fullname)

# list of tuples
result = session.execute(statement).all()

It’s important to note that while methods of Query such as Query.all() and will return instances of ORM mapped objects directly in the case that only a single complete entity were requested, the Result object returned by Session.execute() will always deliver rows (named tuples) by default; this is so that results against single or multiple ORM objects, columns, tables, etc. may all be handled identically.

If only one ORM entity was queried, the rows returned will have exactly one column, consisting of the ORM-mapped object instance for each row. To convert these rows into object instances without the tuples, the Result.scalars() method is used to first apply a “scalars” filter to the result; then the Result can be iterated or deliver rows via standard methods such as Result.all(), Result.first(), etc.

Adding New or Existing Items

Session.add() is used to place instances in the session. For transient (i.e. brand new) instances, this will have the effect of an INSERT taking place for those instances upon the next flush. For instances which are persistent (i.e. were loaded by this session), they are already present and do not need to be added. Instances which are detached (i.e. have been removed from a session) may be re-associated with a session using this method:

user1 = User(name='user1')
user2 = User(name='user2')

session.commit()     # write changes to the database

To add a list of items to the session at once, use Session.add_all():

session.add_all([item1, item2, item3])

The Session.add() operation cascades along the save-update cascade. For more details see the section Cascades.


The Session.delete() method places an instance into the Session’s list of objects to be marked as deleted:

# mark two objects to be deleted

# commit (or flush)

Session.delete() marks an object for deletion, which will result in a DELETE statement emitted for each primary key affected. Before the pending deletes are flushed, objects marked by “delete” are present in the Session.deleted collection. After the DELETE, they are expunged from the Session, which becomes permanent after the transaction is committed.

There are various important behaviors related to the Session.delete() operation, particularly in how relationships to other objects and collections are handled. There’s more information on how this works in the section Cascades, but in general the rules are:

  • Rows that correspond to mapped objects that are related to a deleted object via the relationship() directive are not deleted by default. If those objects have a foreign key constraint back to the row being deleted, those columns are set to NULL. This will cause a constraint violation if the columns are non-nullable.

  • To change the “SET NULL” into a DELETE of a related object’s row, use the delete cascade on the relationship().

  • Rows that are in tables linked as “many-to-many” tables, via the relationship.secondary parameter, are deleted in all cases when the object they refer to is deleted.

  • When related objects include a foreign key constraint back to the object being deleted, and the related collections to which they belong are not currently loaded into memory, the unit of work will emit a SELECT to fetch all related rows, so that their primary key values can be used to emit either UPDATE or DELETE statements on those related rows. In this way, the ORM without further instruction will perform the function of ON DELETE CASCADE, even if this is configured on Core ForeignKeyConstraint objects.

  • The relationship.passive_deletes parameter can be used to tune this behavior and rely upon “ON DELETE CASCADE” more naturally; when set to True, this SELECT operation will no longer take place, however rows that are locally present will still be subject to explicit SET NULL or DELETE. Setting relationship.passive_deletes to the string "all" will disable all related object update/delete.

  • When the DELETE occurs for an object marked for deletion, the object is not automatically removed from collections or object references that refer to it. When the Session is expired, these collections may be loaded again so that the object is no longer present. However, it is preferable that instead of using Session.delete() for these objects, the object should instead be removed from its collection and then delete-orphan should be used so that it is deleted as a secondary effect of that collection removal. See the section Notes on Delete - Deleting Objects Referenced from Collections and Scalar Relationships for an example of this.

See also

delete - describes “delete cascade”, which marks related objects for deletion when a lead object is deleted.

delete-orphan - describes “delete orphan cascade”, which marks related objects for deletion when they are de-associated from their lead object.

Notes on Delete - Deleting Objects Referenced from Collections and Scalar Relationships - important background on Session.delete() as involves relationships being refreshed in memory.


When the Session is used with its default configuration, the flush step is nearly always done transparently. Specifically, the flush occurs before any individual SQL statement is issued as a result of a Query or a 2.0-style Session.execute() call, as well as within the Session.commit() call before the transaction is committed. It also occurs before a SAVEPOINT is issued when Session.begin_nested() is used.

A Session flush can be forced at any time by calling the Session.flush() method:


The flush which occurs automatically within the scope of certain methods is known as autoflush. Autoflush is defined as a configurable, automatic flush call which occurs at the beginning of methods including:

There are also points at which flushes occur unconditionally; these points are within key transactional boundaries which include:

The autoflush behavior, as applied to the previous list of items, can be disabled by constructing a Session or sessionmaker passing the Session.autoflush parameter as False:

Session = sessionmaker(autoflush=False)

Additionally, autoflush can be temporarily disabled within the flow of using a Session using the Session.no_autoflush context manager:

with mysession.no_autoflush:

To reiterate: The flush process always occurs when transactional methods such as Session.commit() and Session.begin_nested() are called, regardless of any “autoflush” settings, when the Session has remaining pending changes to process.

The flush process always occurs within a transaction, (subject to the isolation level of the database transaction), provided that the DBAPI is not in driver level autocommit mode. This includes even if the Session has been configured with the deprecated Session.autocommit setting, which disables the session’s persistent transactional state. If no transaction is present, Session.flush() creates its own transaction and commits it. This means that assuming the database connection is providing for atomicity within its transactional settings, if any individual DML statement inside the flush fails, the entire operation will be rolled back.

Outside of using Session.autocommit, when a failure occurs within a flush, in order to continue using that same Session, an explicit call to Session.rollback() is required after a flush fails, even though the underlying transaction will have been rolled back already (even if the database driver is technically in driver-level autocommit mode). This is so that the overall nesting pattern of so-called “subtransactions” is consistently maintained. The FAQ section “This Session’s transaction has been rolled back due to a previous exception during flush.” (or similar) contains a more detailed description of this behavior.

Expiring / Refreshing

An important consideration that will often come up when using the Session is that of dealing with the state that is present on objects that have been loaded from the database, in terms of keeping them synchronized with the current state of the transaction. The SQLAlchemy ORM is based around the concept of an identity map such that when an object is “loaded” from a SQL query, there will be a unique Python object instance maintained corresponding to a particular database identity. This means if we emit two separate queries, each for the same row, and get a mapped object back, the two queries will have returned the same Python object:

>>> u1 = session.query(User).filter(id=5).first()
>>> u2 = session.query(User).filter(id=5).first()
>>> u1 is u2

Following from this, when the ORM gets rows back from a query, it will skip the population of attributes for an object that’s already loaded. The design assumption here is to assume a transaction that’s perfectly isolated, and then to the degree that the transaction isn’t isolated, the application can take steps on an as-needed basis to refresh objects from the database transaction. The FAQ entry at I’m re-loading data with my Session but it isn’t seeing changes that I committed elsewhere discusses this concept in more detail.

When an ORM mapped object is loaded into memory, there are three general ways to refresh its contents with new data from the current transaction:

  • the expire() method - the Session.expire() method will erase the contents of selected or all attributes of an object, such that they will be loaded from the database when they are next accessed, e.g. using a lazy loading pattern:

    u1.some_attribute  # <-- lazy loads from the transaction
  • the refresh() method - closely related is the Session.refresh() method, which does everything the Session.expire() method does but also emits one or more SQL queries immediately to actually refresh the contents of the object:

    session.refresh(u1)  # <-- emits a SQL query
    u1.some_attribute  # <-- is refreshed from the transaction
  • the populate_existing() method - this method is actually on the Query object as Query.populate_existing() and indicates that it should return objects that are unconditionally re-populated from their contents in the database:

    u2 = session.query(User).populate_existing().filter(id=5).first()

Further discussion on the refresh / expire concept can be found at Refreshing / Expiring.

UPDATE and DELETE with arbitrary WHERE clause

The sections above on Session.flush() and Session.delete() detail how rows can be inserted, updated and deleted in the database, based on primary key identities that are referred towards by mapped Python objects in the application. The Session can also emit UPDATE and DELETE statements with arbitrary WHERE clauses as well, and at the same time refresh locally present objects which match those rows.

To emit an ORM-enabled UPDATE in 1.x style, the Query.update() method may be used:

session.query(User).filter( == "squidward").\
    update({"name": "spongebob"}, synchronize_session="fetch")

Above, an UPDATE will be emitted against all rows that match the name “squidward” and be updated to the name “spongebob”. The Query.update.synchronize_session parameter referring to “fetch” indicates the list of affected primary keys should be fetched either via a separate SELECT statement or via RETURNING if the backend database supports it; objects locally present in memory will be updated in memory based on these primary key identities.

For ORM-enabled UPDATEs in 2.0 style, Session.execute() is used with the Core Update construct:

from sqlalchemy import update

stmt = update(User).where( == "squidward").values(name="spongebob").\

result = session.execute(stmt)

Above, the Update.execution_options() method may be used to establish execution-time options such as “synchronize_session”.

The result object returned is an instance of CursorResult; to retrieve the number of rows matched by any UPDATE or DELETE statement, use CursorResult.rowcount:

num_rows_matched = result.rowcount

DELETEs work in the same way as UPDATE except there is no “values / set” clause established. When synchronize_session is used, matching objects within the Session will be marked as deleted and expunged.

ORM-enabled delete, 1.x style:

session.query(User).filter( == "squidward").\

ORM-enabled delete, 2.0 style:

from sqlalchemy import delete

stmt = delete(User).where( == "squidward").execution_options(synchronize_session="fetch")


Selecting a Synchronization Strategy

With both the 1.x and 2.0 form of ORM-enabled updates and deletes, the following values for synchronize_session are supported:

  • False - don’t synchronize the session. This option is the most efficient and is reliable once the session is expired, which typically occurs after a commit(), or explicitly using expire_all(). Before the expiration, objects that were updated or deleted in the database may still remain in the session with stale values, which can lead to confusing results.

  • 'fetch' - Retrieves the primary key identity of affected rows by either performing a SELECT before the UPDATE or DELETE, or by using RETURNING if the database supports it, so that in-memory objects which are affected by the operation can be refreshed with new values (updates) or expunged from the Session (deletes). Note that this synchronization strategy is not available if the given update() or delete() construct specifies columns for UpdateBase.returning() explicitly.

  • 'evaluate' - Evaluate the WHERE criteria given in the UPDATE or DELETE statement in Python, to locate matching objects within the Session. This approach does not add any round trips and in the absence of RETURNING support is more efficient. For UPDATE or DELETE statements with complex criteria, the 'evaluate' strategy may not be able to evaluate the expression in Python and will raise an error. If this occurs, use the 'fetch' strategy for the operation instead.


    The "evaluate" strategy should be avoided if an UPDATE operation is to run on a Session that has many objects which have been expired, because it will necessarily need to refresh those objects as they are located which will emit a SELECT for each one. The Session may have expired objects if it is being used across multiple Session.commit() calls and the Session.expire_on_commit flag is at its default value of True.


Additional Caveats for ORM-enabled updates and deletes

The ORM-enabled UPDATE and DELETE features bypass ORM unit-of-work automation in favor being able to emit a single UPDATE or DELETE statement that matches multiple rows at once without complexity.

  • The operations do not offer in-Python cascading of relationships - it is assumed that ON UPDATE CASCADE and/or ON DELETE CASCADE is configured for any foreign key references which require it, otherwise the database may emit an integrity violation if foreign key references are being enforced.

  • After the UPDATE or DELETE, dependent objects in the Session which were impacted by an ON UPDATE CASCADE or ON DELETE CASCADE on related tables may not contain the current state; this issue is resolved once the Session is expired, which normally occurs upon Session.commit() or can be forced by using Session.expire_all().

  • The 'fetch' strategy, when run on a database that does not support RETURNING such as MySQL or SQLite, results in an additional SELECT statement emitted which may reduce performance. Use SQL echoing when developing to evaluate the impact of SQL emitted.

  • ORM-enabled UPDATEs and DELETEs do not handle joined table inheritance automatically. If the operation is against multiple tables, typically individual UPDATE / DELETE statements against the individual tables should be used. Some databases support multiple table UPDATEs. Similar guidelines as those detailed at Multiple Table Updates may be applied.

  • The WHERE criteria needed in order to limit the polymorphic identity to specific subclasses for single-table-inheritance mappings is included automatically . This only applies to a subclass mapper that has no table of its own.

    Changed in version 1.4: ORM updates/deletes now automatically accommodate for the WHERE criteria added for single-inheritance mappings.

  • The with_loader_criteria() option is supported by ORM update and delete operations; criteria here will be added to that of the UPDATE or DELETE statement being emitted, as well as taken into account during the “synchronize” process.

  • In order to intercept ORM-enabled UPDATE and DELETE operations with event handlers, use the SessionEvents.do_orm_execute() event.

Selecting ORM Objects Inline with UPDATE.. RETURNING or INSERT..RETURNING

This section has moved. See Using INSERT, UPDATE and ON CONFLICT (i.e. upsert) to return ORM Objects.

Auto Begin

New in version 1.4: This section describes a behavior that is new in SQLAlchemy 1.4 and does not apply to previous versions. Further details on the “autobegin” change are at Session features new “autobegin” behavior.

The Session object features a behavior known as autobegin. This indicates that the Session will internally consider itself to be in a “transactional” state as soon as any work is performed with the Session, either involving modifications to the internal state of the Session with regards to object state changes, or with operations that require database connectivity.

When the Session is first constructed, there’s no transactional state present. The transactional state is begun automatically, when a method such as Session.add() or Session.execute() is invoked, or similarly if a Query is executed to return results (which ultimately uses Session.execute()), or if an attribute is modified on a persistent object.

The transactional state can be checked by accessing the Session.in_transaction() method, which returns True or False indicating if the “autobegin” step has proceeded. While not normally needed, the Session.get_transaction() method will return the actual SessionTransaction object that represents this transactional state.

The transactional state of the Session may also be started explicitly, by invoking the Session.begin() method. When this method is called, the Session is placed into the “transactional” state unconditionally. Session.begin() may be used as a context manager as described at Framing out a begin / commit / rollback block.

Changed in version 1.4.12: - autobegin now correctly occurs if object attributes are modified; previously this was not occurring.


Session.commit() is used to commit the current transaction. At its core this indicates that it emits COMMIT on all current database connections that have a transaction in progress; from a DBAPI perspective this means the connection.commit() DBAPI method is invoked on each DBAPI connection.

When there is no transaction in place for the Session, indicating that no operations were invoked on this Session since the previous call to Session.commit(), the method will begin and commit an internal-only “logical” transaction, that does not normally affect the database unless pending flush changes were detected, but will still invoke event handlers and object expiration rules.

The Session.commit() operation unconditionally issues Session.flush() before emitting COMMIT on relevant database connections. If no pending changes are detected, then no SQL is emitted to the database. This behavior is not configurable and is not affected by the Session.autoflush parameter.

Subsequent to that, Session.commit() will then COMMIT the actual database transaction or transactions, if any, that are in place.

Finally, all objects within the Session are expired as the transaction is closed out. This is so that when the instances are next accessed, either through attribute access or by them being present in the result of a SELECT, they receive the most recent state. This behavior may be controlled by the Session.expire_on_commit flag, which may be set to False when this behavior is undesirable.

See also

Auto Begin

Rolling Back

Session.rollback() rolls back the current transaction, if any. When there is no transaction in place, the method passes silently.

With a default configured session, the post-rollback state of the session, subsequent to a transaction having been begun either via autobegin or by calling the Session.begin() method explicitly, is as follows:

  • All transactions are rolled back and all connections returned to the connection pool, unless the Session was bound directly to a Connection, in which case the connection is still maintained (but still rolled back).

  • Objects which were initially in the pending state when they were added to the Session within the lifespan of the transaction are expunged, corresponding to their INSERT statement being rolled back. The state of their attributes remains unchanged.

  • Objects which were marked as deleted within the lifespan of the transaction are promoted back to the persistent state, corresponding to their DELETE statement being rolled back. Note that if those objects were first pending within the transaction, that operation takes precedence instead.

  • All objects not expunged are fully expired - this is regardless of the Session.expire_on_commit setting.

With that state understood, the Session may safely continue usage after a rollback occurs.

Changed in version 1.4: The Session object now features deferred “begin” behavior, as described in autobegin. If no transaction is begun, methods like Session.commit() and Session.rollback() have no effect. This behavior would not have been observed prior to 1.4 as under non-autocommit mode, a transaction would always be implicitly present.

When a Session.flush() fails, typically for reasons like primary key, foreign key, or “not nullable” constraint violations, a ROLLBACK is issued automatically (it’s currently not possible for a flush to continue after a partial failure). However, the Session goes into a state known as “inactive” at this point, and the calling application must always call the Session.rollback() method explicitly so that the Session can go back into a usable state (it can also be simply closed and discarded). See the FAQ entry at “This Session’s transaction has been rolled back due to a previous exception during flush.” (or similar) for further discussion.

See also

Auto Begin


The Session.close() method issues a Session.expunge_all() which removes all ORM-mapped objects from the session, and releases any transactional/connection resources from the Engine object(s) to which it is bound. When connections are returned to the connection pool, transactional state is rolled back as well.

When the Session is closed, it is essentially in the original state as when it was first constructed, and may be used again. In this sense, the Session.close() method is more like a “reset” back to the clean state and not as much like a “database close” method.

It’s recommended that the scope of a Session be limited by a call to Session.close() at the end, especially if the Session.commit() or Session.rollback() methods are not used. The Session may be used as a context manager to ensure that Session.close() is called:

with Session(engine) as session:
    result = session.execute(select(User))

# closes session automatically

Changed in version 1.4: The Session object features deferred “begin” behavior, as described in autobegin. no longer immediately begins a new transaction after the Session.close() method is called.

Session Frequently Asked Questions

By this point, many users already have questions about sessions. This section presents a mini-FAQ (note that we have also a real FAQ) of the most basic issues one is presented with when using a Session.

When do I make a sessionmaker?

Just one time, somewhere in your application’s global scope. It should be looked upon as part of your application’s configuration. If your application has three .py files in a package, you could, for example, place the sessionmaker line in your file; from that point on your other modules say “from mypackage import Session”. That way, everyone else just uses Session(), and the configuration of that session is controlled by that central point.

If your application starts up, does imports, but does not know what database it’s going to be connecting to, you can bind the Session at the “class” level to the engine later on, using sessionmaker.configure().

In the examples in this section, we will frequently show the sessionmaker being created right above the line where we actually invoke Session. But that’s just for example’s sake! In reality, the sessionmaker would be somewhere at the module level. The calls to instantiate Session would then be placed at the point in the application where database conversations begin.

When do I construct a Session, when do I commit it, and when do I close it?


  1. As a general rule, keep the lifecycle of the session separate and external from functions and objects that access and/or manipulate database data. This will greatly help with achieving a predictable and consistent transactional scope.

  2. Make sure you have a clear notion of where transactions begin and end, and keep transactions short, meaning, they end at the series of a sequence of operations, instead of being held open indefinitely.

A Session is typically constructed at the beginning of a logical operation where database access is potentially anticipated.

The Session, whenever it is used to talk to the database, begins a database transaction as soon as it starts communicating. This transaction remains in progress until the Session is rolled back, committed, or closed. The Session will begin a new transaction if it is used again, subsequent to the previous transaction ending; from this it follows that the Session is capable of having a lifespan across many transactions, though only one at a time. We refer to these two concepts as transaction scope and session scope.

It’s usually not very hard to determine the best points at which to begin and end the scope of a Session, though the wide variety of application architectures possible can introduce challenging situations.

Some sample scenarios include:

  • Web applications. In this case, it’s best to make use of the SQLAlchemy integrations provided by the web framework in use. Or otherwise, the basic pattern is create a Session at the start of a web request, call the Session.commit() method at the end of web requests that do POST, PUT, or DELETE, and then close the session at the end of web request. It’s also usually a good idea to set Session.expire_on_commit to False so that subsequent access to objects that came from a Session within the view layer do not need to emit new SQL queries to refresh the objects, if the transaction has been committed already.

  • A background daemon which spawns off child forks would want to create a Session local to each child process, work with that Session through the life of the “job” that the fork is handling, then tear it down when the job is completed.

  • For a command-line script, the application would create a single, global Session that is established when the program begins to do its work, and commits it right as the program is completing its task.

  • For a GUI interface-driven application, the scope of the Session may best be within the scope of a user-generated event, such as a button push. Or, the scope may correspond to explicit user interaction, such as the user “opening” a series of records, then “saving” them.

As a general rule, the application should manage the lifecycle of the session externally to functions that deal with specific data. This is a fundamental separation of concerns which keeps data-specific operations agnostic of the context in which they access and manipulate that data.

E.g. don’t do this:

### this is the **wrong way to do it** ###

class ThingOne(object):
    def go(self):
        session = Session()
            session.query(FooBar).update({"x": 5})

class ThingTwo(object):
    def go(self):
        session = Session()
            session.query(Widget).update({"q": 18})

def run_my_program():

Keep the lifecycle of the session (and usually the transaction) separate and external. The example below illustrates how this might look, and additionally makes use of a Python context manager (i.e. the with: keyword) in order to manage the scope of the Session and its transaction automatically:

### this is a **better** (but not the only) way to do it ###

class ThingOne(object):
    def go(self, session):
        session.query(FooBar).update({"x": 5})

class ThingTwo(object):
    def go(self, session):
        session.query(Widget).update({"q": 18})

def run_my_program():
    with Session() as session:
        with session.begin():

Changed in version 1.4: The Session may be used as a context manager without the use of external helper functions.

Is the Session a cache?

Yeee…no. It’s somewhat used as a cache, in that it implements the identity map pattern, and stores objects keyed to their primary key. However, it doesn’t do any kind of query caching. This means, if you say session.query(Foo).filter_by(name='bar'), even if Foo(name='bar') is right there, in the identity map, the session has no idea about that. It has to issue SQL to the database, get the rows back, and then when it sees the primary key in the row, then it can look in the local identity map and see that the object is already there. It’s only when you say query.get({some primary key}) that the Session doesn’t have to issue a query.

Additionally, the Session stores object instances using a weak reference by default. This also defeats the purpose of using the Session as a cache.

The Session is not designed to be a global object from which everyone consults as a “registry” of objects. That’s more the job of a second level cache. SQLAlchemy provides a pattern for implementing second level caching using dogpile.cache, via the Dogpile Caching example.

How can I get the Session for a certain object?

Use the Session.object_session() classmethod available on Session:

session = Session.object_session(someobject)

The newer Runtime Inspection API system can also be used:

from sqlalchemy import inspect
session = inspect(someobject).session

Is the session thread-safe?

The Session is very much intended to be used in a non-concurrent fashion, which usually means in only one thread at a time.

The Session should be used in such a way that one instance exists for a single series of operations within a single transaction. One expedient way to get this effect is by associating a Session with the current thread (see Contextual/Thread-local Sessions for background). Another is to use a pattern where the Session is passed between functions and is otherwise not shared with other threads.

The bigger point is that you should not want to use the session with multiple concurrent threads. That would be like having everyone at a restaurant all eat from the same plate. The session is a local “workspace” that you use for a specific set of tasks; you don’t want to, or need to, share that session with other threads who are doing some other task.

Making sure the Session is only used in a single concurrent thread at a time is called a “share nothing” approach to concurrency. But actually, not sharing the Session implies a more significant pattern; it means not just the Session object itself, but also all objects that are associated with that Session, must be kept within the scope of a single concurrent thread. The set of mapped objects associated with a Session are essentially proxies for data within database rows accessed over a database connection, and so just like the Session itself, the whole set of objects is really just a large-scale proxy for a database connection (or connections). Ultimately, it’s mostly the DBAPI connection itself that we’re keeping away from concurrent access; but since the Session and all the objects associated with it are all proxies for that DBAPI connection, the entire graph is essentially not safe for concurrent access.

If there are in fact multiple threads participating in the same task, then you may consider sharing the session and its objects between those threads; however, in this extremely unusual scenario the application would need to ensure that a proper locking scheme is implemented so that there isn’t concurrent access to the Session or its state. A more common approach to this situation is to maintain a single Session per concurrent thread, but to instead copy objects from one Session to another, often using the Session.merge() method to copy the state of an object into a new object local to a different Session.