1.x style
2.0 style

These terms are new in SQLAlchemy 1.4 and refer to the SQLAlchemy 1.4-> 2.0 transition plan, described at Migrating to SQLAlchemy 2.0. The term “1.x style” refers to an API used in the way it’s been documented throughout the 1.x series of SQLAlchemy and earlier (e.g. 1.3, 1.2, etc) and the term “2.0 style” refers to the way an API will look in version 2.0. Version 1.4 implements nearly all of 2.0’s API in so-called “transition mode”.

Enabling 2.0 style usage

When using code from a documentation example that indicates 2.0-style, the Engine as well as the Session in use should make use of “future” mode, via the create_engine.future and Session.future flags:

from sqlalchemy import create_engine
from sqlalchemy.orm import sessionmaker

engine = create_engine("mysql://user:pass@host/dbname", future=True)
Session = sessionmaker(bind=engine, future=True)

ORM Queries in 2.0 style

Besides the above changes to Engine and Session, probably the most major API change implied by 1.x->2.0 is the migration from using the Query object for ORM SELECT statements and instead using the select() construct in conjunction with the Session.execute() method. The general change looks like the following. Given a Session and a Query against that Session:

list_of_users = session.query(User).join(User.addresses).all()

The new style constructs the query separately from the Session using the select() construct; when populated with ORM entities like the User class from the ORM Tutorial, the resulting Select construct receives additional “plugin” state that allows it to work like the Query:

from sqlalchemy import select

# a Core select statement with ORM entities is
# now ORM-enabled at the compiler level
stmt = select(User).join(User.addresses)

session = Session(engine)

result = session.execute(stmt)

# Session returns a Result that has ORM entities
list_of_users = result.scalars().all()
ACID model

An acronym for “Atomicity, Consistency, Isolation, Durability”; a set of properties that guarantee that database transactions are processed reliably. (via Wikipedia)


Annotations are a concept used internally by SQLAlchemy in order to store additional information along with ClauseElement objects. A Python dictionary is associated with a copy of the object, which contains key/value pairs significant to various internal systems, mostly within the ORM:

some_column = Column("some_column", Integer)
some_column_annotated = some_column._annotate({"entity": User})

The annotation system differs from the public dictionary in that the above annotation operation creates a copy of the new Column, rather than considering all annotation values to be part of a single unit. The ORM creates copies of expression objects in order to apply annotations that are specific to their context, such as to differentiate columns that should render themselves as relative to a joined-inheritance entity versus those which should render relative to their immediate parent table alone, as well as to differentiate columns within the “join condition” of a relationship where the column in some cases needs to be expressed in terms of one particular table alias or another, based on its position within the join expression.

association relationship

A two-tiered relationship which links two tables together using an association table in the middle. The association relationship differs from a many to many relationship in that the many-to-many table is mapped by a full class, rather than invisibly handled by the sqlalchemy.orm.relationship() construct as in the case with many-to-many, so that additional attributes are explicitly available.

For example, if we wanted to associate employees with projects, also storing the specific role for that employee with the project, the relational schema might look like:

CREATE TABLE employee (
    name VARCHAR(30)

CREATE TABLE project (
    name VARCHAR(30)

CREATE TABLE employee_project (
    employee_id INTEGER PRIMARY KEY,
    project_id INTEGER PRIMARY KEY,
    role_name VARCHAR(30),
    FOREIGN KEY employee_id REFERENCES employee(id),
    FOREIGN KEY project_id REFERENCES project(id)

A SQLAlchemy declarative mapping for the above might look like:

class Employee(Base):
    __tablename__ = "employee"

    id = Column(Integer, primary_key=True)
    name = Column(String(30))

class Project(Base):
    __tablename__ = "project"

    id = Column(Integer, primary_key=True)
    name = Column(String(30))

class EmployeeProject(Base):
    __tablename__ = "employee_project"

    employee_id = Column(Integer, ForeignKey(""), primary_key=True)
    project_id = Column(Integer, ForeignKey(""), primary_key=True)
    role_name = Column(String(30))

    project = relationship("Project", backref="project_employees")
    employee = relationship("Employee", backref="employee_projects")

Employees can be added to a project given a role name:

proj = Project(name="Client A")

emp1 = Employee(name="emp1")
emp2 = Employee(name="emp2")

        EmployeeProject(employee=emp1, role_name="tech lead"),
        EmployeeProject(employee=emp2, role_name="account executive"),

See also

many to many


Atomicity is one of the components of the ACID model, and requires that each transaction is “all or nothing”: if one part of the transaction fails, the entire transaction fails, and the database state is left unchanged. An atomic system must guarantee atomicity in each and every situation, including power failures, errors, and crashes. (via Wikipedia)


Indicates an ORM object that is presently associated with a specific Session.

bidirectional relationship

An extension to the relationship system whereby two distinct relationship() objects can be mutually associated with each other, such that they coordinate in memory as changes occur to either side. The most common way these two relationships are constructed is by using the relationship() function explicitly for one side and specifying the backref keyword to it so that the other relationship() is created automatically. We can illustrate this against the example we’ve used in one to many as follows:

class Department(Base):
    __tablename__ = "department"
    id = Column(Integer, primary_key=True)
    name = Column(String(30))
    employees = relationship("Employee", backref="department")

class Employee(Base):
    __tablename__ = "employee"
    id = Column(Integer, primary_key=True)
    name = Column(String(30))
    dep_id = Column(Integer, ForeignKey(""))

A backref can be applied to any relationship, including one to many, many to one, and many to many.

bound parameter
bound parameters
bind parameter
bind parameters

Bound parameters are the primary means in which data is passed to the DBAPI database driver. While the operation to be invoked is based on the SQL statement string, the data values themselves are passed separately, where the driver contains logic that will safely process these strings and pass them to the backend database server, which may either involve formatting the parameters into the SQL string itself, or passing them to the database using separate protocols.

The specific system by which the database driver does this should not matter to the caller; the point is that on the outside, data should always be passed separately and not as part of the SQL string itself. This is integral both to having adequate security against SQL injections as well as allowing the driver to have the best performance.

See also

Prepared Statement - at Wikipedia

bind parameters - at Use The Index, Luke!

Sending Parameters - in the SQLAlchemy 1.4 / 2.0 Tutorial

candidate key

A relational algebra term referring to an attribute or set of attributes that form a uniquely identifying key for a row. A row may have more than one candidate key, each of which is suitable for use as the primary key of that row. The primary key of a table is always a candidate key.

cartesian product

Given two sets A and B, the cartesian product is the set of all ordered pairs (a, b) where a is in A and b is in B.

In terms of SQL databases, a cartesian product occurs when we select from two or more tables (or other subqueries) without establishing any kind of criteria between the rows of one table to another (directly or indirectly). If we SELECT from table A and table B at the same time, we get every row of A matched to the first row of B, then every row of A matched to the second row of B, and so on until every row from A has been paired with every row of B.

Cartesian products cause enormous result sets to be generated and can easily crash a client application if not prevented.


A term used in SQLAlchemy to describe how an ORM persistence action that takes place on a particular object would extend into other objects which are directly associated with that object. In SQLAlchemy, these object associations are configured using the relationship() construct. relationship() contains a parameter called relationship.cascade which provides options on how certain persistence operations may cascade.

The term “cascades” as well as the general architecture of this system in SQLAlchemy was borrowed, for better or worse, from the Hibernate ORM.

See also


check constraint

A check constraint is a condition that defines valid data when adding or updating an entry in a table of a relational database. A check constraint is applied to each row in the table.

(via Wikipedia)

A check constraint can be added to a table in standard SQL using DDL like the following:

ALTER TABLE distributors ADD CONSTRAINT zipchk CHECK (char_length(zipcode) = 5);
columns clause

The portion of the SELECT statement which enumerates the SQL expressions to be returned in the result set. The expressions follow the SELECT keyword directly and are a comma-separated list of individual expressions.


FROM user_account WHERE = 'fred'

Above, the list of columns, is the columns clause of the SELECT.

composite primary key

A primary key that has more than one column. A particular database row is unique based on two or more columns rather than just a single value.

See also

primary key


Consistency is one of the components of the ACID model, and ensures that any transaction will bring the database from one valid state to another. Any data written to the database must be valid according to all defined rules, including but not limited to constraints, cascades, triggers, and any combination thereof. (via Wikipedia)


Rules established within a relational database that ensure the validity and consistency of data. Common forms of constraint include primary key constraint, foreign key constraint, and check constraint.

correlated subquery
correlated subqueries

A subquery is correlated if it depends on data in the enclosing SELECT.

Below, a subquery selects the aggregate value MIN( from the email_address table, such that it will be invoked for each value of, correlating the value of this column against the email_address.user_account_id column:

 FROM user_account
 JOIN email_address ON
 WHERE = (
    SELECT MIN( FROM email_address AS a

The above subquery refers to the user_account table, which is not itself in the FROM clause of this nested query. Instead, the user_account table is received from the enclosing query, where each row selected from user_account results in a distinct execution of the subquery.

A correlated subquery is in most cases present in the WHERE clause or columns clause of the immediately enclosing SELECT statement, as well as in the ORDER BY or HAVING clause.

In less common cases, a correlated subquery may be present in the FROM clause of an enclosing SELECT; in these cases the correlation is typically due to the enclosing SELECT itself being enclosed in the WHERE, ORDER BY, columns or HAVING clause of another SELECT, such as:

        SELECT AS id, child.parent_id AS parent_id, child.pos AS pos
        FROM child
        WHERE child.parent_id = ORDER BY child.pos
    LIMIT 3)
WHERE id = 7)

Correlation from one SELECT directly to one which encloses the correlated query via its FROM clause is not possible, because the correlation can only proceed once the original source rows from the enclosing statement’s FROM clause are available.


An acronym meaning “Create, Update, Delete”. The term in SQL refers to the set of operations that create, modify and delete data from the database, also known as DML, and typically refers to the INSERT, UPDATE, and DELETE statements.


A control structure that enables traversal over the records in a database. In the Python DBAPI, the cursor object is in fact the starting point for statement execution as well as the interface used for fetching results.

cyclomatic complexity

A measure of code complexity based on the number of possible paths through a program’s source code.


DBAPI is shorthand for the phrase “Python Database API Specification”. This is a widely used specification within Python to define common usage patterns for all database connection packages. The DBAPI is a “low level” API which is typically the lowest level system used in a Python application to talk to a database. SQLAlchemy’s dialect system is constructed around the operation of the DBAPI, providing individual dialect classes which service a specific DBAPI on top of a specific database engine; for example, the create_engine() URL postgresql+psycopg2://@localhost/test refers to the psycopg2 DBAPI/dialect combination, whereas the URL mysql+mysqldb://@localhost/test refers to the MySQL for Python DBAPI/dialect combination.


An acronym for Data Definition Language. DDL is the subset of SQL that relational databases use to configure tables, constraints, and other permanent objects within a database schema. SQLAlchemy provides a rich API for constructing and emitting DDL expressions.


This describes one of the major object states which an object can have within a Session; a deleted object is an object that was formerly persistent and has had a DELETE statement emitted to the database within a flush to delete its row. The object will move to the detached state once the session’s transaction is committed; alternatively, if the session’s transaction is rolled back, the DELETE is reverted and the object moves back to the persistent state.


In Python, a descriptor is an object attribute with “binding behavior”, one whose attribute access has been overridden by methods in the descriptor protocol. Those methods are __get__(), __set__(), and __delete__(). If any of those methods are defined for an object, it is said to be a descriptor.

In SQLAlchemy, descriptors are used heavily in order to provide attribute behavior on mapped classes. When a class is mapped as such:

class MyClass(Base):
    __tablename__ = "foo"

    id = Column(Integer, primary_key=True)
    data = Column(String)

The MyClass class will be mapped when its definition is complete, at which point the id and data attributes, starting out as Column objects, will be replaced by the instrumentation system with instances of InstrumentedAttribute, which are descriptors that provide the above mentioned __get__(), __set__() and __delete__() methods. The InstrumentedAttribute will generate a SQL expression when used at the class level:

>>> print( == 5)
data = :data_1

and at the instance level, keeps track of changes to values, and also lazy loads unloaded attributes from the database:

>>> m1 = MyClass()
>>> = 5
>>> = "some data"

>>> from sqlalchemy import inspect
>>> inspect(m1)
"some data"

This describes one of the major object states which an object can have within a Session; a detached object is an object that has a database identity (i.e. a primary key) but is not associated with any session. An object that was previously persistent and was removed from its session either because it was expunged, or the owning session was closed, moves into the detached state. The detached state is generally used when objects are being moved between sessions or when being moved to/from an external object cache.


In SQLAlchemy, the “dialect” is a Python object that represents information and methods that allow database operations to proceed on a particular kind of database backend and a particular kind of Python driver (or DBAPI) for that database. SQLAlchemy dialects are subclasses of the Dialect class.


A result-set column which is used during polymorphic loading to determine what kind of mapped class should be applied to a particular incoming result row. In SQLAlchemy, the classes are always part of a hierarchy mapping using inheritance mapping.


An acronym for Data Manipulation Language. DML is the subset of SQL that relational databases use to modify the data in tables. DML typically refers to the three widely familiar statements of INSERT, UPDATE and DELETE, otherwise known as CRUD (acronym for “Create, Read, Update, Delete”).

domain model

A domain model in problem solving and software engineering is a conceptual model of all the topics related to a specific problem. It describes the various entities, their attributes, roles, and relationships, plus the constraints that govern the problem domain.

(via Wikipedia)


An acronym for Data Query Language. DQL is the subset of SQL that relational databases use to read the data in tables. DQL almost exclusively refers to the SQL SELECT construct as the top level SQL statement in use.


Durability is a property of the ACID model which means that once a transaction has been committed, it will remain so, even in the event of power loss, crashes, or errors. In a relational database, for instance, once a group of SQL statements execute, the results need to be stored permanently (even if the database crashes immediately thereafter). (via Wikipedia)

eager load
eager loads
eager loaded
eager loading

In object relational mapping, an “eager load” refers to an attribute that is populated with its database-side value at the same time as when the object itself is loaded from the database. In SQLAlchemy, “eager loading” usually refers to related collections of objects that are mapped using the relationship() construct. Eager loading is the opposite of lazy loading.


In the SQLAlchemy ORM, refers to when the data in a persistent or sometimes detached object is erased, such that when the object’s attributes are next accessed, a lazy load SQL query will be emitted in order to refresh the data for this object as stored in the current ongoing transaction.


An object that serves as a front-facing interface masking more complex underlying or structural code.

foreign key constraint

A referential constraint between two tables. A foreign key is a field or set of fields in a relational table that matches a candidate key of another table. The foreign key can be used to cross-reference tables. (via Wikipedia)

A foreign key constraint can be added to a table in standard SQL using DDL like the following:

FOREIGN KEY (employee) REFERENCES department (dep_id)
FROM clause

The portion of the SELECT statement which indicates the initial source of rows.

A simple SELECT will feature one or more table names in its FROM clause. Multiple sources are separated by a comma:

SELECT, address.email_address
FROM user, address

The FROM clause is also where explicit joins are specified. We can rewrite the above SELECT using a single FROM element which consists of a JOIN of the two tables:

SELECT, address.email_address
FROM user JOIN address ON

A term that SQLAlchemy uses to refer what’s normally known as method chaining; see that term for details.

identity key

A key associated with ORM-mapped objects that identifies their primary key identity within the database, as well as their unique identity within a Session identity map.

identity map

A mapping between Python objects and their database identities. The identity map is a collection that’s associated with an ORM Session object, and maintains a single instance of every database object keyed to its identity. The advantage to this pattern is that all operations which occur for a particular database identity are transparently coordinated onto a single object instance. When using an identity map in conjunction with an isolated transaction, having a reference to an object that’s known to have a particular primary key can be considered from a practical standpoint to be a proxy to the actual database row.


In the SQLAlchemy ORM, these terms refer to two different styles of mapping Python classes to database tables.


Instrumentation refers to the process of augmenting the functionality and attribute set of a particular class. Ideally, the behavior of the class should remain close to a regular class, except that additional behaviors and features are made available. The SQLAlchemy mapping process, among other things, adds database-enabled descriptors to a mapped class each of which represents a particular database column or relationship to a related class.

isolation level

The isolation property of the ACID model ensures that the concurrent execution of transactions results in a system state that would be obtained if transactions were executed serially, i.e. one after the other. Each transaction must execute in total isolation i.e. if T1 and T2 execute concurrently then each should remain independent of the other. (via Wikipedia)

lazy initialization

A tactic of delaying some initialization action, such as creating objects, populating data, or establishing connectivity to other services, until those resources are required.

lazy load
lazy loads
lazy loaded
lazy loading

In object relational mapping, a “lazy load” refers to an attribute that does not contain its database-side value for some period of time, typically when the object is first loaded. Instead, the attribute receives a memoization that causes it to go out to the database and load its data when it’s first used. Using this pattern, the complexity and time spent within object fetches can sometimes be reduced, in that attributes for related tables don’t need to be addressed immediately. Lazy loading is the opposite of eager loading.

many to many

A style of sqlalchemy.orm.relationship() which links two tables together via an intermediary table in the middle. Using this configuration, any number of rows on the left side may refer to any number of rows on the right, and vice versa.

A schema where employees can be associated with projects:

CREATE TABLE employee (
    name VARCHAR(30)

CREATE TABLE project (
    name VARCHAR(30)

CREATE TABLE employee_project (
    employee_id INTEGER PRIMARY KEY,
    project_id INTEGER PRIMARY KEY,
    FOREIGN KEY employee_id REFERENCES employee(id),
    FOREIGN KEY project_id REFERENCES project(id)

Above, the employee_project table is the many-to-many table, which naturally forms a composite primary key consisting of the primary key from each related table.

In SQLAlchemy, the sqlalchemy.orm.relationship() function can represent this style of relationship in a mostly transparent fashion, where the many-to-many table is specified using plain table metadata:

class Employee(Base):
    __tablename__ = "employee"

    id = Column(Integer, primary_key=True)
    name = Column(String(30))

    projects = relationship(
            Column("employee_id", Integer, ForeignKey(""), primary_key=True),
            Column("project_id", Integer, ForeignKey(""), primary_key=True),

class Project(Base):
    __tablename__ = "project"

    id = Column(Integer, primary_key=True)
    name = Column(String(30))

Above, the Employee.projects and back-referencing Project.employees collections are defined:

proj = Project(name="Client A")

emp1 = Employee(name="emp1")
emp2 = Employee(name="emp2")

proj.employees.extend([emp1, emp2])
many to one

A style of relationship() which links a foreign key in the parent mapper’s table to the primary key of a related table. Each parent object can then refer to exactly zero or one related object.

The related objects in turn will have an implicit or explicit one to many relationship to any number of parent objects that refer to them.

An example many to one schema (which, note, is identical to the one to many schema):

CREATE TABLE department (
    name VARCHAR(30)

CREATE TABLE employee (
    name VARCHAR(30),
    dep_id INTEGER REFERENCES department(id)

The relationship from employee to department is many to one, since many employee records can be associated with a single department. A SQLAlchemy mapping might look like:

class Department(Base):
    __tablename__ = "department"
    id = Column(Integer, primary_key=True)
    name = Column(String(30))

class Employee(Base):
    __tablename__ = "employee"
    id = Column(Integer, primary_key=True)
    name = Column(String(30))
    dep_id = Column(Integer, ForeignKey(""))
    department = relationship("Department")
mapped class

We say a class is “mapped” when it has been passed through the mapper() function. This process associates the class with a database table or other selectable construct, so that instances of it can be persisted and loaded using a Session.

data marshalling

The process of transforming the memory representation of an object to a data format suitable for storage or transmission to another part of a system, when data must be moved between different parts of a computer program or from one program to another. In terms of SQLAlchemy, we often need to “marshal” data into a format appropriate for passing into the relational database.

See also

Marshalling (via Wikipedia)

Augmenting Existing Types - SQLAlchemy’s TypeDecorator is commonly used for data marshalling as data is sent into the database for INSERT and UPDATE statements, and “unmarshalling” data as it is retrieved using SELECT statements.

database metadata
table metadata

The term “metadata” generally refers to “data that describes data”; data that itself represents the format and/or structure of some other kind of data. In SQLAlchemy, the term “metadata” typically refers to the MetaData construct, which is a collection of information about the tables, columns, constraints, and other DDL objects that may exist in a particular database.

method chaining

An object-oriented technique whereby the state of an object is constructed by calling methods on the object. The object features any number of methods, each of which return a new object (or in some cases the same object) with additional state added to the object.

The two SQLAlchemy objects that make the most use of method chaining are the Select object and the Query object. For example, a Select object can be assigned two expressions to its WHERE clause as well as an ORDER BY clause by calling upon the Select.where() and Select.order_by() methods:

stmt = select(\
            where( > 5).\

Each method call above returns a copy of the original Select object with additional qualifiers added.

See also


N plus one problem
N plus one

The N plus one problem is a common side effect of the lazy load pattern, whereby an application wishes to iterate through a related attribute or collection on each member of a result set of objects, where that attribute or collection is set to be loaded via the lazy load pattern. The net result is that a SELECT statement is emitted to load the initial result set of parent objects; then, as the application iterates through each member, an additional SELECT statement is emitted for each member in order to load the related attribute or collection for that member. The end result is that for a result set of N parent objects, there will be N + 1 SELECT statements emitted.

The N plus one problem is alleviated using eager loading.

one to many

A style of relationship() which links the primary key of the parent mapper’s table to the foreign key of a related table. Each unique parent object can then refer to zero or more unique related objects.

The related objects in turn will have an implicit or explicit many to one relationship to their parent object.

An example one to many schema (which, note, is identical to the many to one schema):

CREATE TABLE department (
    name VARCHAR(30)

CREATE TABLE employee (
    name VARCHAR(30),
    dep_id INTEGER REFERENCES department(id)

The relationship from department to employee is one to many, since many employee records can be associated with a single department. A SQLAlchemy mapping might look like:

class Department(Base):
    __tablename__ = "department"
    id = Column(Integer, primary_key=True)
    name = Column(String(30))
    employees = relationship("Employee")

class Employee(Base):
    __tablename__ = "employee"
    id = Column(Integer, primary_key=True)
    name = Column(String(30))
    dep_id = Column(Integer, ForeignKey(""))

This describes one of the major object states which an object can have within a Session; a pending object is a new object that doesn’t have any database identity, but has been recently associated with a session. When the session emits a flush and the row is inserted, the object moves to the persistent state.


This describes one of the major object states which an object can have within a Session; a persistent object is an object that has a database identity (i.e. a primary key) and is currently associated with a session. Any object that was previously pending and has now been inserted is in the persistent state, as is any object that’s been loaded by the session from the database. When a persistent object is removed from a session, it is known as detached.


“plugin-enabled” or “plugin-specific” generally indicates a function or method in SQLAlchemy Core which will behave differently when used in an ORM context.

SQLAlchemy allows Core constructs such as Select objects to participate in a “plugin” system, which can inject additional behaviors and features into the object that are not present by default.

Specifically, the primary “plugin” is the “orm” plugin, which is at the base of the system that the SQLAlchemy ORM makes use of Core constructs in order to compose and execute SQL queries that return ORM results.


Refers to a function that handles several types at once. In SQLAlchemy, the term is usually applied to the concept of an ORM mapped class whereby a query operation will return different subclasses based on information in the result set, typically by checking the value of a particular column in the result known as the discriminator.

Polymorphic loading in SQLAlchemy implies that a one or a combination of three different schemes are used to map a hierarchy of classes; “joined”, “single”, and “concrete”. The section Mapping Class Inheritance Hierarchies describes inheritance mapping fully.

primary key
primary key constraint

A constraint that uniquely defines the characteristics of each row in a table. The primary key has to consist of characteristics that cannot be duplicated by any other row. The primary key may consist of a single attribute or multiple attributes in combination. (via Wikipedia)

The primary key of a table is typically, though not always, defined within the CREATE TABLE DDL:

CREATE TABLE employee (
     emp_id INTEGER,
     emp_name VARCHAR(30),
     dep_id INTEGER,
     PRIMARY KEY (emp_id)
read committed

One of the four database isolation levels, read committed features that the transaction will not be exposed to any data from other concurrent transactions that has not been committed yet, preventing so-called “dirty reads”. However, under read committed there can be non-repeatable reads, meaning data in a row may change when read a second time if another transaction has committed changes.

read uncommitted

One of the four database isolation levels, read uncommitted features that changes made to database data within a transaction will not become permanent until the transaction is committed. However, within read uncommitted, it may be possible for data that is not committed in other transactions to be viewable within the scope of another transaction; these are known as “dirty reads”.


In SQLAlchemy, this term refers to the feature of querying a database’s schema catalogs in order to load information about existing tables, columns, constraints, and other constructs. SQLAlchemy includes features that can both provide raw data for this information, as well as that it can construct Core/ORM usable Table objects from database schema catalogs automatically.

See also

Reflecting Database Objects - complete background on database reflection.


An object, typically globally accessible, that contains long-lived information about some program state that is generally useful to many parts of a program.

relational algebra

An algebraic system developed by Edgar F. Codd that is used for modelling and querying the data stored in relational databases.


A connecting unit between two mapped classes, corresponding to some relationship between the two tables in the database.

The relationship is defined using the SQLAlchemy function relationship(). Once created, SQLAlchemy inspects the arguments and underlying mappings involved in order to classify the relationship as one of three types: one to many, many to one, or many to many. With this classification, the relationship construct handles the task of persisting the appropriate linkages in the database in response to in-memory object associations, as well as the job of loading object references and collections into memory based on the current linkages in the database.


In the context of SQLAlchemy, the term “released” refers to the process of ending the usage of a particular database connection. SQLAlchemy features the usage of connection pools, which allows configurability as to the lifespan of database connections. When using a pooled connection, the process of “closing” it, i.e. invoking a statement like connection.close(), may have the effect of the connection being returned to an existing pool, or it may have the effect of actually shutting down the underlying TCP/IP connection referred to by that connection - which one takes place depends on configuration as well as the current state of the pool. So we used the term released instead, to mean “do whatever it is you do with connections when we’re done using them”.

The term will sometimes be used in the phrase, “release transactional resources”, to indicate more explicitly that what we are actually “releasing” is any transactional state which as accumulated upon the connection. In most situations, the process of selecting from tables, emitting updates, etc. acquires isolated state upon that connection as well as potential row or table locks. This state is all local to a particular transaction on the connection, and is released when we emit a rollback. An important feature of the connection pool is that when we return a connection to the pool, the connection.rollback() method of the DBAPI is called as well, so that as the connection is set up to be used again, it’s in a “clean” state with no references held to the previous series of operations.

repeatable read

One of the four database isolation levels, repeatable read features all of the isolation of read committed, and additionally features that any particular row that is read within a transaction is guaranteed from that point to not have any subsequent external changes in value (i.e. from other concurrent UPDATE statements) for the duration of that transaction.


This is a non-SQL standard clause provided in various forms by certain backends, which provides the service of returning a result set upon execution of an INSERT, UPDATE or DELETE statement. Any set of columns from the matched rows can be returned, as though they were produced from a SELECT statement.

The RETURNING clause provides both a dramatic performance boost to common update/select scenarios, including retrieval of inline- or default- generated primary key values and defaults at the moment they were created, as well as a way to get at server-generated default values in an atomic way.

An example of RETURNING, idiomatic to PostgreSQL, looks like:

INSERT INTO user_account (name) VALUES ('new name') RETURNING id, timestamp

Above, the INSERT statement will provide upon execution a result set which includes the values of the columns and user_account.timestamp, which above should have been generated as default values as they are not included otherwise (but note any series of columns or SQL expressions can be placed into RETURNING, not just default-value columns).

The backends that currently support RETURNING or a similar construct are PostgreSQL, SQL Server, Oracle, and Firebird. The PostgreSQL and Firebird implementations are generally full featured, whereas the implementations of SQL Server and Oracle have caveats. On SQL Server, the clause is known as “OUTPUT INSERTED” for INSERT and UPDATE statements and “OUTPUT DELETED” for DELETE statements; the key caveat is that triggers are not supported in conjunction with this keyword. On Oracle, it is known as “RETURNING…INTO”, and requires that the value be placed into an OUT parameter, meaning not only is the syntax awkward, but it can also only be used for one row at a time.

SQLAlchemy’s UpdateBase.returning() system provides a layer of abstraction on top of the RETURNING systems of these backends to provide a consistent interface for returning columns. The ORM also includes many optimizations that make use of RETURNING when available.


A term used in SQLAlchemy to describe a SQL construct that represents a collection of rows. It’s largely similar to the concept of a “relation” in relational algebra. In SQLAlchemy, objects that subclass the Selectable class are considered to be usable as “selectables” when using SQLAlchemy Core. The two most common constructs are that of the Table and that of the Select statement.


One of the four database isolation levels, serializable features all of the isolation of repeatable read, and additionally within a lock-based approach guarantees that so-called “phantom reads” cannot occur; this means that rows which are INSERTed or DELETEd within the scope of other transactions will not be detectable within this transaction. A row that is read within this transaction is guaranteed to continue existing, and a row that does not exist is guaranteed that it cannot appear of inserted from another transaction.

Serializable isolation typically relies upon locking of rows or ranges of rows in order to achieve this effect and can increase the chance of deadlocks and degrade performance. There are also non-lock based schemes however these necessarily rely upon rejecting transactions if write collisions are detected.


The container or scope for ORM database operations. Sessions load instances from the database, track changes to mapped instances and persist changes in a single unit of work when flushed.

scalar subquery

Refers to a SELECT statement that is embedded within an enclosing SELECT.

A subquery comes in two general flavors, one known as a “scalar select” which specifically must return exactly one row and one column, and the other form which acts as a “derived table” and serves as a source of rows for the FROM clause of another select. A scalar select is eligible to be placed in the WHERE clause, columns clause, ORDER BY clause or HAVING clause of the enclosing select, whereas the derived table form is eligible to be placed in the FROM clause of the enclosing SELECT.


  1. a scalar subquery placed in the columns clause of an enclosing SELECT. The subquery in this example is a correlated subquery because part of the rows which it selects from are given via the enclosing statement.

    SELECT id, (SELECT name FROM address WHERE
    FROM user
  2. a scalar subquery placed in the WHERE clause of an enclosing SELECT. This subquery in this example is not correlated as it selects a fixed result.

    SELECT id, name FROM user
    WHERE status=(SELECT status_id FROM status_code WHERE code='C')
  3. a derived table subquery placed in the FROM clause of an enclosing SELECT. Such a subquery is almost always given an alias name.

    SELECT,, ad_subq.email_address
        user JOIN
        (select user_id, email_address FROM address WHERE address_type='Q') AS ad_subq
        ON = ad_subq.user_id

This describes one of the major object states which an object can have within a Session; a transient object is a new object that doesn’t have any database identity and has not been associated with a session yet. When the object is added to the session, it moves to the pending state.

unique constraint
unique key index

A unique key index can uniquely identify each row of data values in a database table. A unique key index comprises a single column or a set of columns in a single database table. No two distinct rows or data records in a database table can have the same data value (or combination of data values) in those unique key index columns if NULL values are not used. Depending on its design, a database table may have many unique key indexes but at most one primary key index.

(via Wikipedia)

unit of work

A software architecture where a persistence system such as an object relational mapper maintains a list of changes made to a series of objects, and periodically flushes all those pending changes out to the database.

SQLAlchemy’s Session implements the unit of work pattern, where objects that are added to the Session using methods like Session.add() will then participate in unit-of-work style persistence.

For a walk-through of what unit of work persistence looks like in SQLAlchemy, start with the section Data Manipulation with the ORM in the SQLAlchemy 1.4 / 2.0 Tutorial. Then for more detail, see Basics of Using a Session in the general reference documentation.

version id column

In SQLAlchemy, this refers to the use of a particular table column that tracks the “version” of a particular row, as the row changes values. While there are different kinds of relational patterns that make use of a “version id column” in different ways, SQLAlchemy’s ORM includes a particular feature that allows for such a column to be configured as a means of testing for stale data when a row is being UPDATEd with new information. If the last known “version” of this column does not match that of the row when we try to put new data into the row, we know that we are acting on stale information.

There are also other ways of storing “versioned” rows in a database, often referred to as “temporal” data. In addition to SQLAlchemy’s versioning feature, a few more examples are also present in the documentation, see the links below.

See also

Configuring a Version Counter - SQLAlchemy’s built-in version id feature.

Versioning Objects - other examples of mappings that version rows temporally.

WHERE clause

The portion of the SELECT statement which indicates criteria by which rows should be filtered. It is a single SQL expression which follows the keyword WHERE.

FROM user_account
WHERE = 'fred' AND user_account.status = 'E'

Above, the phrase WHERE = 'fred' AND user_account.status = 'E' comprises the WHERE clause of the SELECT.