Working with Transactions and the DBAPI¶
With the Engine
object ready to go, we may now proceed
to dive into the basic operation of an Engine
and
its primary interactive endpoints, the Connection
and
Result
. We will additionally introduce the ORM’s
facade for these objects, known as the Session
.
Note to ORM readers
When using the ORM, the Engine
is managed by another
object called the Session
. The Session
in
modern SQLAlchemy emphasizes a transactional and SQL execution pattern that
is largely identical to that of the Connection
discussed
below, so while this subsection is Core-centric, all of the concepts here
are essentially relevant to ORM use as well and is recommended for all ORM
learners. The execution pattern used by the Connection
will be contrasted with that of the Session
at the end
of this section.
As we have yet to introduce the SQLAlchemy Expression Language that is the
primary feature of SQLAlchemy, we will make use of one simple construct within
this package called the text()
construct, which allows us to write
SQL statements as textual SQL. Rest assured that textual SQL in
day-to-day SQLAlchemy use is by far the exception rather than the rule for most
tasks, even though it always remains fully available.
Getting a Connection¶
The sole purpose of the Engine
object from a user-facing
perspective is to provide a unit of
connectivity to the database called the Connection
. When
working with the Core directly, the Connection
object
is how all interaction with the database is done. As the Connection
represents an open resource against the database, we want to always limit
the scope of our use of this object to a specific context, and the best
way to do that is by using Python context manager form, also known as
the with statement.
Below we illustrate “Hello World”, using a textual SQL statement. Textual
SQL is emitted using a construct called text()
that will be discussed
in more detail later:
>>> from sqlalchemy import text
>>> with engine.connect() as conn:
... result = conn.execute(text("select 'hello world'"))
... print(result.all())
BEGIN (implicit)
select 'hello world'
[...] ()
[('hello world',)]
ROLLBACK
In the above example, the context manager provided for a database connection
and also framed the operation inside of a transaction. The default behavior of
the Python DBAPI includes that a transaction is always in progress; when the
scope of the connection is released, a ROLLBACK is emitted to end the
transaction. The transaction is not committed automatically; when we want
to commit data we normally need to call Connection.commit()
as we’ll see in the next section.
Tip
“autocommit” mode is available for special cases. The section Setting Transaction Isolation Levels including DBAPI Autocommit discusses this.
The result of our SELECT was also returned in an object called
Result
that will be discussed later, however for the moment
we’ll add that it’s best to ensure this object is consumed within the
“connect” block, and is not passed along outside of the scope of our connection.
Committing Changes¶
We just learned that the DBAPI connection is non-autocommitting. What if
we want to commit some data? We can alter our above example to create a
table and insert some data, and the transaction is then committed using
the Connection.commit()
method, invoked inside the block
where we acquired the Connection
object:
# "commit as you go"
>>> with engine.connect() as conn:
... conn.execute(text("CREATE TABLE some_table (x int, y int)"))
... conn.execute(
... text("INSERT INTO some_table (x, y) VALUES (:x, :y)"),
... [{"x": 1, "y": 1}, {"x": 2, "y": 4}],
... )
... conn.commit()
BEGIN (implicit)
CREATE TABLE some_table (x int, y int)
[...] ()
<sqlalchemy.engine.cursor.CursorResult object at 0x...>
INSERT INTO some_table (x, y) VALUES (?, ?)
[...] ((1, 1), (2, 4))
<sqlalchemy.engine.cursor.CursorResult object at 0x...>
COMMIT
Above, we emitted two SQL statements that are generally transactional, a
“CREATE TABLE” statement [1] and an “INSERT” statement that’s parameterized
(the parameterization syntax above is discussed a few sections below in
Sending Multiple Parameters). As we want the work we’ve done to be
committed within our block, we invoke the
Connection.commit()
method which commits the transaction. After
we call this method inside the block, we can continue to run more SQL
statements and if we choose we may call Connection.commit()
again for subsequent statements. SQLAlchemy refers to this style as commit as
you go.
There is also another style of committing data, which is that we can declare
our “connect” block to be a transaction block up front. For this mode of
operation, we use the Engine.begin()
method to acquire the
connection, rather than the Engine.connect()
method. This method
will both manage the scope of the Connection
and also
enclose everything inside of a transaction with COMMIT at the end, assuming
a successful block, or ROLLBACK in case of exception raise. This style
may be referred towards as begin once:
# "begin once"
>>> with engine.begin() as conn:
... conn.execute(
... text("INSERT INTO some_table (x, y) VALUES (:x, :y)"),
... [{"x": 6, "y": 8}, {"x": 9, "y": 10}],
... )
BEGIN (implicit)
INSERT INTO some_table (x, y) VALUES (?, ?)
[...] ((6, 8), (9, 10))
<sqlalchemy.engine.cursor.CursorResult object at 0x...>
COMMIT
“Begin once” style is often preferred as it is more succinct and indicates the intention of the entire block up front. However, within this tutorial we will normally use “commit as you go” style as it is more flexible for demonstration purposes.
Basics of Statement Execution¶
We have seen a few examples that run SQL statements against a database, making
use of a method called Connection.execute()
, in conjunction with
an object called text()
, and returning an object called
Result
. In this section we’ll illustrate more closely the
mechanics and interactions of these components.
Most of the content in this section applies equally well to modern ORM
use when using the Session.execute()
method, which works
very similarly to that of Connection.execute()
, including that
ORM result rows are delivered using the same Result
interface used by Core.
Fetching Rows¶
We’ll first illustrate the Result
object more closely by
making use of the rows we’ve inserted previously, running a textual SELECT
statement on the table we’ve created:
>>> with engine.connect() as conn:
... result = conn.execute(text("SELECT x, y FROM some_table"))
... for row in result:
... print(f"x: {row.x} y: {row.y}")
BEGIN (implicit)
SELECT x, y FROM some_table
[...] ()
x: 1 y: 1
x: 2 y: 4
x: 6 y: 8
x: 9 y: 10
ROLLBACK
Above, the “SELECT” string we executed selected all rows from our table.
The object returned is called Result
and represents an
iterable object of result rows.
Result
has lots of methods for
fetching and transforming rows, such as the Result.all()
method illustrated previously, which returns a list of all Row
objects. It also implements the Python iterator interface so that we can
iterate over the collection of Row
objects directly.
The Row
objects themselves are intended to act like Python
named tuples.
Below we illustrate a variety of ways to access rows.
Tuple Assignment - This is the most Python-idiomatic style, which is to assign variables to each row positionally as they are received:
result = conn.execute(text("select x, y from some_table")) for x, y in result: # ...
Integer Index - Tuples are Python sequences, so regular integer access is available too:
result = conn.execute(text("select x, y from some_table")) for row in result: x = row[0]
Attribute Name - As these are Python named tuples, the tuples have dynamic attribute names matching the names of each column. These names are normally the names that the SQL statement assigns to the columns in each row. While they are usually fairly predictable and can also be controlled by labels, in less defined cases they may be subject to database-specific behaviors:
result = conn.execute(text("select x, y from some_table")) for row in result: y = row.y # illustrate use with Python f-strings print(f"Row: {row.x} {y}")
Mapping Access - To receive rows as Python mapping objects, which is essentially a read-only version of Python’s interface to the common
dict
object, theResult
may be transformed into aMappingResult
object using theResult.mappings()
modifier; this is a result object that yields dictionary-likeRowMapping
objects rather thanRow
objects:result = conn.execute(text("select x, y from some_table")) for dict_row in result.mappings(): x = dict_row["x"] y = dict_row["y"]
Sending Parameters¶
SQL statements are usually accompanied by data that is to be passed with the
statement itself, as we saw in the INSERT example previously. The
Connection.execute()
method therefore also accepts parameters,
which are referred towards as bound parameters. A rudimentary example
might be if we wanted to limit our SELECT statement only to rows that meet a
certain criteria, such as rows where the “y” value were greater than a certain
value that is passed in to a function.
In order to achieve this such that the SQL statement can remain fixed and
that the driver can properly sanitize the value, we add a WHERE criteria to
our statement that names a new parameter called “y”; the text()
construct accepts these using a colon format “:y
”. The actual value for
“:y
” is then passed as the second argument to
Connection.execute()
in the form of a dictionary:
>>> with engine.connect() as conn:
... result = conn.execute(text("SELECT x, y FROM some_table WHERE y > :y"), {"y": 2})
... for row in result:
... print(f"x: {row.x} y: {row.y}")
BEGIN (implicit)
SELECT x, y FROM some_table WHERE y > ?
[...] (2,)
x: 2 y: 4
x: 6 y: 8
x: 9 y: 10
ROLLBACK
In the logged SQL output, we can see that the bound parameter :y
was
converted into a question mark when it was sent to the SQLite database.
This is because the SQLite database driver uses a format called “qmark parameter style”,
which is one of six different formats allowed by the DBAPI specification.
SQLAlchemy abstracts these formats into just one, which is the “named” format
using a colon.
Always use bound parameters
As mentioned at the beginning of this section, textual SQL is not the usual way we work with SQLAlchemy. However, when using textual SQL, a Python literal value, even non-strings like integers or dates, should never be stringified into SQL string directly; a parameter should always be used. This is most famously known as how to avoid SQL injection attacks when the data is untrusted. However it also allows the SQLAlchemy dialects and/or DBAPI to correctly handle the incoming input for the backend. Outside of plain textual SQL use cases, SQLAlchemy’s Core Expression API otherwise ensures that Python literal values are passed as bound parameters where appropriate.
Sending Multiple Parameters¶
In the example at Committing Changes, we executed an INSERT
statement where it appeared that we were able to INSERT multiple rows into the
database at once. For statements that operate upon data, but do not return
result sets, namely DML statements such as “INSERT” which don’t
include a phrase like “RETURNING”, we can send multi params to the
Connection.execute()
method by passing a list of dictionaries
instead of a single dictionary, thus allowing the single SQL statement to
be invoked against each parameter set individually:
>>> with engine.connect() as conn:
... conn.execute(
... text("INSERT INTO some_table (x, y) VALUES (:x, :y)"),
... [{"x": 11, "y": 12}, {"x": 13, "y": 14}],
... )
... conn.commit()
BEGIN (implicit)
INSERT INTO some_table (x, y) VALUES (?, ?)
[...] ((11, 12), (13, 14))
<sqlalchemy.engine.cursor.CursorResult object at 0x...>
COMMIT
Behind the scenes, the Connection
objects uses a DBAPI feature
known as cursor.executemany(). This method performs the
equivalent operation of invoking the given SQL statement against each parameter
set individually. The DBAPI may optimize this operation in a variety of ways,
by using prepared statements, or by concatenating the parameter sets into a
single SQL statement in some cases. Some SQLAlchemy dialects may also use
alternate APIs for this case, such as the psycopg2 dialect for PostgreSQL which uses more performant APIs
for this use case.
Tip
you may have noticed this section isn’t tagged as an ORM concept. That’s because the “multiple parameters” use case is usually used for INSERT statements, which when using the ORM are invoked in a different way. Multiple parameters also may be used with UPDATE and DELETE statements to emit distinct UPDATE/DELETE operations on a per-row basis, however again when using the ORM, there is a different technique generally used for updating or deleting many individual rows separately.
Executing with an ORM Session¶
As mentioned previously, most of the patterns and examples above apply to use with the ORM as well, so here we will introduce this usage so that as the tutorial proceeds, we will be able to illustrate each pattern in terms of Core and ORM use together.
The fundamental transactional / database interactive object when using the
ORM is called the Session
. In modern SQLAlchemy, this object
is used in a manner very similar to that of the Connection
,
and in fact as the Session
is used, it refers to a
Connection
internally which it uses to emit SQL.
When the Session
is used with non-ORM constructs, it
passes through the SQL statements we give it and does not generally do things
much differently from how the Connection
does directly, so
we can illustrate it here in terms of the simple textual SQL
operations we’ve already learned.
The Session
has a few different creational patterns, but
here we will illustrate the most basic one that tracks exactly with how
the Connection
is used which is to construct it within
a context manager:
>>> from sqlalchemy.orm import Session
>>> stmt = text("SELECT x, y FROM some_table WHERE y > :y ORDER BY x, y")
>>> with Session(engine) as session:
... result = session.execute(stmt, {"y": 6})
... for row in result:
... print(f"x: {row.x} y: {row.y}")
BEGIN (implicit)
SELECT x, y FROM some_table WHERE y > ? ORDER BY x, y
[...] (6,)
x: 6 y: 8
x: 9 y: 10
x: 11 y: 12
x: 13 y: 14
ROLLBACK
The example above can be compared to the example in the preceding section
in Sending Parameters - we directly replace the call to
with engine.connect() as conn
with with Session(engine) as session
,
and then make use of the Session.execute()
method just like we
do with the Connection.execute()
method.
Also, like the Connection
, the Session
features
“commit as you go” behavior using the Session.commit()
method,
illustrated below using a textual UPDATE statement to alter some of
our data:
>>> with Session(engine) as session:
... result = session.execute(
... text("UPDATE some_table SET y=:y WHERE x=:x"),
... [{"x": 9, "y": 11}, {"x": 13, "y": 15}],
... )
... session.commit()
BEGIN (implicit)
UPDATE some_table SET y=? WHERE x=?
[...] ((11, 9), (15, 13))
COMMIT
Above, we invoked an UPDATE statement using the bound-parameter, “executemany” style of execution introduced at Sending Multiple Parameters, ending the block with a “commit as you go” commit.
Tip
The Session
doesn’t actually hold onto the
Connection
object after it ends the transaction. It
gets a new Connection
from the Engine
the next time it needs to execute SQL against the database.
The Session
obviously has a lot more tricks up its sleeve
than that, however understanding that it has a Session.execute()
method that’s used the same way as Connection.execute()
will
get us started with the examples that follow later.
See also
Basics of Using a Session - presents basic creational and usage patterns with
the Session
object.
SQLAlchemy 1.4 / 2.0 Tutorial
Next Tutorial Section: Working with Database Metadata