Skip to main content

Embrace SQL keeps your SQL queries in SQL files. An anti-ORM inspired by HugSQL and PugSQL

Project description

Does writing complex queries in an ORM feel like driving with the handbrake on? Embrace SQL! Put your SQL queries in regular .sql files, and embrace will load them.


pip install embrace


import embrace

# Connect to your database, using any db-api connector.
# If python supports it, so does embrace.
conn = psycopg2.connect("postgresql:///mydb")

# Create a module populated with queries from a collection of *.sql files:
queries = embrace.module("resources/sql")

# Run a query
users = queries.list_users(conn, order_by='created_at')

Your query would be specified like this:

-- :name list_users :many
select * from users where active = :active order by :identifier:order_by

Embrace returns rows using the underlying db-api cursor. Most db-api libraries have cursor types that return dicts or namedtuples. For example in Postgresql you could do this:

conn = psycopg2.connect(

What is the format of a query SQL file?

Embrace-SQL tries to stick close to the format used by HugSQL and PugSQL. SQL files normally contain special comments to specify the query name and result type, and an SQL query:

-- :name get_user_count
-- :result :scalar
SELECT count(1) FROM users

If a result type is omitted, it will default to cursor. Also, the result type can be included directly after the name:

-- :name get_user_count :scalar

If :name is omitted, it will default to the filename without the file extension.

A single file may contain multiple queries, separated by a structured SQL comment. For example to create two query objects accessible as queries.list_users() and queries.get_user_by_id():

-- :name list_users :many
select * from users

-- :name get_user_by_id :one
select * from users where id=:id

But if you don’t have the separating comment, embrace-sql can run multiple statements in a single query call, returning the result from just the last one.

Why? Because it makes this possible in MySQL:

-- :result :column
insert into users (name, email) values (:name, :email);
select last_insert_id();

What can queries return?

The following result types are supported:

:affected, :n

The number of rows affected


The first row, as returned by cursor.fetchone(), or None if no row is found.

:one, :1

A single row, as returned by cursor.fetchone(), usually as a tuple (but most db-api modules have extensions allowing you to access rows as dicts or named tuples.

If no row is generated by the query, embrace.exceptions.NoResultFound will be raised. If more than one row is generated by the query, embrace.exceptions.MultipleResultsFound will be raised.

exactly-one`, ``:=1

Synonyms for :one, retained for compatibility


As one, but returns None if no row is returned by the query.

:many, :*

An iterator over a number of rows. Each row will be the value returned by cursor.fetchone(), usually a tuple.

:cursor, :raw

The cursor object.


The value of the first column of the first row returned by the query.

If no row is generated by the query, a NoResultFound will be raised.


An iterator over the values in the first column returned.

You can override the return type specified by the query from Python code by using one of the following methods on the Query object:

  • affected

  • one

  • exactlyone

  • many

  • cursor

  • scalar

  • column

How do I map rows onto objects?

Embrace lets you do very simple ORM style mapping.


import embrace
from dataclasses import dataclass

class User:
    id: int
    name: str

query = queries.query("SELECT * from users").returning(User)
users = query.many(conn)

You can also map multiple classes in a single query, for example:

query = queries.query(
    "SELECT * from posts join users on posts.user_id ="
).returning((Post, User))
for post, user in query.many(conn):

By default embrace will look for fields named id (case insensitive) to split up the row. If you need to split the row differently, pass a split_on argument:

query = queries.query(
    "SELECT * from posts join users on posts.user_id ="
).returning((Post, User), split_on=['post_id', 'user_id'])
for post, user in query.many(conn):

You can also tell embrace to populate join relationships:

from embrace import one_to_many
from embrace import one_to_one

query = queries.query(
    SELECT users.*, orders.*, products.*
    FROM users
    JOIN orders ON orders.user_id =
    JOIN products ON orders.product_id =
    # Each row of this query returns data for a User, Order and Product
    # object
    (User, Order, Product),
        # Populate User.orders with the list of Order objects
        one_to_many(User, 'orders', Order),

        # Populate Order.product with the product object
        one_to_one(Order, 'product', Product),

    # Construct objects using keyword arguments based on returned column
    # names (this is the default)

    # Rows with the same key column values are guaranteed to map to the
    # same object within the scope of this query.
    key_columns=[('id',), ('id',), ('id',)]

for user in query.many(conn):
    for order in user.order:
        product = order.product

How do parameters work?

Placeholders inserted using the :name syntax are escaped by the db-api driver:

-- Outputs `select * from user where name = 'o''brien'`;
select * from users where name = :name

You can interpolate lists and tuples too:

:tuple: creates a placeholder like this (?, ?, ?)

:value*: creates a placeholder like this ?, ?, ?

:tuple*: creates a placeholder like this (?, ?, ?), (?, ?, ?), (useful for multiple insert queries)

-- Call this with `queries.insert_foo(data=(1, 2, 3))`
INSERT INTO foo (a, b, c) VALUES :tuple:data

-- Call this with `queries.get_matching_users(names=("carolyn", "douglas"))`
SELECT * from users WHERE name in (:value*:names)

You can escape identifiers with :identifier:, like this:

-- Outputs `select * from "some random table"`
select * from :identifier:table_name

You can pass through raw sql too. This leaves you open to SQL injection attacks if you allow user input into such parameters:

-- Outputs `select * from users order by name desc`
select * from users order by :raw:order_clause

How do I handle connections? Transactions?

You must pass a db-api connection object every time you call a query. You can manage these connections yourself, but Embrace also offers a connection pooling module.

from embrace import pool

# Create a connection pool
connection_pool = pool.ConnectionPool(
    partial(psycopg2.connect, database='mydb'),

# Example 1 - explicit calls to getconn/release
conn = connection_pool.getconn()

# Example 2 - context manager
with connection_pool.connect() as conn:

Transaction handling may be handled manually by calling commit() or rollback() on the connection object, or you can also use the transaction context run to queries in a transaction:

with queries.transaction(conn) as q:

The transaction will be commited when the with block exits, or rolled back if an exception occurred.

How do I reload queries when the underlying files change?

Pass auto_reload=True when constructing a module:

m = module('resources/sql', auto_reload=True)


Exceptions raised from the underlying db-api connection are wrapped in exception classes from embrace.exceptions, with PEP-249 compliant names. You can use this to catch exceptions, for example:

    queries.execute("SELECT 1.0 / 0.0")
except embrace.exceptions.DataError:

The original exception will be available in the __cause__ attribute of the embrace exception object.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

embrace-3.0.0.tar.gz (20.4 kB view hashes)

Uploaded Source

Built Distribution

embrace-3.0.0-py3-none-any.whl (24.5 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page