DDSQL

DDSQL is a Python library for building SQL queries with Jinja2 template rendering and database adapter support. Query results are automatically deserialized into typed models.

Installation

Install the library using pip:

pip install ddsql

Development

The project runs entirely in Docker. Requires Docker and Fabric on the host:

fab build      # build the dev image
fab tests      # run pytest
fab linters    # run ruff (with --fix), ty and complexipy
fab shell      # IPython inside the container
fab bash       # bash inside the container

This project was generated from dd-lib-stub; run copier update to pull in template updates.

Serializer

The serializer converts Python types to their SQL representations. You can use one of the built-in serializers (PostgresSerializer, ClickhouseSerializer) or create your own by inheriting from BaseSerializer.

Serialization Table

String values are escaped according to the dialect rules, so quotes, backslashes and control characters cannot break the query:

• BaseSerializer doubles single quotes (O'Brien → 'O''Brien'), as defined by the SQL standard.
• ClickhouseSerializer escapes backslashes, single quotes and all control characters with backslash sequences (O'Brien → 'O\'Brien', newline → \n, other control characters → \xHH).
• PostgresSerializer doubles single quotes; strings containing backslashes or control characters are emitted using the escape string syntax (C:\dir → E'C:\\dir'), which is interpreted the same way regardless of the standard_conforming_strings server setting. A NUL (0x00) character raises ValueError, since PostgreSQL cannot store it in text values.

The serialized literal is always a single printable line.

datetime values keep their microseconds and UTC offset:

• ClickhouseSerializer renders parseDateTime64BestEffort('...', 6), which yields a DateTime64(6) literal; a timezone-aware value is converted according to its offset, a naive value is interpreted in the server time zone. Inserting such a literal into a DateTime column silently truncates it to seconds, and comparisons with DateTime columns work as expected.
• PostgresSerializer renders a timezone-aware value as '...+03:00'::timestamptz, so its offset is honoured; a naive value is rendered as '...'::timestamp, as before.

To customize escaping in your own serializer, override the escape_string method.

If you need to serialize a type not listed in the table, override the serialize_other_object method in your serializer:

from ddsql.serializers import BaseSerializer


class CustomSerializer(BaseSerializer):
    def serialize_other_object(self, value):
        if isinstance(value, CustomType):
            return ...
        ...

To serialize values in SQL templates, wrap parameters with serialize_value:

SELECT * 
FROM users
WHERE 
    name = {{ serialize_value(name) }}
    AND created_at > {{ serialize_value(created_at) }}

To add custom functions to templates, override the template_functions property:

from ddsql.serializers import BaseSerializer


class CustomSerializer(BaseSerializer):
    @property
    def template_functions(self):
        return {
            **super().template_functions,
            'some_function': ...,
        }

Connections

ddsql.connections sits between a ScopedRegistry (from ddutils) and the code that runs queries. A registry holds one connection per scope: a pooled client per process, a session per task. The connections module adds two things on top:

• ConnectionManagerFactory(registries, default, connection_manager_class=ConnectionManager) resolves a connection alias to its registry, like Django's using. Calling it opens a connection manager on the chosen registry; registry(alias) returns the registry itself.
• ConnectionManager(registry) is the default connection manager: an async context manager that hands out the connection of the registry's current scope and releases nothing on exit. Enough for clients that are pools themselves, such as ClickHouse.

Databases with transactions pass their own connection_manager_class: a unit of work that begins, commits or rolls back and closes the session on exit. It depends on the driver, so it lives in the project.

import os
from enum import Enum

from ddutils.scoped_registry import ScopedRegistry

from ddsql.connections import ConnectionManagerFactory


class ClickhouseDB(str, Enum):
    PRIMARY = 'primary'


client_registries = {
    db: ScopedRegistry(create_func=make_client_for(db), scope_func=os.getpid, destructor_method_name='close')
    for db in ClickhouseDB
}
clickhouse = ConnectionManagerFactory(client_registries, default=ClickhouseDB.PRIMARY)


async with clickhouse() as client:
    result = await client.query(sql)


# a transactional database: `Atomic` is your unit of work over `registry.scope()`
atomic = ConnectionManagerFactory(session_registries, default=PostgresDB.PRIMARY, connection_manager_class=Atomic)


async with atomic(alias=PostgresDB.REPLICA) as session:
    ...

Adapter

Adapter encapsulates database interactions. To create an adapter, inherit from the Adapter base class and define two required elements:

• serializer – an instance of a serializer for converting Python types to SQL representations;
• _execute method – the database-specific query execution logic.

from collections.abc import Sequence
from typing import Any

from sqlalchemy import text

from ddsql.adapter import Adapter
from ddsql.serializers import PostgresSerializer


class PostgresAdapter(Adapter):
    serializer = PostgresSerializer()

    async def _execute(self) -> Sequence[dict[str, Any]]:
        async with atomic(alias=self.alias) as session:
            query = await self.get_query()  # the rendered SQL query
            result = await session.execute(text(query))
            return [dict(zip(result.keys(), row)) for row in result.fetchall()]

Intermediate adapter classes without _execute are abstract and need no serializer; only concrete adapters are validated.

Choosing the database

Adapter.using(alias) picks the connection alias right before executing, like Django's using. _execute reads self.alias (None means the default) and passes it to whatever opens the connection, as in the example above:

await SQL(query).postgres.execute()                               # default connection
await SQL(query).postgres.using(PostgresDB.REPLICA).execute()

SQLBase

SQLBase is configured once per project and defines which adapters are available for query execution. It serves as the central point that connects queries with database adapters.

Create a subclass with one or more adapters:

from ddsql.sqlbase import SQLBase
from ddsql.adapter import AdapterDescriptor


class SQL(SQLBase):
    postgres: PostgresAdapter = AdapterDescriptor(PostgresAdapter)
    clickhouse: ClickhouseAdapter = AdapterDescriptor(ClickhouseAdapter)

Execution example:

from ddsql.query import Query


query = Query(...)
result = await SQL(query=query).with_params(email='test@test.test', is_deleted=False).postgres.execute()

Query

Query knows where to get the template from and how to render a SQL query. It also handles result deserialization via the build_result method, which wraps raw database rows into the specified model (called internally by Adapter.execute).

Required parameters:

• model – a declarative class (e.g., dataclass) describing the output result structure;
• text or path – the SQL template source (inline string or path to a file).

Inline Template (text)

from ddsql.query import Query


query = Query(
    model=User,
    text='SELECT user_id, name FROM users WHERE user_id = {{ serialize_value(user_id) }}'
)

File Template (path)

Pass the path to the SQL file as Path. The file is checked at construction time, so a wrong path fails on import rather than on the first query:

from pathlib import Path

from ddsql.query import Query

SQL_TEMPLATES_DIR = Path(__file__).parent / 'templates' / 'sql'

query = Query(
    model=User,
    path=SQL_TEMPLATES_DIR / 'users' / 'get_by_id.sql',
)

{% include %} and {% import %} inside the template resolve relative to the template's directory.

Result

The result of query execution is a Result object that wraps the data into the specified model:

• get() – returns the first row as a model instance, or None if empty;
• get_list() – returns all rows as a tuple of model instances;
• rows – attribute for accessing raw data.

Complete Example

from dataclasses import dataclass
from datetime import datetime

from ddsql.adapter import Adapter, AdapterDescriptor
from ddsql.query import Query
from ddsql.serializers import PostgresSerializer
from ddsql.sqlbase import SQLBase


class PostgresAdapter(Adapter):
    serializer = PostgresSerializer()

    async def _execute(self):
        ...


class SQL(SQLBase):
    postgres: PostgresAdapter = AdapterDescriptor(PostgresAdapter)


@dataclass
class User:
    user_id: int
    name: str
    email: str | None
    created_at: datetime
    is_deleted: bool


query = Query(
    model=User,
    text='''
        SELECT *
        FROM users
        WHERE 
            created_at > {{ serialize_value(created_after) }}
        LIMIT {{ limit }}
    '''
)


async def get_users():
    result = await (
        SQL(query=query)
        .with_params(created_after=datetime(2025, 1, 1))
        .with_params(limit=10)
        .postgres
        .execute()
    )
    return result.get_list()