预制查询

1.1 新版功能.

Baked queries are used to boost execution performance for constantly-used queries. Similar to the Baked Queries in SQLAlchemy, GINO could also cache the object’s construction and string-compilation steps. Furthermore, GINO automatically manages a prepared statement for each baked query in every active connection in the pool. Executing baked queries is at least 40% faster than running normal queries, but you need to bake them before creating the engine.

GINO provides two approaches for baked queries:

1. Low-level Bakery API

2. High-level Gino.bake() integration

Use Bakery with Bare Engine

First, we need a bakery:

import gino

bakery = gino.Bakery()

Then, let's bake some queries:

db_time = bakery.bake("SELECT now()")

Or queries with parameters:

user_query = bakery.bake("SELECT * FROM users WHERE id = :uid")

Let's assume we have this users table defined in SQLAlchemy Core:

import sqlalchemy as sa

metadata = sa.MetaData()
user_table = sa.Table(
    "users", metadata,
    sa.Column("id", sa.Integer, primary_key=True),
    sa.Column("name", sa.String),
)

Now we can bake a similar query with SQLAlchemy Core:

user_query = bakery.bake(
    sa.select([user_table]).where(user.c.id == sa.bindparam("uid"))
)

These baked queries are usually global, and supposed to be shared across the application. To run them, we need an engine with the bakery:

engine = await gino.create_engine("postgresql://localhost/", bakery=bakery)

By doing so, GINO will bake the queries in the bakery. As new connections are added to the DB pool, the prepared statements are automatically created behind the scene.

To execute the baked queries, you could treat the BakedQuery instances as if they are the queries themselves, for example:

now = await engine.scalar(db_time)

Pass in parameter values:

row = await engine.first(user_query, uid=123)

Use the Gino Integration

In a more common scenario, there will be a Gino instance, which has usually a bind set - either explicitly or by the Web framework extensions:

from gino import Gino

db = Gino()

async def main():
    async with db.with_bind("postgresql://localhost/"):
        ...

A Bakery is automatically created in the db instance, and fed to the engine implicitly. You can immediately start to bake queries without further ado:

class User(db.Model):
    __tablename__ = "users"

    id = db.Column(db.Integer, primary_key=True)
    name = db.Column(db.String)

db_time = db.bake("SELECT now()")
user_getter = db.bake(User.query.where(User.id == db.bindparam("uid")))

And the execution is also simplified with the same bind magic:

async def main():
    async with db.with_bind("postgresql://localhost/"):
        print(await db_time.scalar())

        user: User = await user_getter.first(uid=1)
        print(user.name)

To make things easier, you could even define the baked queries directly on the model:

class User(db.Model):
    __tablename__ = "users"

    id = db.Column(db.Integer, primary_key=True)
    name = db.Column(db.String)

    @db.bake
    def getter(cls):
        return cls.query.where(cls.id == db.bindparam("uid"))

    @classmethod
    async def get(cls, uid):
        return await cls.getter.one_or_none(uid=uid)

Here GINO treats the getter() as a declared_attr() with with_table=True, therefore it takes one positional argument cls for the User class.

How to customize loaders?

If possible, you could bake the additional execution options into the query:

user_getter = db.bake(
    User.query.where(User.id == db.bindparam("uid")).execution_options(
        loader=User.load(comment="Added by loader.")
    )
)

The bake() method accepts keyword arguments as execution options to e.g. simplify the example above into:

user_getter = db.bake(
    User.query.where(User.id == db.bindparam("uid")),
    loader=User.load(comment="Added by loader."),
)

If the query construction is complex, bake() could also be used as a decorator:

@db.bake
def user_getter():
    return User.query.where(User.id == db.bindparam("uid")).execution_options(
        loader=User.load(comment="Added by loader.")
    )

Or with short execution options:

@db.bake(loader=User.load(comment="Added by loader."))
def user_getter():
    return User.query.where(User.id == db.bindparam("uid"))

Meanwhile, it is also possible to override the loader at runtime:

user: User = await user_getter.load(User).first(uid=1)
print(user.name)  # no more comment on user!

提示

This override won't affect the baked query - it's used only in this execution.

What APIs are available on BakedQuery?

BakedQuery is a GinoExecutor, so it inherited all the APIs like all(), first(), one(), one_or_none(), scalar(), status(), load(), timeout(), etc.

GinoExecutor is actually the chained .gino helper API seen usually in queries like this:

user = await User.query.where(User.id == 123).gino.first()

So a BakedQuery can be seen as a normal query with the .gino suffix, plus it is directly executable.

参见

Please see API document of gino.bakery for more information.

I don't want the prepared statements.

If you don't need all the baked queries (m) to create prepared statements for all the active database connections (n) in the beginning, you could set prebake=False in the engine initialization to prevent the default initial m x n prepare calls:

e = await gino.create_engine("postgresql://...", bakery=bakery, prebake=False)

Or if you're using bind:

await db.set_bind("postgresql://...", prebake=False)

This is useful when you're depending on db.gino.create_all() to create the tables, because the prepared statements can only be created after the table creation.

The prepared statements will then be created and cached lazily on demand.