Micro ODM for MongoDB

Hi. Using the documented mode of sorting sort(Class.field) results in a warning in PyCharm. Is something off in the type definition. It seems to work fine.

Motor 3.0.0 was released a few days ago: https://www.mongodb.com/community/forums/t/mongodb-motor-3-0-0-released/160708. Do you plan to support the new version in the near future?

If the field type is Document subclass, then only id should be stored there and the whole subdocument must be stored to the separated collection.

Example:

class Window(Document):
  width: int
  height: int

class House(Document):
  address : str
  windows: List[Window]
  favorite_window: Window

Problems:

• fetching/ lazy fetching
• find by subfield. Example: House.find(House.favorite_window.width == 1)
• updates of the subdocument. Example: house.set({House.favorite_window.width: 1})

Describe the bug Since 1.14.0 all my documents get inserted into one collection named "Documents". All documents also got a _class_id value with .UserEntry

To Reproduce

class UserEntry(Document, ABC):
    id: int
    value: Optional[str] = None
    ...

    class Settings:
        name = 'user_data'
        use_state_management = True

user_data = await UserEntry(...)
user_data.value = 'Test'
await user_data.insert()

Expected behavior The document get's inserted into a "user_data" collection.

Additional context https://github.com/roman-right/beanie/compare/1.13.1...1.14.0

Bug in action of save method I wanna save my document to database and use it after insert but I have problem with this situation:

This is My code

class Child(BaseModel):
    child_field: str


class Sample(Document):
    field: Dict[str, Child]

  instance1 = Sample(field={"Bar": Child(child_field="Foo")})
  print(instance1)
  await instance1.save()
  print(instance1)

Expected behavior

# first print :
id=None revision_id=None field={'Bar': Child(child_field='Foo')}
# second print:
id=ObjectId('636b9d2997bb72433b944ef4') revision_id=None field={'Bar': Child(child_field='Foo')}

But I got this:

# first print :
id=None revision_id=None field={'Bar': Child(child_field='Foo')}
# second print:
id=ObjectId('636b9d2997bb72433b944ef4') revision_id=None field={'Bar': {'child_field': 'Foo'}}

field={'Bar': {'child_field': 'Foo'}} != field={'Bar': Child(child_field='Foo')}

It's okay when I wanna fetch from db and my field filled by Child model but when I want to save to db, this situation happened!!!

When used with FastAPI, all parent required fields become child required fields when creating a new document. For example:

Models

from beanie import Document, Indexed, Link


class Organization(Document):
    slug: Indexed(str, unique=True)

    class Settings:
        name = "github-organization"
        use_revision = True


class Repository(Document):
    organization: Link[Organization]
    name: Indexed(str, unique=True)

    class Settings:
        name = "github-repository"
        use_revision = True

Routers

from fastapi import APIRouter

from .models import Organization, Repository

router = APIRouter()


@router.post(
    "/organization",
    response_description="Add a new GitHub Organization",
    response_model=Organization,
)
async def create_organization(organization: Organization):
    await organization.create()
    return organization


@router.post(
    "/repository",
    response_description="Add a new GitHub Repository",
    response_model=Repository,
)
async def create_repository(repository: Repository):
    await repository.create()
    return repository

I can create an Organization by POSTing the following:

{
  "slug": "roman-right"
}

Let's say that generated a Mongo object ID of 62de9ecf2fa3d30007e3b5ce.

However, I cannot create a Repository with just the following:

{
  "organization": {
    "id": "62de9ecf2fa3d30007e3b5ce"
  },
  "name": "beanie"
}

That results in 422, unprocessable entity, with details that the Organization SLUG is required. I have tried the a few variations - id, _id, and also specifying organization as a string (the ID) vs an object (dictionary) - all with the same 422 error and similar messages.

Specifying any string for slug seems to work. For example:

{
  "organization": {
    "id": "62de9ecf2fa3d30007e3b5ce",
    "slug": ""
  },
  "name": "beanie"
}

(Repository document created as expected with link to correct Organization document)

It would seem to just be a parameter validating issue (required parent fields are required parameters even though only the id is used).

There are ORM functionality which I don't see in beanie which are mentioned below,

1. How to add default fields in the model which is updated from whenever records are added. eg. created and updated fields.
2. I want certain fields to be added while insertion and certain fields only when record is updated. eg. created must be added value only at insertion of the records but updated field must be always updated whenever records is updated.

Is the above feature already present, if yes can you point me to the documentation.

Hi @roman-right! Our team is trying to get Beanie to fetch deep-nested Links when fetch_links=True. To accomplish this, I added additional queries to the MongoDB aggregation. However, I saw your comments in the open issues related to nested Links, and I'm not sure this solution gets around the challenges you had mentioned. Would super appreciate your feedback - thank you! :-)

Describe the bug I noticed, that since I updated to beanie 1.14.0, my program does not work with AWS DocumentDB anymore. This has not been a problem prior, and the same code works perfectly with 1.13.1

Additionally, the code works perfectly fine with 1.14.0 against the local MongoDB test database in version 5.0.10.

The error message is not very helpful, the requested resources can simply be not found (although they are there)

NotFound '<some_OID>' for '<class 'mongodb.model.user.odm.User'>' 
not found in database 'User' with id '<some_OID>' not found in database

To verify, that the resource is there I use a tool like NoSQLBooster or Robo3T

db.user.find( {"_id" : ObjectId("<some_OID>")}  )
   .projection({})
   .sort({_id:-1})
   .limit(100)

To Reproduce

# Nothing special, just a simple find command
result = await model.find_one(model.id == oid)

Expected behavior I expected beanie 1.14.0 to work with AWS DocumentDB the same way as 1.13.1

Additional context I am glad to provide further information, or I can make some tests against DocumentDB if someone can give me hints what to do.

So I'm trying to save an entry that has a datetime.date field. I keep getting two errors:

ValueError: [
  TypeError("'datetime.date' object is not iterable"), 
  TypeError('vars() argument must have __dict__ attribute')
]

This is very easy to recreate. Just try to insert this mode:

class TestInsert(beanie.Document):
    name: str
    recorded_date: datetime.date
    items: List[object] = pydantic.Field(default_factory=list)

    class Collection:
        name = "test_remove_after"

With these values, like this:

async def test_date_insert_async():
    ti = TestInsert(
        name="Test",
        recorded_date=datetime.date.today()
    )
    return await ti.insert()

First of all, thank you so much for the great package!!!

Describe the bug beanie fails to properly convert an ElemMatch Query on a Document property of Type Optional[List[Link]] and raises 'beanie\odm\utils\relations.py", line 65, in convert_ids and k.split(".")[1] == "id" IndexError: list index out of range'

To Reproduce Python 3.9

requirements.txt beanie==1.15.4 pydantic==1.9.2 pymongo[srv]==4.1.1

from typing import Optional
from beanie import Document, Link
from beanie.operators import ElemMatch

import asyncio


class DocToLink(Document):

    class Settings:
        name = 'LinkExample'

    child_name: str


class DocWithLinkAttribute(Document):

    class Settings:
        name = 'ParentExample'

    parent_name: str
    linked_docs: Optional[list[Link[DocToLink]]]


async def add_and_query(child_name: str, parent_name: str) -> Optional[list[DocWithLinkAttribute]]:
    """"""

    child: DocToLink = DocToLink(child_name=child_name)
    await child.insert()

    parent: DocWithLinkAttribute = DocWithLinkAttribute(parent_name=parent_name, linked_docs=[child])
    await parent.insert()

    queried_doc:  list[DocWithLinkAttribute] = await DocWithLinkAttribute.find(
            ElemMatch(DocWithLinkAttribute.linked_docs, DocToLink.child_name == child_name), fetch_links=True
            ).to_list()
    return queried_doc


async def init_mongo_client(models: list[Document]):
    """"""
    mongoClient = mdb(secretName=secretMONGO)
    await init_beanie(database=mongoClient.client[mongoClient.databaseInfo["database"]],
                      document_models=models)
    mongoClient.client.get_io_loop = asyncio.get_running_loop


models: list = [DocWithLinkAttribute, DocToLink]
asyncio.run(init_mongo_client(models))

docs = asyncio.run(add_and_query("I'm a child", "I'm the parent"))

Traceback (most recent call last): File "", line 55, in File "Python\Python39\lib\asyncio\runners.py", line 44, in run return loop.run_until_complete(main) File "Python\Python39\lib\asyncio\base_events.py", line 642, in run_until_complete return future.result() File "", line 35, in add_and_query File "\venv\lib\site-packages\beanie\odm\queries\cursor.py", line 71, in to_list cursor = self.motor_cursor File "\venv\lib\site-packages\beanie\odm\queries\find.py", line 609, in motor_cursor aggregation_pipeline.append({"$match": self.get_filter_query()}) File "\venv\lib\site-packages\beanie\odm\queries\find.py", line 105, in get_filter_query self.prepare_find_expressions() File "\venv\lib\site-packages\beanie\odm\queries\find.py", line 93, in prepare_find_expressions self.find_expressions[i] = convert_ids( File "\venv\lib\site-packages\beanie\odm\utils\relations.py", line 65, in convert_ids and k.split(".")[1] == "id" IndexError: list index out of range

Expected behavior The query (debugger shows: {'linked_docs': {'$elemMatch': {'child_name': "I'm a child"}}}) should return the correct (list) of documents. In, fact if ' and k.split(".")[1] == "id"' is commented out and new_k is set to k as in line 72 of relations.py, the output is given correctly. E.g.

  for k, v in query.items():
      if (
          isinstance(k, ExpressionField)
          and doc.get_link_fields() is not None
          and k.split(".")[0] in doc.get_link_fields().keys()  # type: ignore
          # and k.split(".")[1] == "id"
      ):
          if fetch_links:
              new_k = k
              # new_k = f"{k.split('.')[0]}._id"
          else:
              new_k = f"{k.split('.')[0]}.$id"
      else:
          new_k = k

Unfortunately, however the error prevents this from happening. I am not sure where the "id" should come from or how to fix it properly as of now.

I have seen the "# TODO add all the cases" in "convert_ids()" but I was not aware whether this came to your mind already.

Additional context Thank you very much! I really appreciate the package. Let me know if I can help you on this:) Best, Thilo

Describe the bug In 1.11.9 state was changed and always defaults being not None, instead it got a dict with the default values. This causes check_if_state_saved to never throw an error.

save_changes on a new created document that isn't saved in the database silently does nothing.

To Reproduce

user_data = UserEntry(...)
user_data.x = ...
await user_data.save_changes()

Expected behavior StateNotSaved("No state was saved") Additional context https://github.com/roman-right/beanie/compare/1.11.8...1.11.9 https://canary.discord.com/channels/822196934973456394/822196935435747332/1042243293662158970

Describe the bug when using multiple ExpressionField only the first sorting argument reach mongo

To Reproduce

import asyncio

import structlog
from beanie import Document, init_beanie
from beanie.odm.enums import SortDirection
from motor.motor_asyncio import AsyncIOMotorClient
from pymongo import monitoring
from pymongo.monitoring import CommandStartedEvent, CommandSucceededEvent, CommandFailedEvent

logger = structlog.get_logger("beanie test")


class CommandLogger(monitoring.CommandListener):

    def started(self, event: CommandStartedEvent):
        if event.command_name == "find":
            logger.debug(
                f"mongo {event.command_name} started",
                command=event.command,
            )

    def succeeded(self, event: CommandSucceededEvent):
        pass

    def failed(self, event: CommandFailedEvent):
        pass


class MyModel(Document):
    int_number: int
    is_boolean: bool

    class Settings:
        name = "my_model"


async def main():
    await init_beanie(AsyncIOMotorClient(event_listeners=[CommandLogger()])["test_hs_das"], document_models=[MyModel])
    sort1 = [+MyModel.int_number, +MyModel.is_boolean]
    sort2 = [("int_number", SortDirection.ASCENDING), "-is_boolean"]
    logger.info(sort1)
    logger.info(sort2)
    await MyModel.find().sort(sort1).to_list()
    await MyModel.find().sort(sort2).to_list()


if __name__ == '__main__':
    asyncio.run(main())

Expected behavior I expect that when using

await MyModel.find().sort( [+MyModel.int_number,+MyModel.is_boolean]).to_list()

the logger output will be

2023-01-02 16:19:22 [debug ] mongo find started command=SON([('find', 'my_model'), ('filter', {}), ('sort', SON([('int_number', <SortDirection.ASCENDING: 1>), ('is_boolean', <SortDirection.DESCENDING: -1>)])), ...

but is

2023-01-02 16:19:22 [debug ] mongo find started command=SON([('find', 'my_model'), ('filter', {}), ('sort', SON([('int_number', <SortDirection.ASCENDING: 1>)]))

it does work when NOT using ExpressionField

await MyModel.find().sort([("int_number", SortDirection.ASCENDING), "-is_boolean"]).to_list()

Beanie is all sound and good until there is a need to connect to existing DB where beanie is/was not used. For instance, one might have SNS -> SQS -> lambda (that uses any other client other than beanie) -> DocDB / Mongo. In case like this, the existing data would not have beanie's internal data field such as '_class_id".

This makes switching to Beanie difficult as it requires full DB migration, especially if DB is used for event-sourcing.

This PR allows (1) custom internal data field name to be set and (2) custom name to be set in Union doc children classes.

Usage:

class Parent(UnionDoc):
    class Settings:
        class_id = "event_type" <-
        name = "eventsource"

class ChildCreated(Document):
    event_type: Literal["created"] = "created" <- event_type is used instead of _class_id 
    sent_at: datetime

    class Settings:
        name = "created" <- This was option was ignored before
        union_doc = Parent

class ChildDeleted(Document):
    event_type: Literal["created"] = "deleted" <- event_type is used instead of _class_id 
    sent_at: datetime

    class Settings:
        name = "deleted" <- This was option was ignored before
        union_doc = Parent

Example:

Schema

class Parent(UnionDoc):
    class Settings:
        class_id = "event_type"
        name = "collection"


class One(Document):
    test: int

    class Settings:
        name = "two"
        class_id = "event_type"
        union_doc = Parent
        
 class Two(Document): <- without union_doc
    test: int

    class Settings:
        name = "three"
        class_id = "event_type"

Test

In [1]: await db.One(test=1).insert()
Out[1]: One(id=ObjectId('63ae1890edab780d45df0b07'), revision_id=None, test=1)

pymongo
In [17]: list(c["db"]["collection"].find({'test': 1}))
Out[17]: [{'_id': ObjectId('63ae1890edab780d45df0b07'), 'event_type': 'two', 'test': 1}]

In [1]: await db.Two(test=3).insert()
Out[1]: Two(id=ObjectId('63ae19cd8e684a4cd769fb1f'), revision_id=None, test=3)

pymongo
In [23]: list(c["db"]["three"].find({'test': 3}))
Out[23]: [{'_id': ObjectId('63ae19cd8e684a4cd769fb1f'), 'test': 3}]

FYI, https://github.com/roman-right/beanie/pull/206 this PR was not reviewed for a year so I closed it..

Describe the bug

Starlite raises an HTTP 500 error when trying to return a Beanie Document. It seems to be due to the PydanticObjectId type not being JSON serializable. The issue was discussed here on the Starlite repo. Is this an issue that can be fixed within Beanie, or should it be addressed within Starlite?

Thank you for the great work done in Beanie, it simplified my life significantly.

Describe the bug

This is a DB Client I wrote:

class MongoClient:
    def __init__(self):
        client_options = {'appname': self.__class__.__name__}
        self.client = AsyncIOMotorClient(MONGO_URL, **client_options)
        self.client.get_io_loop = asyncio.get_running_loop

        self.is_testing = TESTING == Environments.Testing.value

    @property
    def db_name(self) -> str:
        db_postfix = ENV

        return f'{svc}-{db_postfix}'

    @abstractmethod
    async def initialize(self):
        raise NotImplementedError('MongoClient.initialize() must be implemented')

    @asynccontextmanager
    async def transaction(self) -> AsyncIOMotorClientSession:
        async with await self.client.start_session() as session:
            async with session.start_transaction():
                yield session

The initialize() method is implemented like below:

class MongoClientNumberOne(MongoClient):
    async def initialize(self):
        collections = [
            Service,
            ...
        ]
        await init_beanie(database=self.client[self.db_name], document_models=collections)

The way I use the Mongo client is to declare it as a dependency in FastAPI routes like below:

class MongoClientDependency:
    def __init__(self, db_type: Type[T]):
        if db_type == MongoClientNumberOne:
            self.client = MongoClientNumberOne()
        elif db_type == MongoClientNumberTwo:
            self.client = MongoClientNumberTwo()
        else:
            raise ValueError("Invalid DB Type given")

    async def __call__(self, request: Request) -> MongoClientNumberOne | MongoClientNumberTwo:
        await self.client.initialize()
        return self.client

class Context:
    def __init__(
        self,
        access_token: str | None = None,
        permissions: List[Permissions] | None = None,
        mongo_client: MongoClientNumberOne | None = None,
        mongo_session: AsyncIOMotorClientSession | None = None,
    ):
        self.access_token: str = access_token
        self.permissions: List[Permissions] = permissions

        self.mongo_client: AsyncIOMotorClient = mongo_client
        self.mongo_session: AsyncIOMotorClientSession = mongo_session
        if self.mongo_client:
            self.db_name: str = self.mongo_client.db_name

        self.current_user: User | None = None
        self.current_service: Service | None = None

    @classmethod
    @asynccontextmanager
    async def plain(
        cls, mongo_session: AsyncIOMotorClientSession | None = None, mongo_client: MongoClientNumberOne | None = None
    ) -> Context:
        yield cls(mongo_session=mongo_session, mongo_client=mongo_client)

@router.get(
    '/blog/posts',
    summary='Get latest blog posts',
    tags=['Blog'],
    response_model=List[BlogPost],
    responses={503: {'model': ErrorResponse}},
)
async def get_latest_blog_posts(mongo_client: MongoClientNumberOne = Depends(MongoClientDependency(MongoClientNumberOne))):
    async with mongo_client.transaction():
        async with Context.plain():
            return await Blog.get_latest_posts()

The problem that I see in Sentry is that at times, read/write queries encountered the motor_collection in ItemSettings is None. A screenshot for this below:

Unfortunately this happens randomly and intermittent. Usually when there's no traffic and new traffic comes, these errors popped up.

To Reproduce

At the start, we almost succeeded in replicating this by creating this test, not so many requests but very high concurrency. Before the codes I wrote above, we have an async context manager that was buggy which is now fixed and the test below passed.

@pytest.mark.asyncio
async def test_concurrency(client: TestClient, individual_user: (RegisterRequest, str, str)):
    token = individual_user[2]

    async def get_profile():
        headers = {'Authorization': f'Bearer {token}'}
        response = await asyncify(client.get)(headers=headers, url='/me')
        assert response.status_code == 200

    async def get_wishlist():
        headers = {'Authorization': f'Bearer {token}'}
        response = await asyncify(client.get)(headers=headers, url='/wishlist')
        assert response.status_code == 200

    async def get_txn():
        headers = {'Authorization': f'Bearer {token}'}
        url = '/transactions?limit=10&skip=0&statuses=submitted,accepted,waitlisted,pending-signature,pending-funding,funds-received,in-execution,completed,unsuccessful'
        response = await asyncify(client.get)(headers=headers, url=url)
        assert response.status_code == 200

    for i in range(10):
        print(f'Iteration {i} - Enter')
        await asyncio.gather(get_profile(), get_wishlist(), get_txn(), get_profile(), get_wishlist(), get_txn())
        print(f'Iteration {i} - Exit')

The bug is still happening though, motor_collection sometimes still returns None.

Expected behavior

self.document_model.get_motor_collection() should return the collection.

Additional context

I'm happy to elaborate more with code examples.

Describe the bug Inserting/replacing documents using insert_many or replace_many not triggers any of event-based actions.

To Reproduce

import asyncio
from typing import cast

from beanie import Document, Insert, Replace, before_event, init_beanie
from motor.motor_asyncio import AsyncIOMotorClient


class Test(Document):
    name: str

    @before_event([Insert, Replace])
    def name_force_upper(self) -> None:
        self.name = self.name.upper()


async def test():
    client = AsyncIOMotorClient("mongodb://root:root@localhost:27017")

    # drop db...
    await client.drop_database("test_db")

    await init_beanie(database=client.test_db, document_models=[Test])

    test_doc = Test(name="lowcase_string")
    await test_doc.insert_many([test_doc])

    test_from_db = cast(Test, await Test.find_one())
    assert test_doc.name.upper() == test_from_db.name, f"{test_doc.name = }, {test_from_db.name = }"


if __name__ == "__main__":
    asyncio.run(test())

Expected behavior test_from_db.name in uppercase, which means, that name_force_upper were called

Additional context This behaviour not described in docs, and there are no issues about it, so i think this is a bug.