Description
First Check
- I added a very descriptive title to this issue.I used the GitHub search to find a similar issue and didn't find it.I searched the SQLModel documentation, with the integrated search.I already searched in Google "How to X in SQLModel" and didn't find any information.I already read and followed all the tutorial in the docs and didn't find an answer.I already checked if it is not related to SQLModel but to Pydantic.I already checked if it is not related to SQLModel but to SQLAlchemy.
Commit to Help
- I commit to help with one of those options 👆
Example Code
class Team(SQLModel):
id: int = Field(default=None, primary_key=True)
heroes: List["Hero"] = Relationship(back_populates="team", sa_relationship_kwargs={"lazy": "selectin"})
class Hero(SQLModel):
name: str
team_id: Optional[int] = Field(default=None, foreign_key="team.id")
team: Optional[Team] = Relationship(back_populates="incidents")
app = FastAPI()
@app.get("/teams/", response_model=List[Team])
def get_teams() -> List[Team]:
with Session(engine) as session:
return session.exec(select(Team)).all()
desired_output = [
{
"id": 1,
"heroes": [
{"name": "Batman"},
{"name": "Superman"}
]
}
]
Description
When returning a list of Teams the Relationship
fields are not covered in the output. I believe the response is using pydantic.main.BaseModel.json()
to get the data (and there it will only cover non-Relationship
fields).
It would be great if there was an option to also return the Relationship
fields especially if {"lazy": "selectin"}
was enabled for the relation.
It's an easy workaround to walk the objects and return the complete data as json in the endpoint, but since we already have the models that is not very satisfying (also because the response_model for openapi will not be easily achievable).
Did I miss an option to enable having the Relationship
attributes in the response, or is that something that could be worked on?
I would also be happy to contribute but would need a pointer on where to start.
Operating System
Linux
Operating System Details
No response
SQLModel Version
0.0.7
Python Version
3.10.4
Additional Context
No response
Activity
daniil-berg commentedon Sep 9, 2022
First off, if you haven't already, I suggest reading this section of the docs. It addresses the issue of infinite recursion (among other things) that can easily happen, if you don't think carefully about what data you actually want to get.
Second, I assume you forgot to copy the
table=True
in your example models. Without those, you can't use the models for any database queries.The cyclic dilemma
There are already quite a few issues around this topic of including relationships in such a way. I haven't looked deeply into the code for this yet. But as I understand it, it is not at all clear, how exactly you should tackle the problem of cyclical relationship graphs. Not that there aren't any theoretical solutions. It seems to be more of a question of what a "sensible" approach may look like. In the end, it would have to be up to the user to decide, when exactly to stop walking along the relationship graph.
The most straightforward solution, and the one used right now is to simply avoid the issue altogether and force the user to e.g. set up his own response models (not tables) with the desired nesting of sub-models. (See the section "What Data to Include" on the aforementioned docs page.)
Be explicit yourself
So I would suggest closely following the advice about model inheritance from the docs for now. That means for your desired output:
Under the hood, FastAPI will basically call the
TeamReadWithHeroes.from_orm()
method on eachTeam
instance returned by the database query. This will populate theheroes
field on that model and give you the desired result:busykoala commentedon Sep 9, 2022
Thanks for the elaborate reply!
table=True
is indeed missing sorry for that mistake.Cyclic Dilemma
I've read the documentation regarding the issue, it would be up to the implementation whether that is an issue or not. It would only be one if the table design is accordingly.
Regarding the Suggestion
The suggested code would in my case duplicate every single model with the exact same attributes, in which case it would probably be smarter to work with sqlalchemy and pydantic only.
Thanks for the explanation and hopefully there will be a possibility in the future to directly decide upon whether to include the relations or not.
daniil-berg commentedon Sep 9, 2022
Yes, I know what you mean. I've been thinking about this a lot recently.
Maybe someone will figure out an intuitive design for this. Because I don't think the issue here is that implementation would be hard. It's the question of "how should the interface look"? If you or anyone else has an idea, like "could we maybe have a function like this ...", I am genuinely interested.
Regarding duplication, I can't imagine that it would actually force you to duplicate every attribute in every case. 👀 I don't know your specific use case, but I can't imagine SQLModel not leading to significantly less code (compared to Pydantic+SQLAlchemy). Even with the relationship issue... But you'll know best what works for you.
antont commentedon Sep 9, 2022
Yeah @busykoala how does it duplicate, when with the inheritance you get reuse? I'm using this pattern quite happily.
busykoala commentedon Sep 9, 2022
As a workaround, I played around a bit and the below code was the outcome. The SQLModel class could provide something similar (implemented a bit cleaner) to walk the objects. Also, this way an additional counter variable would make it easy to stop infinite recursion if that was desired.
The below code only walks relations that have lazy selectin enabled like
Relationship(back_populates="team", sa_relationship_kwargs={"lazy": "selectin"})
, if not enabled the relations will be skipped.I've provided an example serializer to show how that could be customized if needed as well.
I'm probably forgetting about many things, but that might be a start anyways.
xavipolo commentedon Sep 11, 2022
Does anyone have the models separated in different files?
When I separate the Hero from the Team in two different files I get an error when open "/docs/"
In one file all is running ok, but not with separated files.
All is running OK with separated files, without the HeroReadWithTeam / TeamReadWithHeroes
Note: I'm using TYPE_CHECKING with imports.
dict()
method for SQLModel objects does not include relationships fractal-analytics-platform/fractal-server#489