@Marvin I have a pydantic class that uses a generic and serialization fails:

Task run 'google_download_file-c9c' - Encountered an error while serializing result for transaction 'ccda71d380b8bf741ee2e6fa41c9005a': Failed to serialize object of type 'APIResult[TypeVar]' with serializer 'json'

Is there a way to handle this?

✅ thought for 243.4 seconds

can you show your type definition?

Short answer: yes. This error happens because the JSON result serializer doesn’t play nicely with Pydantic generics that aren’t fully concrete/importable. You’ve got a few good options depending on whether you really need JSON. What’s happening - With Prefect 3.x, the JSON serializer tries to encode the class and then decode it later. For Pydantic generics, if the instance’s class looks like

APIResult[TypeVar]

(or even

APIResult[User]

as a dynamic generic alias), JSON encoding/decoding can fail. - Pickle handles these cases fine because it doesn’t need to reconstruct the class from an importable name. Ways to fix it 1) Switch to pickle (easiest) - This is the default in Prefect 3.x, so if you set

result_serializer="json"

somewhere, just remove it or set it to pickle explicitly.

from prefect import task

@task(result_serializer="pickle")  # or omit; pickle is default in Prefect 3.x
def my_task():
    ...

- For large results, consider:

@task(result_serializer="compressed/pickle")

2) Keep JSON, return a JSON-able dict - Dump the model to a dict so the serializer stores plain JSON data.

from prefect import task

@task(result_serializer="json")
def my_task():
    result = APIResult<data=User(name="alice", age=30|User>)
    return result.model_dump(mode="json")  # or result.model_dump()

- If you prefer a JSON string:

return result.model_dump_json()

3) Keep JSON, but use a named concrete subclass (importable) - Avoid returning a dynamic generic like

APIResult[User]

and instead define a concrete, top-level subclass so Prefect can re-import it on load.

from pydantic import BaseModel
from typing import Generic, TypeVar

T = TypeVar("T")

class APIResult(BaseModel, Generic[T]):
    data: T
    status: str = "success"

class User(BaseModel):
    name: str
    age: int

# Define at module top-level so it's importable
class UserResult(APIResult[User]):
    pass

from prefect import task

@task(result_serializer="json")
def my_task() -> UserResult:
    return UserResult(data=User(name="alice", age=30))

4) Don’t persist results - If you don’t need result persistence, you can disable it:

@task(persist_result=False)
def my_task():
    return APIResult<...|User>

Helpful docs - Results and serializers in Prefect 3.x: Prefect results and serializers A couple of clarifying questions so I can suggest the best fit: - Do you need JSON-persisted results for downstream systems, or is pickle acceptable? - Where is

result_serializer="json"

set (on the task/flow, or via

PREFECT_RESULTS_DEFAULT_SERIALIZER

)? - Are your models Pydantic v2 (Prefect 3.x assumes v2)?

@Marvin if the serialization fails will this impact caching?

✅ thought for 140.4 seconds