PrefectResult

stores something directly in the database column which has to be a string (ie JSON) not bytes.

I think this is because we don’t want to store pickles in the database

It’s dangerous since unpickling can execute arbitary code

Following the b64 encoding approach, with how

PrefectResult

leverages

location

and

value

, the Location Result in the UI has a b64 encoded string. I’m curious if I could just reassign

location

to a string representation of my value.

Here is some sample code:

class CloudPicklePrefectResult(Result):
    def __init__(self, **kwargs: Any) -> None:
        super().__init__(**kwargs)

    def _encode(self, value: str) -> bytes:
        return value.encode('utf-8')
    
    def _decode(self, value: bytes) -> str:
        return value.decode('utf-8')

    def read(self, location: str) -> Result:
        new = self.copy()
        new.value = (
            cloudpickle.loads(
                base64.b64decode(
                    self._encode(location)
                )
            )
        )

        try:
            new.location = str(new.value)
        except:
            new.location = location

        return new

    def write(self, value_: Any, **kwargs: Any) -> Result:
        new = self.copy()
        new.value = value_
        new.location = (
            self._decode(
                base64.b64encode(
                    cloudpickle.dumps(new.value)
                )
            )
        )
        return new

Ignore my last statement. I was thinking of results that read from databases won’t work. This will work perfectly fine. I see the issue you are saying now with the Result Location. Will dig in.

Ok so I talked to Michael and this is easier to do (without displaying the location as a string) if you subclass the JSONSerializer and create one that uses cloudpickle. Example:

class PickleJSONSerializer(JSONSerializer):
    def serialize(self, value: Any) -> bytes:
        return base64.b64encode(cloudpickle.dumps(value))

    def deserialize(self, value: bytes) -> Any:
        return cloudpickle.loads(base64.b64decode(value))

@task
def abc():
    return "this string"

with Flow("result_test", result = PrefectResult(serializer = PickleJSONSerializer())) as flow:
    abc()

But if you really want the result location to be a string in the dashboard, I am not 100% sure but it seemed to render right with this implementation (no serializer attached):

class CloudPicklePrefectResult(Result):
    def __init__(self, **kwargs: Any) -> None:
        super().__init__(**kwargs)

    def read(self, location: str) -> Result:
        new = self.copy()
        new.value = (
            cloudpickle.loads(
                base64.b64decode((location.encode("utf-8"))
                )
            )
        )
        new.location = location

        return new

    def write(self, value_: Any, **kwargs: Any) -> Result:
        new = self.copy()
        new.value = base64.b64encode(cloudpickle.dumps(value_)).decode("utf-8")
        new.location = value_
        return new