Type Mismatching while Serializing Dataclass with Union

[mao3267]

Tracking issue

Closes flyteorg/flyte#5910

Why are the changes needed?

While converting dataclasses from Python types to literal types through the to_literal function, the dataclass will pass through the _make_dataclass_serializable function to ensure it can be serialized later. For dataclasses with Union properties, under the if UnionTransformer.is_optional_type(python_type): statement, it uses get_args(python_type)[0] as the type of the property without verifying its compatibility. This will cause an error if the first argument is not the expected type.

What changes were proposed in this pull request?

1. Do typing matching instead of using the first type from get_args(python_type)
2. Add unit tests for more possible union types

How was this patch tested?

1. Run on local and remote execution

• Union[None, FlyteFile]
  flytefile_serialize.py

from dataclasses import dataclass
from flytekit import task
from flytekit.types.file import FlyteFile
from typing import Union
from flytekit.image_spec import ImageSpec

flytekit_hash = "f647bd3a1b727082210454c3f5d1e652a29b1083"
flytekit = f"git+https://github.com/mao3267/flytekit.git@{flytekit_hash}"

image_spec = ImageSpec(
    name="serialize-union-1",
    packages=[flytekit],
    builder="default",
    registry="localhost:30000",
    apt_packages=["git", "gh"],
)

@dataclass
class InnerDC:
    ff: Union[None, FlyteFile]

@dataclass
class DC:
    inner_dc: InnerDC

@task(container_image=image_spec)
def t_dc() -> DC:
    return DC(inner_dc=InnerDC(ff="s3://path"))

• Union[None, int, str]
  serialize_union.py

from dataclasses import dataclass
from flytekit import task
from typing import Union
from flytekit.image_spec import ImageSpec

flytekit_hash = "f647bd3a1b727082210454c3f5d1e652a29b1083"
flytekit = f"git+https://github.com/mao3267/flytekit.git@{flytekit_hash}"

image_spec = ImageSpec(
    name="serialize-union-1",
    packages=[flytekit],
    builder="default",
    registry="localhost:30000",
    apt_packages=["git", "gh"],
)

@dataclass
class InnerDC:
    ff: Union[None, int, str]

@dataclass
class DC:
    inner_dc: InnerDC

@task(container_image=image_spec)
def t_dc() -> DC:
    return DC(inner_dc=InnerDC(ff="string"))

2. Add unit tests in test_type_engine.py

Setup process

git clone https://github.com/flyteorg/flytekit.git
gh pr checkout 2859
pip install -e .

Screenshots

• Local
  • Union[None, FlyteFile]
    
  • Union[None, int, str]
    
• Remote

  • Union[None, int, str]
    

  • Union[None, int, str]
    

Check all the applicable boxes

• I updated the documentation accordingly.
• All new and existing tests passed.
• All commits are signed-off.

Related PRs

None

Docs link

None

[wild-endeavor]

Can you make this

Suggested change

	c_prime: typing.Union[None, FlyteFile]
	c_prime: typing.Union[None, StructuredDataset, int, FlyteFile]

[wild-endeavor]

actually can you write one more unit test for me please? (and add it under test_dataclass.py this file is getting too big).

@dataclass
class A():
  x: int

@dataclass
class B():
   x: FlyteFile

then call _make_dataclass_serializable on Union[None, A, B] where b = B(x="s3://tmp) or something.

[codecov]

Codecov Report

Attention: Patch coverage is 81.81818% with 2 lines in your changes missing coverage. Please review.

Project coverage is 79.15%. Comparing base (3fc51af) to head (4894ae5).
Report is 8 commits behind head on master.

Files with missing lines	Patch %	Lines
flytekit/core/type_engine.py	81.81%	0 Missing and 2 partials ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           master    #2859       +/-   ##
===========================================
+ Coverage   45.53%   79.15%   +33.61%     
===========================================
  Files         196      196               
  Lines       20418    20545      +127     
  Branches     2647     2647               
===========================================
+ Hits         9298    16262     +6964     
+ Misses      10658     3546     -7112     
- Partials      462      737      +275     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[wild-endeavor]

can we make the error

Cannot have more than one Flyte type in the Union when attempting to use the string shortcut. Please specify the full object (e.g. FlyteFile(...)) instead of just passing a string.

instead?

[wild-endeavor]

thank you! left one more comment to update the error message. otherwise good. thank you.

[mao3267]

Although we didn’t expect typing.Union[None, StructuredDataset, int, FlyteFile] to work in our tests due to multiple FlyteTypes ambiguity, we anticipated something like typing.Union[None, int, FlyteFile] might function correctly. However, it seems that while messagepack decoder decodes from a binary to a python value, it fails to identify the target type for Union types with more than two types, resulting a dictionary rather than a FlyteFile. cc @Future-Outlier

[wild-endeavor]

@Future-Outlier - do you know what the problems are?

[mao3267]

I have raised an issue at Fatal1ty/mashumaro#255 describing the bugs we faced. Hope that it will be fixed in a couple of weeks.

[Future-Outlier]

I have raised an issue at Fatal1ty/mashumaro#255 describing the bugs we faced. Hope that it will be fixed in a couple of weeks.

Can you also create an issue in flyte and link this PR and mashumaro's issue?