Dataclass#
When you’ve multiple values that you want to send across Union entities, you can use a dataclass
.
To begin, import the necessary dependencies:
import os
import tempfile
from dataclasses import dataclass
import pandas as pd
import union
from flytekit.types.structured import StructuredDataset
Build your custom image with ImageSpec:
image_spec = union.ImageSpec(
registry="ghcr.io/flyteorg",
packages=["pandas", "pyarrow"],
)
Python types#
We define a dataclass
with int
, str
and dict
as the data types.
@dataclass
class Datum:
x: int
y: str
z: dict[int, str]
You can send a dataclass
between different tasks written in various languages, and input it through the Union UI as raw JSON.
Note
All variables in a data class should be annotated with their type. Failure to do will result in an error.
Once declared, a dataclass can be returned as an output or accepted as an input.
@union.task(container_image=image_spec)
def stringify(s: int) -> Datum:
"""
A dataclass return will be treated as a single complex JSON return.
"""
return Datum(x=s, y=str(s), z={s: str(s)})
@union.task(container_image=image_spec)
def add(x: Datum, y: Datum) -> Datum:
x.z.update(y.z)
return Datum(x=x.x + y.x, y=x.y + y.y, z=x.z)
Union types#
We also define a data class that accepts StructuredDataset
, FlyteFile
and FlyteDirectory
.
@dataclass
class UnionTypes:
dataframe: StructuredDataset
file: union.FlyteFile
directory: union.FlyteDirectory
@union.task(container_image=image_spec)
def upload_data() -> UnionTypes:
df = pd.DataFrame({"Name": ["Tom", "Joseph"], "Age": [20, 22]})
temp_dir = tempfile.mkdtemp(prefix="union-")
df.to_parquet(temp_dir + "/df.parquet")
file_path = tempfile.NamedTemporaryFile(delete=False)
file_path.write(b"Hello, World!")
fs = UnionTypes(
dataframe=StructuredDataset(dataframe=df),
file=union.FlyteFile(file_path.name),
directory=union.FlyteDirectory(temp_dir),
)
return fs
@union.task(container_image=image_spec)
def download_data(res: UnionTypes):
assert pd.DataFrame({"Name": ["Tom", "Joseph"], "Age": [20, 22]}).equals(res.dataframe.open(pd.DataFrame).all())
f = open(res.file, "r")
assert f.read() == "Hello, World!"
assert os.listdir(res.directory) == ["df.parquet"]
A data class supports the usage of data associated with Python types, data classes, FlyteFile, FlyteDirectory and structured dataset.
We define a workflow that calls the tasks created above.
@union.workflow
def dataclass_wf(x: int, y: int) -> (Datum, FlyteTypes):
o1 = add(x=stringify(s=x), y=stringify(s=y))
o2 = upload_data()
download_data(res=o2)
return o1, o2
You can run the workflow locally as follows:
if __name__ == "__main__":
dataclass_wf(x=10, y=20)
To trigger a task that accepts a dataclass as an input with union run
, you can provide a JSON file as an input:
union run \
https://raw.githubusercontent.com/flyteorg/flytesnacks/69dbe4840031a85d79d9ded25f80397c6834752d/examples/data_types_and_io/data_types_and_io/dataclass.py \
add --x dataclass_input.json --y dataclass_input.json