#3720 [Core feature] Show FlyteDeck even when task fails Issue created by peridotml Motivation: Why do you think this is important? Users want to surface failing validation checks on data and models in Decks. For example, • Show which checks failed in pandera • A model did not pass quality checks, i.e. the model returned random results This also allows users to use Decks early in tasks that could be used for debugging. I can see someone saving summary statistics on a dataset to the Deck before training. Then when training fails, they could inspect the deck and be like "Oh lol I had 10000 columns instead of 5" Goal: What should the final outcome look like, ideally? A user will go through this flow 1. See an error was raised that says their validation check failed 2. Click on FlyteDecks 3. Look at the reasons it failed For plugins, like Pandera, the FlyteDeck cloud be built into the plugin! Describe alternatives you've considered Have the validation task succeed and then return a boolean to see if it failed. Then raise in a separate task.

@task(container_image=python_image, disable_deck=False)
def validate_dataset(df: pd.DataFrame) -> bool:
    try:
        InputSchema.validate(df, lazy=True)
    except SchemaErrors as err:
        Deck(
            "Schema Validation Failures",
            build_table(err.failure_cases, "orange_light"),
        )
        return True

    return False

@workflow
def data_validaition_wf(valid: bool):
    df = load_dataset()
    failure = validate_dataset(df=df)
    raise_error(failure=failure) >> etl_task(df=df)

I don't like the extra steps! Propose: Link/Inline OR Additional context slack convo: https://flyte-org.slack.com/archives/CP2HDHKE1/p1684947483699369?thread_ts=1684945136.431699&cid=CP2HDHKE1 Are you sure this issue hasn't been raised already? ☑︎ Yes Have you read the Code of Conduct? ☑︎ Yes flyteorg/flyte