F

Flyte enables you to build & deploy data & ML pipelines, hassle-free. The infinitely scalable and flexible workflow orchestration platform that seamlessly unifies data, ML and analytics stacks. Explore and Join the Flyte Community!

FlyteContainerTask Output Issue

Summary

The user is experiencing an issue with a container task in Flyte, where the output, expected to be a FlyteDirectory (an S3 path), is being interpreted as a single blob. This misinterpretation is causing an error in a subsequent task that uses this output as input, indicating that the specified S3 path is not recognized as a directory.

Status
open
Tags
  • flyte
  • Flyte
  • Container Task
  • S3 path
  • Bug Report
Source
#ask-the-community
    i

    ishasingh

    10/25/2024

    the subsequent task is a regular task not container task. The container task copies the output to flyte backend storage in s3 but it somehow returns a single blob instead of multipart blob

    d

    david.espejo

    10/25/2024

    not sure if it explains the behavior you see

    d

    david.espejo

    10/25/2024

    I think FlyteDirectory is supported as an output of ContainerTask but not as an input (there's <https://github.com/flyteorg/flyte/pull/5715|work in progress>)

    i

    ishasingh

    10/25/2024

    Hi, I have a container task whose output is a FlyteDirectory (an s3 path) and its somehow interpreting it as a single blob. see: {"type":"single blob","uri":"<s3://bucket_name/prefix/results>"} A subsequent task that uses this output as an input is erroring out:

    
        TypeTransformerFailedError: Error converting input 'results_master' at position 0:
      <s3://bucket_name/prefix/results> is not a directory.```