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!

User Feedback on ContainerError Implementation

Summary

The user agrees with Rafael's suggestion to use ContainerError in Flytekit's agent service and is open to implementing it. They note that while it isn't technically a ContainerError since it doesn't involve a container, porting a task from pod to agent would result in a breaking change due to differences in error reporting. The user mentions the need to evaluate fields under ContainerError to determine if a retry is necessary and expresses a desire for improved error reporting mechanisms.

Status
resolved
Tags
  • User Feedback
  • flyte
  • Error Reporting
  • Developer
  • Rafael
  • Developer Help
Source
#flyte-agents
    r

    rafaelraposo

    10/25/2024

    Ok so if I understood correctly. https://github.com/flyteorg/flyte/pull/5916 Here's a good way, still not totally convinced we would need all those fields but is customisable now.

    r

    rafaelraposo

    10/24/2024

    The breaking change is how it's reporting the errors back. As it stands right now I would have to include it in the message or in the Custom Field that I forget the name.

    Ideally if we had this https://github.com/flyteorg/flyte/blob/master/flyteidl/protos/flyteidl/core/execution.proto#L61 that would be sweet

    r

    rafaelraposo

    10/24/2024

    Happy to try and implement it. :slightly_smiling_face:

    Even though it's not a ContainerError since it's a not a container, but the usecase here is that porting a task from pod to agent, as it is right now it would be a breaking change.

    We evaluate the fields under the ContainerError to evaluate if we need to retry or not.

    e

    eric901201

    10/24/2024

    Rafael and I just discussed the idea of using ContainerError in Flytekit's agent service, and I completely agree with the suggestion. I'd love to hear what others think about it.

    For reference:

    • ContainerError: Link
    • Agent service: Link