Skip to content

Vertex AI Pipeline - PipelineJob run is still logging in background when sync is set to False #688

Description

@SaschaHeyer

Hello Google Team,

when running a PipelineJob in async mode it still logs the pipeline job status in the background.
This is particularly bad if we just want to use the Client Library to start a pipeline run. I would expect that no action is happening in the background when setting the sync mode to False

pl = PipelineJob(
 ...
)

# Execute pipeline in Vertex
pl.run(
  # Whether this function call should be synchronous (wait for pipeline run to finish before terminating)
  # or asynchronous (return immediately)
  sync=False
)

Best regards
Sascha

Activity

  1. changed the title [-]PipelineJob run is still logging in background when sync is set to False[/-] [+]Vertex AI Pipeline - PipelineJob run is still logging in background when sync is set to False[/+] on Sep 3, 2021
  2. SaschaHeyer commented on Sep 28, 2021

    @SaschaHeyer
    ContributorAuthor

    Morning
    any feedback on that?

  3. SaschaHeyer commented on Oct 5, 2021

    @SaschaHeyer
    ContributorAuthor

    Morning
    any feedback on that?

  4. sasha-gitg commented on Oct 6, 2021

    @sasha-gitg
    Member

    Hi Sascha,

    We plan to add a create API to handle the case where the intention is to create a PipelineJob but not monitor it's completion.

    Thanks,
    Sasha

  5. SaschaHeyer commented on Oct 12, 2021

    @SaschaHeyer
    ContributorAuthor

    Hi @sasha-gitg,

    looking very much forward to this as part of the API.

    Best regards
    Sascha

  6. added
    type: feature request‘Nice-to-have’ improvement, new feature or different behavior or design.
    on Oct 18, 2021
  7. adesso-alex commented on Oct 25, 2021

    @adesso-alex

    Hi everyone,

    I just ran into the same bug with the "sync=False" flag not having the intended effect here, and would very much support any solution/possibility to trigger an async pipeline job run.

    Best Regards,
    Alex

  8. ji-yaqi commented on Oct 25, 2021

    @ji-yaqi
    Contributor

    Hi @adesso-alex, by 'not having the intended effect', do you mean the job is not async?

    If you do

    pl = PipelineJob(
     ...
    )
    
    # Execute pipeline in Vertex
    pl.run(
      sync=False
    )
    print('next action')
    

    'next action' was printed immediately after async call. I understand you might get confused over some pipelineJob comments being sent out after async, but that's just the side effect. In fact, the next line (if you have any) is run immediately after.

  9. adesso-alex commented on Oct 25, 2021

    @adesso-alex

    Hi @ji-yaqi,

    you are right, that's the way it works right now. I see that this is the expected behavior.
    It would be cool if, as requested, there would be a way to avoid a background thread just being created to do the logging/monitoring here.

    Thanks!

  10. ji-yaqi commented on Oct 25, 2021

    @ji-yaqi
    Contributor

    @adesso-alex we are working on a solution right now :) Thanks for you patience

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Labels

api: aiplatformIssues related to the AI Platform API.type: feature request‘Nice-to-have’ improvement, new feature or different behavior or design.

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions