beam-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From María GH (JIRA) <j...@apache.org>
Subject [jira] [Created] (BEAM-802) Support Dynamic PipelineOptions for python
Date Mon, 24 Oct 2016 04:15:58 GMT
María GH created BEAM-802:
-----------------------------

             Summary: Support Dynamic PipelineOptions for python
                 Key: BEAM-802
                 URL: https://issues.apache.org/jira/browse/BEAM-802
             Project: Beam
          Issue Type: New Feature
          Components: beam-model
            Reporter: María GH
            Priority: Minor


Goal:  Enable users to run pipelines from templates filled via CL (pipeline options)
Background: Currently, the Runner creates the JSON pipeline description which can be sent
to the worker as is, since everything is already defined there (with links to gs:// for input
and binaries). With the parametrized approach, those descriptions are empty and filled by
the user or defaulted, so the pipeline needs to be stored somewhere first until the values
become available.
Tasks:
1- Create template-style pipeline description (TemplateRunner)
The graph description is now a template (some parts are not filled) that needs to be saved
in GCS. 
2- Define values to inject to the template (ValueProviders API)
The placeholders can be filled with default values (static) or with dynamic key/value pairs
provided at runtime (dynamic)
3- Adapt service
	1- Allow various IO classes to use accept placeholders and defer validation for dynamic values:

For Text: file.open(<some value>)
For BigQ: table.select(<some value>)
        2- Have a mechanism for the service to instantiate the template with the actual values,
so that the worker can start working on the pipeline.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message