Stephen Reichling created BEAM-1068:
---------------------------------------
Summary: Service Account Credentials File Specified via Pipeline
Option Ignored
Key: BEAM-1068
URL: https://issues.apache.org/jira/browse/BEAM-1068
Project: Beam
Issue Type: Bug
Components: sdk-py
Reporter: Stephen Reichling
Assignee: Frances Perry
Priority: Minor
When writing a pipeline that authenticates with Google Dataflow APIs using a
service account, specifying the path to that service account's credentials file
in the {{PipelineOptions}} object passed in to the pipeline does not work, it
only works when passed as a command-line flag.
For example, if I write code like so:
{code}
pipelineOptions = options.PipelineOptions()
gcOptions = pipelineOptions.view_as(options.GoogleCloudOptions)
gcOptions.service_account_name = 'My Service Account Name'
gcOptions.service_account_key_file = '/some/path/keyfile.p12'
pipeline = beam.Pipeline(options=pipelineOptions)
# ... add stages to the pipeline
p.run()
{code}
and execute it like so:
{{python ./my_pipeline.py}}
...the service account I specify will not be used.
Only if I were to execute the code like so:
{{python ./my_pipeline.py --service_account_name 'My Service Account Name'
--service_account_key_file /some/path/keyfile.p12}}
...does it actually use the service account.
The problem appears to be rooted in `auth.py` which reconstructs the
{{PipelineOptions}} object directly from {{sys.argv}} rather than using the
instance passed in to the pipeline:
https://github.com/apache/incubator-beam/blob/9ded359daefc6040d61a1f33c77563474fcb09b6/sdks/python/apache_beam/internal/auth.py#L129-L130
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
