[
https://issues.apache.org/jira/browse/AIRFLOW-1401?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Peter Dolan updated AIRFLOW-1401:
---------------------------------
Description:
At the moment, there isn't standard usage of operator arguments for Google
Cloud Platform across the contributions, primarily in the usage of the
parameter meaning the GCP project name/id. This makes it difficult to specify
default_arguments that work across all GCP-centric operators in a graph.
Using the command `grep -r project airflow/contrib/*`, we can see these uses:
project_id:
* gcp_dataproc_hook
* datastore_hook
* gcp_api_base_hook
* bigquery_hook
* dataproc_operator
* bigquery_sensor
project:
* gcp_pubsub_hook (here 'project' means project id or project name, which does
not fully understand the distinction within GCP between project id and project
name as elements of the REST api)
* dataflow_operator (see note below)
* pubsub_operator
project_name:
* gcp_cloudml_hook
* cloudml_operator
Notably, the Dataflow Operator diverges from the pattern of using top-level
operator parameters by specifying an options dict, which can be populated by
the dataflow_default_options dict. This can contain 'project', and 'zone.'
This improvement proposes to standardize the above operators (at least) on
* project_id (meaning '<project>' in this example request: GET
https://www.googleapis.com/compute/v1/projects/<project>/zones/<zone>/instances/<instance>)
* region
* zone
This can be done by changing the names of parameters of operators and hooks
that were not included in the 1.8.2 release (cloud ml and pubsub), and by
adding parameters to operators and hooks that were included in 1.8.2 (and
internally copying the old parameter name to the new one, and deprecating the
old one).
was:
At the moment, there isn't perfectly standard usage of operator arguments for
Google Cloud Platform across the contributions, primarily in the usage of the
parameter meaning the GCP project name/id. This makes it difficult to specify
default_arguments that work across all GCP-centric operators in a graph.
Using the command `grep -r project airflow/contrib/*`, we can see these uses:
project_id:
* gcp_dataproc_hook
* datastore_hook
* gcp_api_base_hook
* bigquery_hook
* dataproc_operator
* bigquery_sensor
project:
* gcp_pubsub_hook (here 'project' means project id or project name, which does
not fully understand the distinction within GCP between project id and project
name as elements of the REST api)
* dataflow_operator (see note below)
* pubsub_operator
project_name:
* gcp_cloudml_hook
* cloudml_operator
Notably, the Dataflow Operator diverges from the pattern of using top-level
operator parameters by specifying an options dict, which can be populated by
the dataflow_default_options dict. This can contain 'project', and 'zone.'
This improvement proposes to standardize the above operators (at least) on
* project_id (meaning '<project>' in this example request: GET
https://www.googleapis.com/compute/v1/projects/<project>/zones/<zone>/instances/<instance>)
* region
* zone
This can be done by changing the names of parameters of operators and hooks
that were not included in the 1.8.2 release, and by adding parameters to
operators and hooks that were included in 1.8.2 (and internally copying the old
parameter name to the new one, and deprecating the old one).
> Standardize GCP project, region, and zone argument names
> --------------------------------------------------------
>
> Key: AIRFLOW-1401
> URL: https://issues.apache.org/jira/browse/AIRFLOW-1401
> Project: Apache Airflow
> Issue Type: Improvement
> Components: contrib
> Affects Versions: 1.8.1
> Reporter: Peter Dolan
> Assignee: Peter Dolan
>
> At the moment, there isn't standard usage of operator arguments for Google
> Cloud Platform across the contributions, primarily in the usage of the
> parameter meaning the GCP project name/id. This makes it difficult to specify
> default_arguments that work across all GCP-centric operators in a graph.
> Using the command `grep -r project airflow/contrib/*`, we can see these uses:
> project_id:
> * gcp_dataproc_hook
> * datastore_hook
> * gcp_api_base_hook
> * bigquery_hook
> * dataproc_operator
> * bigquery_sensor
> project:
> * gcp_pubsub_hook (here 'project' means project id or project name, which
> does not fully understand the distinction within GCP between project id and
> project name as elements of the REST api)
> * dataflow_operator (see note below)
> * pubsub_operator
> project_name:
> * gcp_cloudml_hook
> * cloudml_operator
> Notably, the Dataflow Operator diverges from the pattern of using top-level
> operator parameters by specifying an options dict, which can be populated by
> the dataflow_default_options dict. This can contain 'project', and 'zone.'
> This improvement proposes to standardize the above operators (at least) on
> * project_id (meaning '<project>' in this example request: GET
> https://www.googleapis.com/compute/v1/projects/<project>/zones/<zone>/instances/<instance>)
> * region
> * zone
> This can be done by changing the names of parameters of operators and hooks
> that were not included in the 1.8.2 release (cloud ml and pubsub), and by
> adding parameters to operators and hooks that were included in 1.8.2 (and
> internally copying the old parameter name to the new one, and deprecating the
> old one).
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
