Florian FERREIRA created AIRFLOW-6086:
-----------------------------------------
Summary: SparkSubmitOperator - Unable to override spark_binary
Key: AIRFLOW-6086
URL: https://issues.apache.org/jira/browse/AIRFLOW-6086
Project: Apache Airflow
Issue Type: Bug
Components: contrib, core
Affects Versions: 1.10.6
Reporter: Florian FERREIRA
Hello,
I have a connection "spark2_default" :
|| Conn Id || Conn Type || Host || Port || Is Encrypted || Is Extra Encrypted
|| Extra
| 'spark2_default' | 'spark2' | 'yarn-cluster' | None | False | False |
{"master":"yarn-cluster","deploy-mode":"cluster","spark-binary":"spark2-submit"}
|
Extra contains 'spark-binary' key that was use by airflow 1.10.2 to choose
spark-submit operator. But in version 1.10.6 this config is ignore.
I think that , in class SparkSubmitOperator in init function they has a default
value "spark-submit" for spark_binary parameter.
{code}
spark_binary="spark-submit",
{code}
Therefore in class SparkSubmitHook when we control if spark_binary is empty it
can't be.
{code}
conn_data['spark_binary'] = self._spark_binary or \
extra.get('spark-binary', "spark-submit")
{code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
