Hi,

I'm trying to build an authorization/security extension for spark using the
hooks provided in SPARK-18127 (
https://issues.apache.org/jira/browse/SPARK-18127).

The problem I've encountered is that those hooks aren't available for
pyspark, as the exensions are loaded in the getOrCreate method of the
SparkSession (
https://github.com/apache/spark/blob/v2.3.1/sql/core/src/main/scala/org/apache/spark/sql/SparkSession.scala#L937-L953)
and the SparkSession used for pyspark is created through the constructor
using py4j which initializes the extensions, but it never adds configured
values to it.

https://github.com/apache/spark/blob/v2.3.1/python/pyspark/sql/session.py#L216
https://github.com/apache/spark/blob/v2.3.1/sql/core/src/main/scala/org/apache/spark/sql/SparkSession.scala#L87-L89

(I'm pointing to the latest stable release (2.3.1) but this is also the
case in master at the moment.)

I haven't found a jira issue or info in the mailing list regarding this,
but wanted to check if this is something that we could hopefuly push to
newer releases to maintain consistency of the extensions feature across
python/scala.

Do you think this is worth a jira issue? I can provide a very small patch
if that helps.

Thanks,
Maximiliano

Reply via email to