Hyukjin Kwon created SPARK-62129:
------------------------------------
Summary: Raise the exceptions of the reads of a client's
attributes by their calls, and read them strictly for the requests of a
failure, as PySpark does, in the Python client
Key: SPARK-62129
URL: https://issues.apache.org/jira/browse/SPARK-62129
Project: Spark
Issue Type: Bug
Components: Connect
Affects Versions: 5.0.0
Reporter: Hyukjin Kwon
After SPARK-62114, the Python client (pyspark-client-rust) reads the attributes
of a SparkConnectClient again at the end of a nested call, before the requests
of a call's failure, and before the interrupt of a Ctrl-C, as PySpark's
requests read them when they are built. Two cases still differ from
pyspark-client 4.3.0:
- An exception that is not an Exception, raised while the attributes are read
there (a KeyboardInterrupt of a signal that comes then, a SystemExit of a
signal handler), was raised again later as a new SIGINT, or reported as
unraisable, where PySpark raises it from the call whose request it was
building: a SystemExit of the read before the details of a call's failure was
lost, and the call raised the server's error.
- The read before the requests of a call's failure took its values best-effort,
so an attribute that became one that protobuf refuses during the call (an
extension that is not an Any) was ignored, where PySpark's request of the
details raises that TypeError in place of the server's error, and sends neither
the request of the details nor those of the configs of the stack trace.
Expected: as pyspark-client, such an exception is raised by the call whose
request was built, and the read before the failure requests reads the
extensions strictly, of PySpark's error.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]