[
https://issues.apache.org/jira/browse/SPARK-48517?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Prabhu Joseph updated SPARK-48517:
----------------------------------
Description:
PythonWorkerFactory does not print the error stream in case the daemon fails
before executing the main daemon.py#mainI(). It throws SparkException like
below but does not print the error stream which has the failure why
pyspark.daemon failed to start.
{code:java}
org.apache.spark.SparkException:
2024-05-07T16:04:53.169524256Z stderr F Bad data in pyspark.daemon's standard
output. Invalid port number:
2024-05-07T16:04:53.169530703Z stderr F 1097887852 (0x4170706c){code}
The error stream is being [read|#L303]] after throwing SparkException. It has
to be captured during the exception as well.
*Simple Repro:*
1. Run a sample pyspark job by setting wrong
spark.python.daemon.module like pyspark.wrongdaemon instead of default one
pyspark.daemon.
2. The forked python process will fail with
{*}"{*}{*}/opt/python/3.9.2/bin/python: No module named pyspark.wrongdaemon"{*}
but it is not captured by PythonWorkerFactory.
was:
PythonWorkerFactory does not print the error stream in case the daemon fails
before calling the main daemon.py#method.
It throws SparkException like below but does not print the error stream which
has the failure why pyspark.daemon failed to start.
{code:java}
org.apache.spark.SparkException:
2024-05-07T16:04:53.169524256Z stderr F Bad data in pyspark.daemon's standard
output. Invalid port number:
2024-05-07T16:04:53.169530703Z stderr F 1097887852 (0x4170706c){code}
The error stream is being
[read|[https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala#L303]]
after throwing SparkException. It has to be captured during the exception as
well.
*Simple Repro:*
1. Run a sample pyspark job by setting wrong
spark.python.daemon.module like pyspark.wrongdaemon instead of default one
pyspark.daemon.
2. The forked python process will fail with
{*}"{*}{*}/opt/python/3.9.2/bin/python: No module named pyspark.wrongdaemon"{*}
but it is not captured by PythonWorkerFactory.
> PythonWorkerFactory does not print error stream in case the daemon fails
> before the main daemon.py#main()
> ---------------------------------------------------------------------------------------------------------
>
> Key: SPARK-48517
> URL: https://issues.apache.org/jira/browse/SPARK-48517
> Project: Spark
> Issue Type: Bug
> Components: PySpark
> Affects Versions: 3.3.2
> Reporter: Prabhu Joseph
> Priority: Major
>
> PythonWorkerFactory does not print the error stream in case the daemon fails
> before executing the main daemon.py#mainI(). It throws SparkException like
> below but does not print the error stream which has the failure why
> pyspark.daemon failed to start.
> {code:java}
> org.apache.spark.SparkException:
> 2024-05-07T16:04:53.169524256Z stderr F Bad data in pyspark.daemon's standard
> output. Invalid port number:
> 2024-05-07T16:04:53.169530703Z stderr F 1097887852 (0x4170706c){code}
> The error stream is being [read|#L303]] after throwing SparkException. It has
> to be captured during the exception as well.
>
> *Simple Repro:*
> 1. Run a sample pyspark job by setting wrong
> spark.python.daemon.module like pyspark.wrongdaemon instead of default one
> pyspark.daemon.
>
> 2. The forked python process will fail with
> {*}"{*}{*}/opt/python/3.9.2/bin/python: No module named
> pyspark.wrongdaemon"{*} but it is not captured by PythonWorkerFactory.
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
