Uploaded image for project: 'Spark'
  1. Spark
  2. SPARK-28749

Fix PySpark tests not to require kafka-0-8 in branch-2.4

    XMLWordPrintableJSON

    Details

    • Type: Bug
    • Status: Resolved
    • Priority: Minor
    • Resolution: Fixed
    • Affects Version/s: 2.4.3
    • Fix Version/s: 2.4.5
    • Component/s: PySpark, Tests
    • Labels:
      None

      Description

      As noted in SPARK-27550 we want to encourage testing of Spark 2.4.x with Scala-2.12, and kafka-0-8 does not support Scala-2.12.

      Currently, the PySpark tests invoked by `python/run-tests` demand the presence of kafka-0-8 libraries. If not present, this failure message will be generated:

      Traceback (most recent call last):
       File "/usr/lib64/python2.7/runpy.py", line 174, in _run_module_as_main
       "__main__", fname, loader, pkg_name)
       File "/usr/lib64/python2.7/runpy.py", line 72, in _run_code
       exec code in run_globals
       File "spark/python/pyspark/streaming/tests.py", line 1579, in <module>
       kafka_assembly_jar = search_kafka_assembly_jar()
       File "spark/python/pyspark/streaming/tests.py", line 1524, in search_kafka_assembly_jar
       "You need to build Spark with "
       Exception: Failed to find Spark Streaming kafka assembly jar in spark/external/kafka-0-8-assembly. You need to build Spark with 'build/sbt -Pkafka-0-8 assembly/package streaming-kafka-0-8-assembly/assembly' or 'build/mvn -DskipTests -Pkafka-0-8 package' before running this test.
      
      Had test failures in pyspark.streaming.tests with spark/py_virtenv/bin/python; see logs.
       Process exited with code 255
      

      This change is only targeted at branch-2.4, as most kafka-0-8 related materials have been removed in master and this problem no longer occurs there.

      PROPOSED SOLUTION

      The proposed solution is to make the kafka-0-8 stream testing optional for pyspark testing, exactly the same as the Kinesis stream testing currently is, in file `python/pyspark/streaming/tests.py`. This is only a few lines of change.

      Ideally it would be limited to when SPARK_SCALA_VERSION >= 2.12, but it turns out to be somewhat onerous to reliably obtain that value from within the python test env, and no other python test code currently does so. So my proposed solution simply makes the use of the kafka-0-8 profile optional, and leaves it to the tester to include it for Scala-2.11 test builds and exclude it for Scala-2.12 test builds.

      PR will be available in a day or so.

        Attachments

          Issue Links

            Activity

              People

              • Assignee:
                mattf Matt Foley
                Reporter:
                mattf Matt Foley
                Shepherd:
                Dongjoon Hyun
              • Votes:
                0 Vote for this issue
                Watchers:
                3 Start watching this issue

                Dates

                • Created:
                  Updated:
                  Resolved: