dc4761fd8f
Because the call to the constructor of HiveClientImpl crosses class loader boundaries, different versions of the same class (Configuration in this case) were loaded, and that caused a runtime error when instantiating the client. By using a safer type in the signature of the constructor, it's possible to avoid the problem. I considered removing 'sharesHadoopClasses', but it may still be desired (even though there are 0 users of it since it was not working). When Spark starts to support Hadoop 3, it may be necessary to use that option to load clients for older Hive metastore versions that don't know about Hadoop 3. Tested with added unit test. Author: Marcelo Vanzin <vanzin@cloudera.com> Closes #20169 from vanzin/SPARK-17088. |
||
---|---|---|
.. | ||
compatibility/src/test/scala/org/apache/spark/sql/hive/execution | ||
src | ||
pom.xml |