Hi Shawn I am not aware of the rationale for using *savedClassLoader.getParent()*. As next steps on resolving this issue, can you create a bug and provide details of the problem you are seeing. This will help us validate the fix.
For now, I made the change to *AtlasPluginClassLoader *and did some basic sanity tests. The tests were all positive. I hope these are satisfactory next steps. I will attach the patch to the JIRA and then open review item. Best regards, *~ *ashutosh On Wed, Mar 20, 2019 at 4:52 PM Shawn Weeks <[email protected]> wrote: > Hi, I'm tracking down an issue with Hive ClassLoading for UDFs and I > discovered that after calling the Atlas Hook Hive's current thread had a > different class loader than before the hook. I'm looking in > AtlasPluginClassLoader.java at the deactivate method and it looks like it's > setting the threads class loader back to savedClassLoader.getParent() > instead of savedClassLoader. Why is this? It's causing Hive to add UDF Jars > to the wrong classloader. > > Thanks > Shawn Weeks > > -- Ashutosh Mestry <[email protected]> . Staff Software Engineer . +1-310-988-0670 . Cloudera, Inc. .......... *"No hurry, no pause." -- Tim Ferriss, Author*
