pan3793 commented on PR #8522: URL: https://github.com/apache/hadoop/pull/8522#issuecomment-4795344080
> This is complex but oracle are forcing it on us, and the path doesn't get exercised until java25, so it'll be left for people testing hadoop 3.5.1+ on java25 to find bugs asap. @stoty actually pointed out this issue years ago, but I didn't realize its severity at the time. I had verified that Spark on YARN ran correctly on a Kerberos-enabled Hadoop cluster using JDK 25, so I concluded that Spark was unaffected; it wasn't until I recently tried running Spark on K8s that I discovered this change presented a much trickier problem than I had thought. (PR's description "How was this patch tested?" section explains why this happens in detail) @steveloughran, that's true. I verified this patch in our internal Spark and Hadoop forks with both Spark on YARN and K8s workloads, but not sure if I missed any edge cases. If we agree with this direction, I am going to revert the `SubjectInheritingThread` related code. And there are some open questions: - Should `SubjectInheritingThread` itself be removed? (1) It was designed as a public API and has been released in 3.5.0, but it is not necessary after this patch. (2) The implementation has a semantic issue - HADOOP-19907 SubjectInheritingThread should capture Subject in constructor instead of start(), if we want to keep it, we must fix this issue. - Someone requested backporting this patch to branch-3.4 (see comments at the end of HADOOP-19486), do we want to do this? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
