[HIVE-23685] Removing user's extra resources while executing File Merge Task

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: All Versions
• Component/s: Physical Optimizer, Query Planning
• Labels: pull-request-available
• Assignee: Qiang.Kang
• Reporter: Qiang.Kang
• Created: 13/Jun/20 04:47
• Updated: 05/Nov/20 05:14
• Resolved: 03/Nov/20 08:17

Description

Hi, we find that MapReduce's file merge map containers will download user's extra resources(such as: added jars, files, archives) before launching task. When these resources are large or the network is busy, file merge jobs will be timeout, causing the query be failed. As we all know, file merge task will run correctly just with hive-exec.jar and MapReduce framework. Therefore, there is no need to download user's resources. The patch below prevents setting `tmpjars` for FileMerge Task.

Issue Links

• links to: GitHub Pull Request #1608