[DRILL-5163] External sort on Mac creates a separate child process per spill via HDFS FS

• Type: Improvement
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 1.8.0
• Fix Version/s: 1.11.0
• Assignee: Paul Rogers
• Reporter: Paul Rogers
• Created: 24/Dec/16 22:02
• Updated: 22/Jun/17 05:22
• Resolved: 22/Jun/17 05:22

Description

The external sort operator spills to disk. Spill files are created and written using the HDFS file system. For performance, HDFS uses native libraries to access the file system. These native libraries are not available on the Mac. As a result, some operations are implemented using a shower, Java-only path. One of these operations (need details) is implemented by forking a child process.

When run in a debugger on the Mac, the behavior shows up as the furious creation and deletion of threads to manage the child processes: one per spill. Because of this behavior, performance of external sort is slow. Of course, no production code uses Drill on a Mac, so this is more of a nuisance than a real bug, which is why it is marked as an improvement.

Issue Links

• Is contained by: DRILL-5325 Implement sub-operator unit tests for managed external sort