[ORC-1232] [C++]Disable metrics collector by default

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.9.0
• Fix Version/s: 1.9.0
• Assignee: ZhangXin
• Reporter: Xinyu Zeng
• Created: 29/Jul/22 09:25
• Updated: 09/Aug/22 06:22
• Resolved: 05/Aug/22 15:41

Description

ORC-961 introduced a metrics collector for the reader. However, it may affect the performance of reading ORC files. It may be helpful to disable it as default.

 

Reproducable experiment result:

Alibaba Cloud ecs.s6-c1m4.xlarge, running Ubuntu 20.04, ESSD PL1 40GB

The original file is 4.1GB csv file with generated string with some degree of repetiveness (the value of one column follows a zipfian distribution). The ORC file with dictionary encoding and no block compression is 319MB.

 

Time of running orc-scan with metrics enabled: 7.5s

Time of running orc-scan with metrics disabled: 1.5s

The action of disable is implemented by adding 

readerOpts.setReaderMetrics(nullptr); 

after https://github.com/apache/orc/blob/02e48107b36b8ed868797dadcd7355a632519d48/tools/src/FileScan.cc#L26

Issue Links

• links to: GitHub Pull Request #1206