Improving columnar scanning of doc values #135711
Description
This meta issue tracks the work to improve columnar scanning of doc values in compute engine with the tsdb doc values codec (which are used with logsdb and tsdb). Improving support for all other index modes will be covered by another meta issue.
Follow up from #128445.
Immediate tasks:
- Push down loading of singleton dense binary doc values in the es819 doc value codec. (@parkertimmins)
- Push down loading of singleton dense boolean field types to the es819 doc value codec.
Bulk reading sparse doc values works in some cases, but not all cases. Tasks around improving sparse single-valued doc values:
- Push down loading of singleton sparse number based field types to the es819 doc value codec.
- Push down loading of singleton sparse keyword field types to the es819 doc value codec.
Currently if a doc values is multi-valued no bulk reading occurs. Tasks around improving scanning multi-valued doc values:
- After upgrading to Lucene 10.3, make use of
IndexedDISI#docIDRunEnd()implementation to in some cases support bulk loading in the es819 doc value codec. - Push down loading of multi-valued number based field types to to the es819 doc value codec
- Push down loading of multi-valued keyword field types to to the es819 doc value codec