[LUCENE-10031] Speedup to SortedDocIDMerger when sorting on low-cardinality fields

• Type: Improvement
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 8.10
• Lucene Fields: New
• Assignee: Unassigned
• Reporter: Adrien Grand
• Created: 21/Jul/21 17:18
• Updated: 28/Aug/22 16:23
• Resolved: 29/Jul/21 13:33

Description

I've been looking at profiles of indexing with index sorting enabled and saw non-negligible time spent in SortedDocIDMerger. This isn't completely surprising as this little class is called on every document whenever merging postings, doc values, stored fields, etc.

I'm especially interested in cases when the sort key is on a low cardinality field, so the priority queue doesn't get reordered often. I've been playing with a change to SortedDocIdMerger that makes merging significantly faster in that case.

Issue Links

• links to: GitHub Pull Request #221