Optimize StoredFields merge strategy based on chunk deletion ratio
- Dominant language
- Java
- Stars
- 3.6k
- Forks
- 1.4k
- Avg merge
- 2d 11h
- Merged PRs (30d)
- 88
Description
### Description
For StoredField merging, the merge strategy has to switch from the `BULK` mode to the `DOC` mode even if only a document is deleted, which significantly increases merging overhead.
https://github.com/apache/lucene/blob/2dee50e21f868aaae436c1dc47d6678564d69b4d/lucene/core/src/java/org/apache/lucene/codecs/lucene90/compressing/Lucene90CompressingStoredFieldsWriter.java#L684
StoredField is compressed at the chunk granularity I propose introducing a new merge strategy:
- For the chunk with no deleted documents, retain `copyChunk `during merging;
- For the chunk containing deleted documents, fall back to `copyOneDoc`.
Furthermore, we can leverage the segment deletion ratio (e.g., a threshold below 5%) as a condition to enable or disable this optimized merging logic.
Contributor guide
Research direction
Start in lucene/core/src/java/org/apache/lucene/codecs/lucene90/compressing/Lucene90CompressingStoredFieldsWriter.java around line 684 and trace the current BULK/DOC merge selection. Confirm the intended behavior for chunks with and without deletions, including the deletion-ratio threshold, and use the existing stored-fields merge tests or benchmarks if found nearby to verify the optimization.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- java
- Domain
- search
- Issue type
- Refactor
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 52/100