Sasilekha opened a new pull request, #16663:
URL: https://github.com/apache/lucene/pull/16663

   MemoryAccountingBitsetCollector used to reserve a bitset large enough for 
the entire leaf upfront, even if the query matched nothing in that leaf. Under 
intra segment concurrency (where each partition worker gets its own collector), 
unmatched partitions were paying the same upfront cost.
   This PR delays the allocation until the first match, so collectors that see 
no matches allocate nothing.
   Similar in spirit to #16452 (which tightened the result bitset in 
Manager.reduce())
   
   tracker.getBytes() on a 100k-doc index, single-hit TermQuery:
   | Scenario | Before | After |
   |---|---|---|
   | Match-all (baseline) | ~26,656 B | ~25,912 B (no regression) |
   | Selective 1-hit, single-threaded | ~14,160 B | ~96 B |
   | Selective 1-hit, 4-way intra-segment | ~61,296 B | ~96 B |
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to