atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-510028794
@jpountz JFYI I ran ant beast in 3 batches of 10 times each and it ran clean:
WARNING: All illegal access
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-510007530
> @atris Can you give some result comparison or other evidence?
Please see the blog post listed above.
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-509928095
Not just the cost of prepopulating the PQ, but also building it in first
place. If total number of hits < number of
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-509655033
I just discovered ant beast that seems like the right tool for ensuring that
such random tests get reproduced and
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-509599328
@jpountz Thanks for looking through!
I refactored the tests to remove the dependency on random() to populate
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-509588693
@jpountz @mikemccand Thanks for your comments, fixed the same.
I am surprised that the test is failing : it
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-509534291
Added a dedicated test for checking order of hits when TopDocs are returned
from just the hits list and no PQ is
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-509525838
@jpountz Thanks, I have fixed the comments. The reason I did not use the
BooleanQuery for the PQ build and not
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-509518620
FYI ant precommit passes -- ran it with latest iteration
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-509207157
@jpountz I have updated the PR per your comments. ant precommit passes.
Apologies, this iteration also got
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-509169211
@jpountz Thanks for the comments, update the PR.
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-509109583
@jpountz I have pushed a new iteration which does as discussed i.e. builds a
hits list and populates hits as long
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-508914746
@jpountz @tokee Thanks for your comments.
I am planning to maintain an ArrayList of ScoreDocs and collect
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-508804369
> Actually I don't think we need a growable priority queue. For such large
number of hits it'd be probably more
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-508784416
> Not prepopulating the hit queue is only one part of the problem, we would
also need to not allocate `numHits`
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-508377686
Any thoughts on this?
This is an automated message
atris commented on issue #754: LUCENE-8875: Introduce Optimized Collector For
Large Number Of Hits
URL: https://github.com/apache/lucene-solr/pull/754#issuecomment-507544094
I ran some tests with N requested as 100K but the matching docs being only
around 10K. This gives a sizeable
17 matches
Mail list logo