Apache
/
lucene
mirror of https://github.com/apache/lucene.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

SOLR-8559: FCS facet performance optimization 

Significantly speeds up processing when terms are high cardinality and the matching docset is small. 
When facet minCount > 0 and the number of matching documents is small (or 0) this enhancement 
prevents considering terms which have a 0 count. Also includes change to move to the next non-zero 
term value when selecting a segment position.



git-svn-id: https://svn.apache.org/repos/asf/lucene/dev/trunk@1725638 13f79535-47bb-0310-9956-ffa450edef68
This commit is contained in:
Dennis Gove 2016-01-19 22:41:49 +00:00
parent cecb9f4e25
commit 2d2582e4f4
2 changed files with 29 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
solr/CHANGES.txt
View File

@ -165,6 +165,12 @@ Optimizations
speedups were up to 2.5x for production of filters, and up to 10x for query evaluation with
embedded term range queres that resulted in filter cache hits. (yonik)
* SOLR-8559: FCS facet performance optimization which significantly speeds up processing when terms
are high cardinality and the matching docset is small. When facet minCount > 0 and the number of
matching documents is small (or 0) this enhancement prevents considering terms which have a 0
count. Also includes change to move to the next non-zero term value when selecting a segment
position. (Keith Laban, Steve Bower, Dennis Gove)
Other Changes
----------------------

29
solr/core/src/java/org/apache/solr/request/PerSegmentSingleValuedFaceting.java
View File

@ -163,7 +163,7 @@ class PerSegmentSingleValuedFaceting {
} else {
seg.pos = seg.startTermIndex;
}
if (seg.pos < seg.endTermIndex) {
if (seg.pos < seg.endTermIndex && (mincount < 1 || seg.hasAnyCount)) {
seg.tenum = seg.si.termsEnum();
seg.tenum.seekExact(seg.pos);
seg.tempBR = seg.tenum.term();
@ -201,14 +201,22 @@ class PerSegmentSingleValuedFaceting {
count += seg.counts[seg.pos - seg.startTermIndex];
}
// TODO: OPTIMIZATION...
// if mincount>0 then seg.pos++ can skip ahead to the next non-zero entry.
seg.pos++;
do{
++seg.pos;
}
while(
(seg.pos < seg.endTermIndex) //stop incrementing before we run off the end
&& (seg.tenum.next() != null || true) //move term enum forward with position -- dont care about value
&& (mincount > 0) //only skip ahead if mincount > 0
&& (seg.counts[seg.pos - seg.startTermIndex] == 0) //check zero count
);
if (seg.pos >= seg.endTermIndex) {
queue.pop();
seg = queue.top();
} else {
seg.tempBR = seg.tenum.next();
seg.tempBR = seg.tenum.term();
seg = queue.updateTop();
}
} while (seg != null && val.get().compareTo(seg.tempBR) == 0);
@ -248,6 +256,10 @@ class PerSegmentSingleValuedFaceting {
int startTermIndex;
int endTermIndex;
int[] counts;
//whether this segment has any non-zero term counts
//used to ignore insignificant segments when mincount>0
boolean hasAnyCount = false;
int pos; // only used when merging
TermsEnum tenum; // only used when merging
@ -285,7 +297,9 @@ class PerSegmentSingleValuedFaceting {
// specialized version when collecting counts for all terms
int doc;
while ((doc = iter.nextDoc()) < DocIdSetIterator.NO_MORE_DOCS) {
counts[1+si.getOrd(doc)]++;
int t = 1+si.getOrd(doc);
hasAnyCount = hasAnyCount || t > 0; //counts[0] == missing counts
counts[t]++;
}
} else {
// version that adjusts term numbers because we aren't collecting the full range
@ -293,7 +307,10 @@ class PerSegmentSingleValuedFaceting {
while ((doc = iter.nextDoc()) < DocIdSetIterator.NO_MORE_DOCS) {
int term = si.getOrd(doc);
int arrIdx = term-startTermIndex;
if (arrIdx>=0 && arrIdx<nTerms) counts[arrIdx]++;
if (arrIdx>=0 && arrIdx<nTerms){
counts[arrIdx]++;
hasAnyCount = true;
}
}
}
}