Manybubbles has submitted this change and it was merged.
Change subject: Phrase optimization
......................................................................
Phrase optimization
If the query is just phrases and the field contains no phrases then skip
trying to highlight it. This works for matched_fields as well.
Change-Id: Ieacf9abb3be9dd7ba1bab89ad754e8ba5cef27f6
---
M
experimental-highlighter-elasticsearch-plugin/src/main/java/org/elasticsearch/search/highlight/ExperimentalHighlighter.java
M
experimental-highlighter-elasticsearch-plugin/src/main/java/org/elasticsearch/search/highlight/FieldWrapper.java
M
experimental-highlighter-elasticsearch-plugin/src/test/java/org/wikimedia/highlighter/experimental/elasticsearch/ExperimentalHighlighterTest.java
M
experimental-highlighter-lucene/src/main/java/org/wikimedia/highlighter/experimental/lucene/hit/weight/BasicQueryWeigher.java
4 files changed, 179 insertions(+), 82 deletions(-)
Approvals:
Chad: Looks good to me, but someone else must approve
Manybubbles: Looks good to me, approved
jenkins-bot: Verified
diff --git
a/experimental-highlighter-elasticsearch-plugin/src/main/java/org/elasticsearch/search/highlight/ExperimentalHighlighter.java
b/experimental-highlighter-elasticsearch-plugin/src/main/java/org/elasticsearch/search/highlight/ExperimentalHighlighter.java
index 286a8a4..3fc545c 100644
---
a/experimental-highlighter-elasticsearch-plugin/src/main/java/org/elasticsearch/search/highlight/ExperimentalHighlighter.java
+++
b/experimental-highlighter-elasticsearch-plugin/src/main/java/org/elasticsearch/search/highlight/ExperimentalHighlighter.java
@@ -30,6 +30,7 @@
import org.wikimedia.search.highlighter.experimental.SnippetChooser;
import org.wikimedia.search.highlighter.experimental.SnippetFormatter;
import org.wikimedia.search.highlighter.experimental.SnippetWeigher;
+import org.wikimedia.search.highlighter.experimental.hit.EmptyHitEnum;
import org.wikimedia.search.highlighter.experimental.hit.MergingHitEnum;
import
org.wikimedia.search.highlighter.experimental.hit.OverlapMergingHitEnumWrapper;
import
org.wikimedia.search.highlighter.experimental.snippet.BasicScoreBasedSnippetChooser;
@@ -184,6 +185,9 @@
private HitEnum buildHitFindingHitEnum() throws IOException {
Set<String> matchedFields =
context.field.fieldOptions().matchedFields();
if (matchedFields == null) {
+ if (!defaultField.canProduceHits()) {
+ return EmptyHitEnum.INSTANCE;
+ }
return defaultField.buildHitEnum();
}
List<HitEnum> toMerge = new
ArrayList<HitEnum>(matchedFields.size());
@@ -195,9 +199,17 @@
} else {
wrapper = new FieldWrapper(this, context, weigher, field);
}
- toMerge.add(wrapper.buildHitEnum());
+ if (wrapper.canProduceHits()) {
+ toMerge.add(wrapper.buildHitEnum());
+ }
extraFields.add(wrapper);
}
+ if (toMerge.size() == 0) {
+ return EmptyHitEnum.INSTANCE;
+ }
+ if (toMerge.size() == 1) {
+ return toMerge.get(0);
+ }
return new MergingHitEnum(toMerge, HitEnum.LessThans.OFFSETS);
}
diff --git
a/experimental-highlighter-elasticsearch-plugin/src/main/java/org/elasticsearch/search/highlight/FieldWrapper.java
b/experimental-highlighter-elasticsearch-plugin/src/main/java/org/elasticsearch/search/highlight/FieldWrapper.java
index 911823a..84d57c6 100644
---
a/experimental-highlighter-elasticsearch-plugin/src/main/java/org/elasticsearch/search/highlight/FieldWrapper.java
+++
b/experimental-highlighter-elasticsearch-plugin/src/main/java/org/elasticsearch/search/highlight/FieldWrapper.java
@@ -29,9 +29,9 @@
import org.wikimedia.search.highlighter.experimental.SourceExtracter;
import org.wikimedia.search.highlighter.experimental.hit.ConcatHitEnum;
import
org.wikimedia.search.highlighter.experimental.hit.PositionBoostingHitEnumWrapper;
-import
org.wikimedia.search.highlighter.experimental.hit.WeightFilteredHitEnumWrapper;
import
org.wikimedia.search.highlighter.experimental.hit.ReplayingHitEnum.HitEnumAndLength;
import org.wikimedia.search.highlighter.experimental.hit.TermWeigher;
+import
org.wikimedia.search.highlighter.experimental.hit.WeightFilteredHitEnumWrapper;
import
org.wikimedia.search.highlighter.experimental.hit.weight.CachingTermWeigher;
import
org.wikimedia.search.highlighter.experimental.hit.weight.MultiplyingTermWeigher;
import org.wikimedia.search.highlighter.experimental.snippet.MultiSegmenter;
@@ -110,8 +110,6 @@
}
public SourceExtracter<String> buildSourceExtracter() throws IOException {
- // TODO loading source is expensive if you don't need it. Delay
- // this.
List<String> fieldValues = getFieldValues();
switch (fieldValues.size()) {
case 0:
@@ -156,6 +154,15 @@
}
}
+ /**
+ * Can this field produce hits? This'll return false when there are no
terms
+ * worth anything that aren't in phrases AND there aren't any phrases on
+ * this field.
+ */
+ public boolean canProduceHits() {
+ return weigher.maxTermWeight() > 0 ||
weigher.areTherePhrasesOnField(context.fieldName);
+ }
+
public HitEnum buildHitEnum() throws IOException {
HitEnum e = buildHitEnumForSource();
diff --git
a/experimental-highlighter-elasticsearch-plugin/src/test/java/org/wikimedia/highlighter/experimental/elasticsearch/ExperimentalHighlighterTest.java
b/experimental-highlighter-elasticsearch-plugin/src/test/java/org/wikimedia/highlighter/experimental/elasticsearch/ExperimentalHighlighterTest.java
index 19a723d..0e357e9 100644
---
a/experimental-highlighter-elasticsearch-plugin/src/test/java/org/wikimedia/highlighter/experimental/elasticsearch/ExperimentalHighlighterTest.java
+++
b/experimental-highlighter-elasticsearch-plugin/src/test/java/org/wikimedia/highlighter/experimental/elasticsearch/ExperimentalHighlighterTest.java
@@ -25,6 +25,7 @@
import static org.hamcrest.Matchers.both;
import static org.hamcrest.Matchers.containsString;
import static org.hamcrest.Matchers.equalTo;
+import static org.hamcrest.Matchers.lessThan;
import java.io.IOException;
import java.util.ArrayList;
@@ -108,6 +109,53 @@
}
}
+ /**
+ * Makes sure we skip fields if the query is just phrases and none of them
are on the field.
+ */
+ @Test
+ public void phraseQueryNoPossiblePhrasesSpeed() throws IOException {
+ buildIndex();
+ // Size has to be big enough to make it really expensive to highlight
+ // the field to exaggerate the issue
+ int size = 100000;
+ int iterations = 100;
+ StringBuilder b = new StringBuilder(5 * size);
+ for (int i = 0; i < size; i++) {
+ b.append("test ");
+ }
+ client().prepareIndex("test", "test", "1").setSource("test",
b.toString(), "test2", "simple test").get();
+ refresh();
+
+ SearchRequestBuilder search =
setHitSource(testSearch(matchPhraseQuery("test2", "simple test")), "analyze");
+ long start = System.currentTimeMillis();
+ for (int i = 0; i < iterations; i++) {
+ SearchResponse response = search.get();
+ assertNotHighlighted(response, 0, "test");
+ }
+ long total = System.currentTimeMillis() - start;
+ // Without the optimization this runs about 7 seconds, with, about 1.8.
+ assertThat(total, lessThan(3000l));
+
+ search.addHighlightedField(new
HighlightBuilder.Field("test2").matchedFields("test.english",
"test.whitespace"));
+ start = System.currentTimeMillis();
+ for (int i = 0; i < iterations; i++) {
+ SearchResponse response = search.get();
+ assertNotHighlighted(response, 0, "test");
+ }
+ total = System.currentTimeMillis() - start;
+ // Without the optimization this runs about 7 seconds, with, about 1.8.
+ assertThat(total, lessThan(3000l));
+
+ search.addHighlightedField(new
HighlightBuilder.Field("test2").matchedFields("test", "test2.whitespace"));
+ start = System.currentTimeMillis();
+ for (int i = 0; i < iterations; i++) {
+ assertNoFailures(search.get());
+ }
+ total = System.currentTimeMillis() - start;
+ // Without the optimization this runs about 7 seconds, with, about 1.8.
+ assertThat(total, lessThan(3000l));
+ }
+
@Test
public void phraseWithSeparations() throws IOException {
buildIndex();
diff --git
a/experimental-highlighter-lucene/src/main/java/org/wikimedia/highlighter/experimental/lucene/hit/weight/BasicQueryWeigher.java
b/experimental-highlighter-lucene/src/main/java/org/wikimedia/highlighter/experimental/lucene/hit/weight/BasicQueryWeigher.java
index 64da7e2..128a34d 100644
---
a/experimental-highlighter-lucene/src/main/java/org/wikimedia/highlighter/experimental/lucene/hit/weight/BasicQueryWeigher.java
+++
b/experimental-highlighter-lucene/src/main/java/org/wikimedia/highlighter/experimental/lucene/hit/weight/BasicQueryWeigher.java
@@ -16,7 +16,6 @@
import org.apache.lucene.util.automaton.ByteRunAutomaton;
import org.apache.lucene.util.automaton.CompiledAutomaton;
import org.wikimedia.highlighter.experimental.lucene.QueryFlattener;
-import org.wikimedia.highlighter.experimental.lucene.QueryFlattener.Callback;
import org.wikimedia.search.highlighter.experimental.HitEnum;
import org.wikimedia.search.highlighter.experimental.hit.PhraseHitEnumWrapper;
import org.wikimedia.search.highlighter.experimental.hit.TermSourceFinder;
@@ -32,6 +31,7 @@
private final List<AutomatonSourceInfo> automata = new
ArrayList<AutomatonSourceInfo>();
private final List<BytesRef> terms = new ArrayList<BytesRef>();
private final TermInfos termInfos;
+ private final float maxTermWeight;
private Map<String, List<PhraseInfo>> phrases;
private CompiledAutomaton acceptable;
@@ -41,83 +41,9 @@
public BasicQueryWeigher(QueryFlattener flattener, final TermInfos
termInfos, IndexReader reader, Query query) {
this.termInfos = termInfos;
- flattener.flatten(query, reader, new Callback() {
- private int[][] phrase;
- private int phrasePosition;
- private int phraseTerm;
-
- @Override
- public void flattened(BytesRef term, float boost, Object
rewritten) {
- int source = rewritten == null ? term.hashCode() :
rewritten.hashCode();
- SourceInfo info = termInfos.get(term);
- if (info == null) {
- info = new SourceInfo();
- info.source = source;
- info.weight = boost;
- termInfos.put(term, info);
- terms.add(BytesRef.deepCopyOf(term));
- } else {
- /*
- * If both terms can't be traced back to the same source we
- * declare that they are from a new source by merging the
- * hashes. This might not be ideal, but it has the
advantage
- * of being consistent.
- */
- if (info.source != source) {
- info.source = source * 31 + source;
- }
- info.weight = Math.max(info.weight, boost);
- }
- if (phrase != null) {
- phrase[phrasePosition][phraseTerm++] = info.source;
- }
- }
-
- @Override
- public void flattened(Automaton automaton, float boost, int
source) {
- AutomatonSourceInfo info = new AutomatonSourceInfo(automaton);
- // Automata don't have a hashcode so we always use the source
- info.source = source;
- info.weight = boost;
- automata.add(info);
- if (phrase != null) {
- phrase[phrasePosition][phraseTerm++] = info.source;
- }
- }
-
- @Override
- public void startPhrase(int termCount) {
- phrase = new int[termCount][];
- phrasePosition = -1;
- }
- @Override
- public void startPhrasePosition(int termCount) {
- phrasePosition++;
- phrase[phrasePosition] = new int[termCount];
- phraseTerm = 0;
- }
- @Override
- public void endPhrasePosition() {
- }
- @Override
- public void endPhrase(String field, int slop, float weight) {
- List<PhraseInfo> phraseList;
- if (phrases == null) {
- phrases = new HashMap<String, List<PhraseInfo>>();
- phraseList = new ArrayList<PhraseInfo>();
- phrases.put(field, phraseList);
- } else {
- phraseList = phrases.get(field);
- if (phraseList == null) {
- phraseList = new ArrayList<PhraseInfo>();
- phrases.put(field, phraseList);
- }
- }
- phraseList.add(new PhraseInfo(phrase, slop, weight));
- phrase = null;
- }
- });
-
+ FlattenerCallback callback = new FlattenerCallback();
+ flattener.flatten(query, reader, callback);
+ maxTermWeight = callback.maxTermWeight;
}
public boolean singleTerm() {
@@ -151,6 +77,27 @@
e = new PhraseHitEnumWrapper(e, phrase.phrase, phrase.weight,
phrase.slop);
}
return e;
+ }
+
+ /**
+ * The maximum weight of a single term without phrase queries.
+ */
+ public float maxTermWeight() {
+ return maxTermWeight;
+ }
+
+ /**
+ * Are there phrases on the provided field?
+ */
+ public boolean areTherePhrasesOnField(String field) {
+ if (phrases == null) {
+ return false;
+ }
+ List<PhraseInfo> phraseList = phrases.get(field);
+ if (phraseList == null) {
+ return false;
+ }
+ return !phraseList.isEmpty();
}
public CompiledAutomaton acceptableTerms() {
@@ -267,4 +214,87 @@
return b.toString();
}
}
+
+ private class FlattenerCallback implements QueryFlattener.Callback {
+ private float maxTermWeight = 0;
+ private int[][] phrase;
+ private int phrasePosition;
+ private int phraseTerm;
+
+ @Override
+ public void flattened(BytesRef term, float boost, Object rewritten) {
+ maxTermWeight = Math.max(maxTermWeight, boost);
+ int source = rewritten == null ? term.hashCode() :
rewritten.hashCode();
+ SourceInfo info = termInfos.get(term);
+ if (info == null) {
+ info = new SourceInfo();
+ info.source = source;
+ info.weight = boost;
+ termInfos.put(term, info);
+ terms.add(BytesRef.deepCopyOf(term));
+ } else {
+ /*
+ * If both terms can't be traced back to the same source we
+ * declare that they are from a new source by merging the
+ * hashes. This might not be ideal, but it has the advantage
+ * of being consistent.
+ */
+ if (info.source != source) {
+ info.source = source * 31 + source;
+ }
+ info.weight = Math.max(info.weight, boost);
+ }
+ if (phrase != null) {
+ phrase[phrasePosition][phraseTerm++] = info.source;
+ }
+ }
+
+ @Override
+ public void flattened(Automaton automaton, float boost, int source) {
+ maxTermWeight = Math.max(maxTermWeight, boost);
+ AutomatonSourceInfo info = new AutomatonSourceInfo(automaton);
+ // Automata don't have a hashcode so we always use the source
+ info.source = source;
+ info.weight = boost;
+ automata.add(info);
+ if (phrase != null) {
+ phrase[phrasePosition][phraseTerm++] = info.source;
+ }
+ }
+
+ @Override
+ public void startPhrase(int termCount) {
+ phrase = new int[termCount][];
+ phrasePosition = -1;
+ }
+
+ @Override
+ public void startPhrasePosition(int termCount) {
+ phrasePosition++;
+ phrase[phrasePosition] = new int[termCount];
+ phraseTerm = 0;
+ }
+
+ @Override
+ public void endPhrasePosition() {
+ }
+
+ @Override
+ public void endPhrase(String field, int slop, float weight) {
+ List<PhraseInfo> phraseList;
+ if (phrases == null) {
+ phrases = new HashMap<String, List<PhraseInfo>>();
+ phraseList = new ArrayList<PhraseInfo>();
+ phrases.put(field, phraseList);
+ } else {
+ phraseList = phrases.get(field);
+ if (phraseList == null) {
+ phraseList = new ArrayList<PhraseInfo>();
+ phrases.put(field, phraseList);
+ }
+ }
+ phraseList.add(new PhraseInfo(phrase, slop, weight));
+ phrase = null;
+ }
+ }
}
--
To view, visit https://gerrit.wikimedia.org/r/132964
To unsubscribe, visit https://gerrit.wikimedia.org/r/settings
Gerrit-MessageType: merged
Gerrit-Change-Id: Ieacf9abb3be9dd7ba1bab89ad754e8ba5cef27f6
Gerrit-PatchSet: 1
Gerrit-Project: search/highlighter
Gerrit-Branch: master
Gerrit-Owner: Manybubbles <[email protected]>
Gerrit-Reviewer: BryanDavis <[email protected]>
Gerrit-Reviewer: Chad <[email protected]>
Gerrit-Reviewer: Manybubbles <[email protected]>
Gerrit-Reviewer: jenkins-bot <>
_______________________________________________
MediaWiki-commits mailing list
[email protected]
https://lists.wikimedia.org/mailman/listinfo/mediawiki-commits
