SOLR-1166: Speed up docset/filter generation by avoiding top-level score() call and iterating over leaf readers with TermDocs.

git-svn-id: https://svn.apache.org/repos/asf/lucene/solr/trunk@774946 13f79535-47bb-0310-9956-ffa450edef68

CHANGES.txt

@@ -247,6 +247,10 @@ Optimizations
  9. SOLR-1108: Remove un-needed synchronization in SolrCore constructor.
     (Noble Paul via shalin)
 
+10. SOLR-1166: Speed up docset/filter generation by avoiding top-level
+    score() call and iterating over leaf readers with TermDocs. (yonik)
+
+
 Bug Fixes
 ----------------------
  1. SOLR-774: Fixed logging level display (Sean Timm via Otis Gospodnetic)

src/java/org/apache/solr/search/DocSetHitCollector.java

@@ -18,7 +18,12 @@
 package org.apache.solr.search;
 
 import org.apache.lucene.search.HitCollector;
+import org.apache.lucene.search.Collector;
+import org.apache.lucene.search.Scorer;
 import org.apache.lucene.util.OpenBitSet;
+import org.apache.lucene.index.IndexReader;
+
+import java.io.IOException;
 
 /**
  * @version $Id$
@@ -76,3 +81,66 @@ final class DocSetHitCollector extends HitCollector {
     }
   }
 }
+
+
+class DocSetCollector extends Collector {
+
+  final float HASHSET_INVERSE_LOAD_FACTOR;
+  final int HASHDOCSET_MAXSIZE;
+
+  int pos=0;
+  OpenBitSet bits;
+  final int maxDoc;
+  int base=0;
+
+  // in case there aren't that many hits, we may not want a very sparse
+  // bit array.  Optimistically collect the first few docs in an array
+  // in case there are only a few.
+  final int[] scratch;
+
+  // todo - could pass in bitset and an operation also...
+  DocSetCollector(float inverseLoadFactor, int maxSize, int maxDoc) {
+    this.maxDoc = maxDoc;
+    HASHSET_INVERSE_LOAD_FACTOR = inverseLoadFactor;
+    HASHDOCSET_MAXSIZE = maxSize;
+    scratch = new int[HASHDOCSET_MAXSIZE];
+  }
+
+  public void collect(int doc) {
+    doc += base;
+    // optimistically collect the first docs in an array
+    // in case the total number will be small enough to represent
+    // as a HashDocSet() instead...
+    // Storing in this array will be quicker to convert
+    // than scanning through a potentially huge bit vector.
+    // FUTURE: when search methods all start returning docs in order, maybe
+    // we could have a ListDocSet() and use the collected array directly.
+    if (pos < scratch.length) {
+      scratch[pos]=doc;
+    } else {
+      // this conditional could be removed if BitSet was preallocated, but that
+      // would take up more memory, and add more GC time...
+      if (bits==null) bits = new OpenBitSet(maxDoc);
+      bits.fastSet(doc);
+    }
+
+    pos++;
+  }
+
+  public DocSet getDocSet() {
+    if (pos<=scratch.length) {
+      return new HashDocSet(scratch,0,pos,HASHSET_INVERSE_LOAD_FACTOR);
+    } else {
+      // set the bits for ids that were collected in the array
+      for (int i=0; i<scratch.length; i++) bits.fastSet(scratch[i]);
+      return new BitDocSet(bits,pos);
+    }
+  }
+
+  public void setScorer(Scorer scorer) throws IOException {
+  }
+
+  public void setNextReader(IndexReader reader, int docBase) throws IOException {
+    this.base = docBase;
+  }
+}

src/java/org/apache/solr/search/SolrIndexSearcher.java

@@ -628,15 +628,26 @@ public class SolrIndexSearcher extends IndexSearcher implements SolrInfoMBean {
   // query must be positive
   protected DocSet getDocSetNC(Query query, DocSet filter) throws IOException {
     if (filter==null) {
-      DocSetHitCollector hc = new DocSetHitCollector(HASHSET_INVERSE_LOAD_FACTOR, HASHDOCSET_MAXSIZE, maxDoc());
+      DocSetCollector hc = new DocSetCollector(HASHSET_INVERSE_LOAD_FACTOR, HASHDOCSET_MAXSIZE, maxDoc());
       if (query instanceof TermQuery) {
         Term t = ((TermQuery)query).getTerm();
-        TermDocs tdocs = null;
-        try {
-          tdocs = reader.termDocs(t);
-          while (tdocs.next()) hc.collect(tdocs.doc(),0.0f);
-        } finally {
-          if (tdocs!=null) tdocs.close();
+        SolrIndexReader[] readers = reader.getLeafReaders();
+        int[] offsets = reader.getLeafOffsets();
+        int[] arr = new int[256];
+        int[] freq = new int[256];
+        for (int i=0; i<readers.length; i++) {
+          SolrIndexReader sir = readers[i];
+          int offset = offsets[i];
+          hc.setNextReader(sir, offset);
+          TermDocs tdocs = sir.termDocs(t);
+          for(;;) {
+            int num = tdocs.read(arr, freq);
+            if (num==0) break;
+            while (--num>=0) {
+              hc.collect(arr[num]);
+            }
+          }
+          tdocs.close();
         }
       } else {
         super.search(query,null,hc);
@@ -645,11 +656,20 @@ public class SolrIndexSearcher extends IndexSearcher implements SolrInfoMBean {
 
     } else {
       // FUTURE: if the filter is sorted by docid, could use skipTo (SkipQueryFilter)
-      final DocSetHitCollector hc = new DocSetHitCollector(HASHSET_INVERSE_LOAD_FACTOR, HASHDOCSET_MAXSIZE, maxDoc());
+      final DocSetCollector hc = new DocSetCollector(HASHSET_INVERSE_LOAD_FACTOR, HASHDOCSET_MAXSIZE, maxDoc());
       final DocSet filt = filter;
-      super.search(query, null, new HitCollector() {
-        public void collect(int doc, float score) {
-          if (filt.exists(doc)) hc.collect(doc,score);
+      super.search(query, null, new Collector() {
+        int base = 0;
+        public void collect(int doc) {
+          doc += base;
+          if (filt.exists(doc)) hc.collect(doc);
+        }
+
+        public void setNextReader(IndexReader reader, int docBase) throws IOException {
+          this.base = docBase;  
+        }
+
+        public void setScorer(Scorer scorer) throws IOException {
         }
       }
       );
@@ -1112,7 +1132,7 @@ public class SolrIndexSearcher extends IndexSearcher implements SolrInfoMBean {
     int[] ids;
     float[] scores;
     final DocSetHitCollector setHC = new DocSetHitCollector(HASHSET_INVERSE_LOAD_FACTOR, HASHDOCSET_MAXSIZE, maxDoc());
-    final HitCollector hitCollector = ( cmd.getTimeAllowed() > 0 ) ? new TimeLimitedCollector( setHC, cmd.getTimeAllowed() ) : setHC;
+    final HitCollector collector = ( cmd.getTimeAllowed() > 0 ) ? new TimeLimitedCollector( setHC, cmd.getTimeAllowed() ) : setHC;
 
     Query query = QueryUtils.makeQueryable(cmd.getQuery());
 
@@ -1137,7 +1157,7 @@ public class SolrIndexSearcher extends IndexSearcher implements SolrInfoMBean {
       try {
         super.search(query, new HitCollector() {
           public void collect(int doc, float score) {
-            hitCollector.collect(doc,score);
+            collector.collect(doc,score);
             if (filt!=null && !filt.exists(doc)) return;
             numHits[0]++;
             if (score > topscore[0]) topscore[0]=score;
@@ -1168,7 +1188,7 @@ public class SolrIndexSearcher extends IndexSearcher implements SolrInfoMBean {
         super.search(query, new HitCollector() {
           private FieldDoc reusableFD;
           public void collect(int doc, float score) {
-            hitCollector.collect(doc,score);
+            collector.collect(doc,score);
             if (filt!=null && !filt.exists(doc)) return;
             numHits[0]++;
             if (reusableFD == null)
@@ -1212,7 +1232,7 @@ public class SolrIndexSearcher extends IndexSearcher implements SolrInfoMBean {
         super.search(query, new HitCollector() {
           private ScoreDoc reusableSD;
           public void collect(int doc, float score) {
-            hitCollector.collect(doc,score);
+            collector.collect(doc,score);
             if (filt!=null && !filt.exists(doc)) return;
               // if docs are always delivered in order, we could use "score>minScore"
               // but might BooleanScorer14 might still be used and deliver docs out-of-order?

src/test/org/apache/solr/search/TestSearchPerf.java

@@ -0,0 +1,106 @@
+/**
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.solr.search;
+
+import org.apache.lucene.index.Term;
+import org.apache.lucene.search.TermQuery;
+import org.apache.lucene.search.Query;
+import org.apache.lucene.search.BooleanQuery;
+import org.apache.lucene.search.BooleanClause;
+import org.apache.solr.util.AbstractSolrTestCase;
+import org.apache.solr.request.SolrQueryRequest;
+
+import java.util.Random;
+
+/**
+ * @version $Id$
+ */
+public class TestSearchPerf extends AbstractSolrTestCase {
+
+  public String getSchemaFile() { return "schema11.xml"; }
+  public String getSolrConfigFile() { return "solrconfig.xml"; }
+
+  public void setUp() throws Exception {
+    super.setUp();
+  }
+  public void tearDown() throws Exception {
+    close();
+    super.tearDown();
+  }
+
+  String t(int tnum) {
+    return String.format("%08d", tnum);
+  }
+
+  Random r = new Random(0);
+  int nDocs;
+  void createIndex(int nDocs) {
+    this.nDocs = nDocs;
+    assertU(delQ("*:*"));
+    for (int i=0; i<nDocs; i++) {
+      assertU(adoc("id", Float.toString(i)
+              ,"foo1_s",t(0)
+              ,"foo2_s",t(r.nextInt(2))
+              ,"foo4_s",t(r.nextInt(3))
+      ));
+    }
+    // assertU(optimize()); // squeeze out any possible deleted docs
+    assertU(commit());
+  }
+
+  SolrQueryRequest req; // used to get a searcher
+  void close() {
+    if (req!=null) req.close();
+    req = null;
+  }
+
+  int doSetGen(int iter, Query q) throws Exception {
+    close();
+    req = lrf.makeRequest("q","*:*");
+
+    SolrIndexSearcher searcher = req.getSearcher();
+
+    long start = System.currentTimeMillis();
+
+    int ret = 0;
+    for (int i=0; i<iter; i++) {
+      DocSet set = searcher.getDocSetNC(q, null);
+      ret += set.size();
+    }
+
+    long end = System.currentTimeMillis();
+    System.out.println("ret="+ret+ " time="+(end-start)+" throughput="+iter*1000/(end-start));
+
+    return ret;
+  }
+
+  // prevent complaints by junit
+  public void testEmpty() {
+  }
+
+  public void XtestSetGenerationPerformance() throws Exception {
+    createIndex(49999);
+    doSetGen(10000, new TermQuery(new Term("foo1_s",t(0))) );
+
+    BooleanQuery bq = new BooleanQuery();
+    bq.add(new TermQuery(new Term("foo2_s",t(0))), BooleanClause.Occur.SHOULD);
+    bq.add(new TermQuery(new Term("foo2_s",t(1))), BooleanClause.Occur.SHOULD);
+    doSetGen(5000, bq); 
+  }
+
+}