Propagate scoring function through random sampler (elastic#116957)

* Propagate scoring function through random sampler. * Update docs/changelog/116957.yaml * Correct score mode in random sampler weight * Fix random sampling with scores and p=1.0 * Unit test with scores * YAML test * Add capability
afoucret · Nov 20, 2024 · dea1e7d · dea1e7d
1 parent c3f73d0
commit dea1e7d
Show file tree

Hide file tree

Showing 8 changed files with 150 additions and 55 deletions.
diff --git a/docs/changelog/116957.yaml b/docs/changelog/116957.yaml
@@ -0,0 +1,5 @@
+pr: 116957
+summary: Propagate scoring function through random sampler
+area: Machine Learning
+type: bug
+issues: [ 110134 ]
diff --git a/modules/aggregations/build.gradle b/modules/aggregations/build.gradle
@@ -20,7 +20,7 @@ esplugin {
 
 restResources {
   restApi {
-    include '_common', 'indices', 'cluster', 'index', 'search', 'nodes', 'bulk', 'scripts_painless_execute', 'put_script'
+    include 'capabilities', '_common', 'indices', 'cluster', 'index', 'search', 'nodes', 'bulk', 'scripts_painless_execute', 'put_script'
   }
   restTests {
     // Pulls in all aggregation tests from core AND the forwards v7's core for forwards compatibility

diff --git a/...ggregations/src/yamlRestTest/resources/rest-api-spec/test/aggregations/random_sampler.yml b/...ggregations/src/yamlRestTest/resources/rest-api-spec/test/aggregations/random_sampler.yml
@@ -142,6 +142,66 @@ setup:
           }
   - match: { aggregations.sampled.mean.value: 1.0 }
 ---
+"Test random_sampler aggregation with scored subagg":
+  - requires:
+      capabilities:
+        - method: POST
+          path: /_search
+          capabilities: [ random_sampler_with_scored_subaggs ]
+      test_runner_features: capabilities
+      reason: "Support for random sampler with scored subaggs capability required"
+  - do:
+      search:
+        index: data
+        size: 0
+        body: >
+          {
+            "query": {
+              "function_score": {
+                "random_score": {}
+              }
+            },
+            "aggs": {
+              "sampled": {
+                "random_sampler": {
+                  "probability": 0.5
+                },
+                "aggs": {
+                  "top": {
+                    "top_hits": {}
+                  }
+                }
+              }
+            }
+          }
+  - is_true: aggregations.sampled.top.hits
+  - do:
+      search:
+        index: data
+        size: 0
+        body: >
+          {
+            "query": {
+              "function_score": {
+                "random_score": {}
+              }
+            },
+            "aggs": {
+              "sampled": {
+                "random_sampler": {
+                  "probability": 1.0
+                },
+                "aggs": {
+                  "top": {
+                    "top_hits": {}
+                  }
+                }
+              }
+            }
+          }
+  - match: { aggregations.sampled.top.hits.total.value: 6 }
+  - is_true: aggregations.sampled.top.hits.hits.0._score
+---
 "Test random_sampler aggregation with poor settings":
   - requires:
       cluster_features: ["gte_v8.2.0"]

diff --git a/server/src/main/java/org/elasticsearch/rest/action/search/SearchCapabilities.java b/server/src/main/java/org/elasticsearch/rest/action/search/SearchCapabilities.java
@@ -41,6 +41,8 @@ private SearchCapabilities() {}
     /** Support multi-dense-vector script field access. */
     private static final String MULTI_DENSE_VECTOR_SCRIPT_ACCESS = "multi_dense_vector_script_access";
 
+    private static final String RANDOM_SAMPLER_WITH_SCORED_SUBAGGS = "random_sampler_with_scored_subaggs";
+
     public static final Set<String> CAPABILITIES;
     static {
         HashSet<String> capabilities = new HashSet<>();
@@ -50,6 +52,7 @@ private SearchCapabilities() {}
         capabilities.add(DENSE_VECTOR_DOCVALUE_FIELDS);
         capabilities.add(TRANSFORM_RANK_RRF_TO_RETRIEVER);
         capabilities.add(NESTED_RETRIEVER_INNER_HITS_SUPPORT);
+        capabilities.add(RANDOM_SAMPLER_WITH_SCORED_SUBAGGS);
         if (MultiDenseVectorFieldMapper.FEATURE_FLAG.isEnabled()) {
             capabilities.add(MULTI_DENSE_VECTOR_FIELD_MAPPER);
             capabilities.add(MULTI_DENSE_VECTOR_SCRIPT_ACCESS);

diff --git a/server/src/main/java/org/elasticsearch/search/aggregations/AggregatorBase.java b/server/src/main/java/org/elasticsearch/search/aggregations/AggregatorBase.java
@@ -40,7 +40,7 @@ public abstract class AggregatorBase extends Aggregator {
 
     protected final String name;
     protected final Aggregator parent;
-    private final AggregationContext context;
+    protected final AggregationContext context;
     private final Map<String, Object> metadata;
 
     protected final Aggregator[] subAggregators;

diff --git a/.../org/elasticsearch/search/aggregations/bucket/sampler/random/RandomSamplerAggregator.java b/.../org/elasticsearch/search/aggregations/bucket/sampler/random/RandomSamplerAggregator.java
@@ -9,19 +9,23 @@
 
 package org.elasticsearch.search.aggregations.bucket.sampler.random;
 
+import org.apache.lucene.search.BooleanClause;
+import org.apache.lucene.search.BooleanQuery;
 import org.apache.lucene.search.CollectionTerminatedException;
 import org.apache.lucene.search.DocIdSetIterator;
+import org.apache.lucene.search.Query;
+import org.apache.lucene.search.ScoreMode;
 import org.apache.lucene.search.Scorer;
 import org.apache.lucene.search.Weight;
 import org.apache.lucene.util.Bits;
-import org.elasticsearch.common.CheckedSupplier;
 import org.elasticsearch.common.util.LongArray;
 import org.elasticsearch.search.aggregations.AggregationExecutionContext;
 import org.elasticsearch.search.aggregations.Aggregator;
 import org.elasticsearch.search.aggregations.AggregatorFactories;
 import org.elasticsearch.search.aggregations.CardinalityUpperBound;
 import org.elasticsearch.search.aggregations.InternalAggregation;
 import org.elasticsearch.search.aggregations.LeafBucketCollector;
+import org.elasticsearch.search.aggregations.LeafBucketCollectorBase;
 import org.elasticsearch.search.aggregations.bucket.BucketsAggregator;
 import org.elasticsearch.search.aggregations.bucket.SingleBucketAggregator;
 import org.elasticsearch.search.aggregations.support.AggregationContext;
@@ -34,14 +38,13 @@ public class RandomSamplerAggregator extends BucketsAggregator implements Single
     private final int seed;
     private final Integer shardSeed;
     private final double probability;
-    private final CheckedSupplier<Weight, IOException> weightSupplier;
+    private Weight weight;
 
     RandomSamplerAggregator(
         String name,
         int seed,
         Integer shardSeed,
         double probability,
-        CheckedSupplier<Weight, IOException> weightSupplier,
         AggregatorFactories factories,
         AggregationContext context,
         Aggregator parent,
@@ -56,10 +59,33 @@ public class RandomSamplerAggregator extends BucketsAggregator implements Single
                 RandomSamplerAggregationBuilder.NAME + " aggregation [" + name + "] must have sub aggregations configured"
             );
         }
-        this.weightSupplier = weightSupplier;
         this.shardSeed = shardSeed;
     }
 
+    /**
+     * This creates the query weight which will be used in the aggregator.
+     *
+     * This weight is a boolean query between {@link RandomSamplingQuery} and the configured top level query of the search. This allows
+     * the aggregation to iterate the documents directly, thus sampling in the background instead of the foreground.
+     * @return weight to be used, is cached for additional usages
+     * @throws IOException when building the weight or queries fails;
+     */
+    private Weight getWeight() throws IOException {
+        if (weight == null) {
+            ScoreMode scoreMode = scoreMode();
+            BooleanQuery.Builder fullQuery = new BooleanQuery.Builder().add(
+                context.query(),
+                scoreMode.needsScores() ? BooleanClause.Occur.MUST : BooleanClause.Occur.FILTER
+            );
+            if (probability < 1.0) {
+                Query sampleQuery = new RandomSamplingQuery(probability, seed, shardSeed == null ? context.shardRandomSeed() : shardSeed);
+                fullQuery.add(sampleQuery, BooleanClause.Occur.FILTER);
+            }
+            weight = context.searcher().createWeight(context.searcher().rewrite(fullQuery.build()), scoreMode, 1f);
+        }
+        return weight;
+    }
+
     @Override
     public InternalAggregation[] buildAggregations(LongArray owningBucketOrds) throws IOException {
         return buildAggregationsForSingleBucket(
@@ -101,22 +127,26 @@ protected LeafBucketCollector getLeafCollector(AggregationExecutionContext aggCt
         if (sub.isNoop()) {
             return LeafBucketCollector.NO_OP_COLLECTOR;
         }
+
+        Scorer scorer = getWeight().scorer(aggCtx.getLeafReaderContext());
+        // This means there are no docs to iterate, possibly due to the fields not existing
+        if (scorer == null) {
+            return LeafBucketCollector.NO_OP_COLLECTOR;
+        }
+        sub.setScorer(scorer);
+
         // No sampling is being done, collect all docs
+        // TODO know when sampling would be much slower and skip sampling: https://github.com/elastic/elasticsearch/issues/84353
         if (probability >= 1.0) {
             grow(1);
-            return new LeafBucketCollector() {
+            return new LeafBucketCollectorBase(sub, null) {
                 @Override
                 public void collect(int doc, long owningBucketOrd) throws IOException {
                     collectExistingBucket(sub, doc, 0);
                 }
             };
         }
-        // TODO know when sampling would be much slower and skip sampling: https://github.com/elastic/elasticsearch/issues/84353
-        Scorer scorer = weightSupplier.get().scorer(aggCtx.getLeafReaderContext());
-        // This means there are no docs to iterate, possibly due to the fields not existing
-        if (scorer == null) {
-            return LeafBucketCollector.NO_OP_COLLECTOR;
-        }
+
         final DocIdSetIterator docIt = scorer.iterator();
         final Bits liveDocs = aggCtx.getLeafReaderContext().reader().getLiveDocs();
         try {
@@ -136,5 +166,4 @@ public void collect(int doc, long owningBucketOrd) throws IOException {
         // Since we have done our own collection, there is nothing for the leaf collector to do
         return LeafBucketCollector.NO_OP_COLLECTOR;
     }
-
 }
diff --git a/...asticsearch/search/aggregations/bucket/sampler/random/RandomSamplerAggregatorFactory.java b/...asticsearch/search/aggregations/bucket/sampler/random/RandomSamplerAggregatorFactory.java
@@ -9,10 +9,6 @@
 
 package org.elasticsearch.search.aggregations.bucket.sampler.random;
 
-import org.apache.lucene.search.BooleanClause;
-import org.apache.lucene.search.BooleanQuery;
-import org.apache.lucene.search.ScoreMode;
-import org.apache.lucene.search.Weight;
 import org.elasticsearch.search.aggregations.Aggregator;
 import org.elasticsearch.search.aggregations.AggregatorFactories;
 import org.elasticsearch.search.aggregations.AggregatorFactory;
@@ -30,7 +26,6 @@ public class RandomSamplerAggregatorFactory extends AggregatorFactory {
     private final Integer shardSeed;
     private final double probability;
     private final SamplingContext samplingContext;
-    private Weight weight;
 
     RandomSamplerAggregatorFactory(
         String name,
@@ -57,41 +52,6 @@ public Optional<SamplingContext> getSamplingContext() {
     @Override
     public Aggregator createInternal(Aggregator parent, CardinalityUpperBound cardinality, Map<String, Object> metadata)
         throws IOException {
-        return new RandomSamplerAggregator(
-            name,
-            seed,
-            shardSeed,
-            probability,
-            this::getWeight,
-            factories,
-            context,
-            parent,
-            cardinality,
-            metadata
-        );
+        return new RandomSamplerAggregator(name, seed, shardSeed, probability, factories, context, parent, cardinality, metadata);
     }
-
-    /**
-     * This creates the query weight which will be used in the aggregator.
-     *
-     * This weight is a boolean query between {@link RandomSamplingQuery} and the configured top level query of the search. This allows
-     * the aggregation to iterate the documents directly, thus sampling in the background instead of the foreground.
-     * @return weight to be used, is cached for additional usages
-     * @throws IOException when building the weight or queries fails;
-     */
-    private Weight getWeight() throws IOException {
-        if (weight == null) {
-            RandomSamplingQuery query = new RandomSamplingQuery(
-                probability,
-                seed,
-                shardSeed == null ? context.shardRandomSeed() : shardSeed
-            );
-            BooleanQuery booleanQuery = new BooleanQuery.Builder().add(query, BooleanClause.Occur.FILTER)
-                .add(context.query(), BooleanClause.Occur.FILTER)
-                .build();
-            weight = context.searcher().createWeight(context.searcher().rewrite(booleanQuery), ScoreMode.COMPLETE_NO_SCORES, 1f);
-        }
-        return weight;
-    }
-
 }
diff --git a/...elasticsearch/search/aggregations/bucket/sampler/random/RandomSamplerAggregatorTests.java b/...elasticsearch/search/aggregations/bucket/sampler/random/RandomSamplerAggregatorTests.java
@@ -11,22 +11,29 @@
 
 import org.apache.lucene.document.LongPoint;
 import org.apache.lucene.document.SortedNumericDocValuesField;
+import org.apache.lucene.index.Term;
+import org.apache.lucene.search.BooleanClause;
+import org.apache.lucene.search.BooleanQuery;
+import org.apache.lucene.search.TermQuery;
 import org.apache.lucene.tests.index.RandomIndexWriter;
 import org.apache.lucene.util.BytesRef;
 import org.elasticsearch.common.Strings;
 import org.elasticsearch.index.mapper.KeywordFieldMapper;
 import org.elasticsearch.index.query.QueryBuilders;
+import org.elasticsearch.search.SearchHit;
 import org.elasticsearch.search.aggregations.AggregationBuilders;
 import org.elasticsearch.search.aggregations.AggregatorTestCase;
 import org.elasticsearch.search.aggregations.bucket.filter.Filter;
 import org.elasticsearch.search.aggregations.metrics.Avg;
 import org.elasticsearch.search.aggregations.metrics.Max;
 import org.elasticsearch.search.aggregations.metrics.Min;
+import org.elasticsearch.search.aggregations.metrics.TopHits;
 import org.hamcrest.Description;
 import org.hamcrest.Matcher;
 import org.hamcrest.TypeSafeMatcher;
 
 import java.io.IOException;
+import java.util.Arrays;
 import java.util.List;
 import java.util.concurrent.atomic.AtomicInteger;
 import java.util.stream.DoubleStream;
@@ -37,6 +44,8 @@
 import static org.hamcrest.Matchers.equalTo;
 import static org.hamcrest.Matchers.greaterThan;
 import static org.hamcrest.Matchers.greaterThanOrEqualTo;
+import static org.hamcrest.Matchers.hasSize;
+import static org.hamcrest.Matchers.lessThan;
 import static org.hamcrest.Matchers.lessThanOrEqualTo;
 import static org.hamcrest.Matchers.not;
 import static org.hamcrest.Matchers.notANumber;
@@ -76,6 +85,35 @@ public void testAggregationSampling() throws IOException {
         assertThat(avgAvg, closeTo(1.5, 0.5));
     }
 
+    public void testAggregationSampling_withScores() throws IOException {
+        long[] counts = new long[5];
+        AtomicInteger integer = new AtomicInteger();
+        do {
+            testCase(RandomSamplerAggregatorTests::writeTestDocs, (InternalRandomSampler result) -> {
+                counts[integer.get()] = result.getDocCount();
+                if (result.getDocCount() > 0) {
+                    TopHits agg = result.getAggregations().get("top");
+                    List<SearchHit> hits = Arrays.asList(agg.getHits().getHits());
+                    assertThat(Strings.toString(result), hits, hasSize(1));
+                    assertThat(Strings.toString(result), hits.get(0).getScore(), allOf(greaterThan(0.0f), lessThan(1.0f)));
+                }
+            },
+                new AggTestConfig(
+                    new RandomSamplerAggregationBuilder("my_agg").subAggregation(AggregationBuilders.topHits("top").size(1))
+                        .setProbability(0.25),
+                    longField(NUMERIC_FIELD_NAME)
+                ).withQuery(
+                    new BooleanQuery.Builder().add(
+                        new TermQuery(new Term(KEYWORD_FIELD_NAME, KEYWORD_FIELD_VALUE)),
+                        BooleanClause.Occur.SHOULD
+                    ).build()
+                )
+            );
+        } while (integer.incrementAndGet() < 5);
+        long avgCount = LongStream.of(counts).sum() / integer.get();
+        assertThat(avgCount, allOf(greaterThanOrEqualTo(20L), lessThanOrEqualTo(70L)));
+    }
+
     public void testAggregationSamplingNestedAggsScaled() throws IOException {
         // in case 0 docs get sampled, which can rarely happen
         // in case the test index has many segments.