First round of optimizations for vector functions. (#46294)

This PR merges the `vectors-optimize-brute-force` feature branch, which makes the following changes to how vector functions are computed: * Precompute the L2 norm of each vector at indexing time. (#45390) * Switch to ByteBuffer for vector encoding. (#45936) * Decode vectors and while computing the vector function. (#46103) * Use an array instead of a List for the query vector. (#46155) * Precompute the normalized query vector when using cosine similarity. (#46190) Co-authored-by: Mayya Sharipova <[email protected]>
elastic · Sep 4, 2019 · 8e588db · 8e588db
1 parent d5ad86d
commit 8e588db
Show file tree

Hide file tree

Showing 15 changed files with 560 additions and 299 deletions.
diff --git a/docs/reference/mapping/types/dense-vector.asciidoc b/docs/reference/mapping/types/dense-vector.asciidoc
@@ -54,4 +54,4 @@ PUT my_index/_doc/2
 
 Internally, each document's dense vector is encoded as a binary
 doc value. Its size in bytes is equal to
-`4 * dims`, where `dims`—the number of the vector's dimensions.
+`4 * dims + 4`, where `dims`—the number of the vector's dimensions.
diff --git a/docs/reference/mapping/types/sparse-vector.asciidoc b/docs/reference/mapping/types/sparse-vector.asciidoc
@@ -56,5 +56,5 @@ PUT my_index/_doc/2
 
 Internally, each document's sparse vector is encoded as a binary
 doc value. Its size in bytes is equal to
-`6 * NUMBER_OF_DIMENSIONS`, where `NUMBER_OF_DIMENSIONS` -
+`6 * NUMBER_OF_DIMENSIONS + 4`, where `NUMBER_OF_DIMENSIONS` -
 number of the vector's dimensions.
diff --git a/...er/src/main/java/org/elasticsearch/common/lucene/search/function/ScriptScoreFunction.java b/...er/src/main/java/org/elasticsearch/common/lucene/search/function/ScriptScoreFunction.java
@@ -25,6 +25,7 @@
 import org.elasticsearch.script.ExplainableScoreScript;
 import org.elasticsearch.script.ScoreScript;
 import org.elasticsearch.script.Script;
+import org.elasticsearch.Version;
 
 import java.io.IOException;
 import java.util.Objects;
@@ -52,22 +53,24 @@ public float score() {
 
     private final int shardId;
     private final String indexName;
-
+    private final Version indexVersion;
 
     public ScriptScoreFunction(Script sScript, ScoreScript.LeafFactory script) {
         super(CombineFunction.REPLACE);
         this.sScript = sScript;
         this.script = script;
         this.indexName = null;
         this.shardId = -1;
+        this.indexVersion = null;
     }
 
-    public ScriptScoreFunction(Script sScript, ScoreScript.LeafFactory script, String indexName, int shardId) {
+    public ScriptScoreFunction(Script sScript, ScoreScript.LeafFactory script, String indexName, int shardId, Version indexVersion) {
         super(CombineFunction.REPLACE);
         this.sScript = sScript;
         this.script = script;
         this.indexName = indexName;
         this.shardId = shardId;
+        this.indexVersion = indexVersion;
     }
 
     @Override
@@ -77,6 +80,7 @@ public LeafScoreFunction getLeafScoreFunction(LeafReaderContext ctx) throws IOEx
         leafScript.setScorer(scorer);
         leafScript._setIndexName(indexName);
         leafScript._setShard(shardId);
+        leafScript._setIndexVersion(indexVersion);
         return new LeafScoreFunction() {
             @Override
             public double score(int docId, float subQueryScore) throws IOException {

diff --git a/...src/main/java/org/elasticsearch/index/query/functionscore/ScriptScoreFunctionBuilder.java b/...src/main/java/org/elasticsearch/index/query/functionscore/ScriptScoreFunctionBuilder.java
@@ -94,7 +94,8 @@ protected ScoreFunction doToFunction(QueryShardContext context) {
         try {
             ScoreScript.Factory factory = context.getScriptService().compile(script, ScoreScript.CONTEXT);
             ScoreScript.LeafFactory searchScript = factory.newFactory(script.getParams(), context.lookup());
-            return new ScriptScoreFunction(script, searchScript, context.index().getName(), context.getShardId());
+            return new ScriptScoreFunction(script, searchScript,
+                context.index().getName(), context.getShardId(), context.indexVersionCreated());
         } catch (Exception e) {
             throw new QueryShardException(context, "script_score: the script could not be loaded", e);
         }

diff --git a/server/src/main/java/org/elasticsearch/script/ScoreScript.java b/server/src/main/java/org/elasticsearch/script/ScoreScript.java
@@ -23,6 +23,7 @@
 import org.elasticsearch.index.fielddata.ScriptDocValues;
 import org.elasticsearch.search.lookup.LeafSearchLookup;
 import org.elasticsearch.search.lookup.SearchLookup;
+import org.elasticsearch.Version;
 
 import java.io.IOException;
 import java.io.UncheckedIOException;
@@ -66,6 +67,7 @@ public abstract class ScoreScript {
     private int docId;
     private int shardId = -1;
     private String indexName = null;
+    private Version indexVersion = null;
 
     public ScoreScript(Map<String, Object> params, SearchLookup lookup, LeafReaderContext leafContext) {
         // null check needed b/c of expression engine subclass
@@ -165,6 +167,19 @@ public String _getIndex() {
         }
     }
 
+    /**
+     *  Starting a name with underscore, so that the user cannot access this function directly through a script
+     *  It is only used within predefined painless functions.
+     * @return index version or throws an exception if the index version is not set up for this script instance
+     */
+    public Version _getIndexVersion() {
+        if (indexVersion != null) {
+            return indexVersion;
+        } else {
+            throw new IllegalArgumentException("index version can not be looked up!");
+        }
+    }
+
     /**
      *  Starting a name with underscore, so that the user cannot access this function directly through a script
      */
@@ -179,6 +194,13 @@ public void _setIndexName(String indexName) {
         this.indexName = indexName;
     }
 
+    /**
+     *  Starting a name with underscore, so that the user cannot access this function directly through a script
+     */
+    public void _setIndexVersion(Version indexVersion) {
+        this.indexVersion = indexVersion;
+    }
+
 
     /** A factory to construct {@link ScoreScript} instances. */
     public interface LeafFactory {

diff --git a/...ck/plugin/src/test/resources/rest-api-spec/test/vectors/20_dense_vector_special_cases.yml b/...ck/plugin/src/test/resources/rest-api-spec/test/vectors/20_dense_vector_special_cases.yml
@@ -46,7 +46,7 @@ setup:
         index: test-index
         id: 2
         body:
-          my_dense_vector: [10.9, 10.9, 10.9]
+          my_dense_vector: [10.5, 10.9, 10.4]
 
   - do:
       indices.refresh: {}

diff --git a/.../vectors/src/main/java/org/elasticsearch/xpack/vectors/mapper/DenseVectorFieldMapper.java b/.../vectors/src/main/java/org/elasticsearch/xpack/vectors/mapper/DenseVectorFieldMapper.java
@@ -13,6 +13,7 @@
 import org.apache.lucene.search.DocValuesFieldExistsQuery;
 import org.apache.lucene.search.Query;
 import org.apache.lucene.util.BytesRef;
+import org.elasticsearch.Version;
 import org.elasticsearch.common.settings.Settings;
 import org.elasticsearch.common.xcontent.XContentBuilder;
 import org.elasticsearch.common.xcontent.XContentParser.Token;
@@ -25,10 +26,11 @@
 import org.elasticsearch.index.mapper.MapperParsingException;
 import org.elasticsearch.index.mapper.ParseContext;
 import org.elasticsearch.index.query.QueryShardContext;
-import org.elasticsearch.xpack.vectors.query.VectorDVIndexFieldData;
 import org.elasticsearch.search.DocValueFormat;
+import org.elasticsearch.xpack.vectors.query.VectorDVIndexFieldData;
 
 import java.io.IOException;
+import java.nio.ByteBuffer;
 import java.time.ZoneId;
 import java.util.List;
 import java.util.Map;
@@ -180,8 +182,11 @@ public void parse(ParseContext context) throws IOException {
 
         // encode array of floats as array of integers and store into buf
         // this code is here and not int the VectorEncoderDecoder so not to create extra arrays
-        byte[] buf = new byte[dims * INT_BYTES];
-        int offset = 0;
+        byte[] bytes = indexCreatedVersion.onOrAfter(Version.V_7_5_0) ? new byte[dims * INT_BYTES + INT_BYTES] : new byte[dims * INT_BYTES];
+
+        ByteBuffer byteBuffer = ByteBuffer.wrap(bytes);
+        double dotProduct = 0f;
+
         int dim = 0;
         for (Token token = context.parser().nextToken(); token != Token.END_ARRAY; token = context.parser().nextToken()) {
             if (dim++ >= dims) {
@@ -190,18 +195,22 @@ public void parse(ParseContext context) throws IOException {
             }
             ensureExpectedToken(Token.VALUE_NUMBER, token, context.parser()::getTokenLocation);
             float value = context.parser().floatValue(true);
-            int intValue = Float.floatToIntBits(value);
-            buf[offset++] = (byte) (intValue >> 24);
-            buf[offset++] = (byte) (intValue >> 16);
-            buf[offset++] = (byte) (intValue >>  8);
-            buf[offset++] = (byte) intValue;
+
+            byteBuffer.putFloat(value);
+            dotProduct += value * value;
         }
         if (dim != dims) {
             throw new IllegalArgumentException("Field [" + name() + "] of type [" + typeName() + "] of doc [" +
                 context.sourceToParse().id() + "] has number of dimensions [" + dim +
                 "] less than defined in the mapping [" +  dims +"]");
         }
-        BinaryDocValuesField field = new BinaryDocValuesField(fieldType().name(), new BytesRef(buf, 0, offset));
+
+        if (indexCreatedVersion.onOrAfter(Version.V_7_5_0)) {
+            // encode vector magnitude at the end
+            float vectorMagnitude = (float) Math.sqrt(dotProduct);
+            byteBuffer.putFloat(vectorMagnitude);
+        }
+        BinaryDocValuesField field = new BinaryDocValuesField(fieldType().name(), new BytesRef(bytes));
         if (context.doc().getByKey(fieldType().name()) != null) {
             throw new IllegalArgumentException("Field [" + name() + "] of type [" + typeName() +
                 "] doesn't not support indexing multiple values for the same field in the same document");

diff --git a/...vectors/src/main/java/org/elasticsearch/xpack/vectors/mapper/SparseVectorFieldMapper.java b/...vectors/src/main/java/org/elasticsearch/xpack/vectors/mapper/SparseVectorFieldMapper.java
@@ -181,7 +181,7 @@ public void parse(ParseContext context) throws IOException {
             }
         }
 
-        BytesRef br = VectorEncoderDecoder.encodeSparseVector(dims, values, dimCount);
+        BytesRef br = VectorEncoderDecoder.encodeSparseVector(indexCreatedVersion, dims, values, dimCount);
         BinaryDocValuesField field = new BinaryDocValuesField(fieldType().name(), br);
         context.doc().addWithKey(fieldType().name(), field);
     }

diff --git a/...in/vectors/src/main/java/org/elasticsearch/xpack/vectors/mapper/VectorEncoderDecoder.java b/...in/vectors/src/main/java/org/elasticsearch/xpack/vectors/mapper/VectorEncoderDecoder.java
@@ -9,6 +9,9 @@
 
 import org.apache.lucene.util.BytesRef;
 import org.apache.lucene.util.InPlaceMergeSorter;
+import org.elasticsearch.Version;
+
+import java.nio.ByteBuffer;
 
 // static utility functions for encoding and decoding dense_vector and sparse_vector fields
 public final class VectorEncoderDecoder {
@@ -19,81 +22,84 @@ private VectorEncoderDecoder() { }
 
     /**
      * Encodes a sparse array represented by values, dims and dimCount into a bytes array - BytesRef
-     * BytesRef: int[] floats encoded as integers values, 2 bytes for each dimension
-     * @param values - values of the sparse array
+     * BytesRef: int[] floats encoded as integers values, 2 bytes for each dimension, length of vector
+     * @param indexVersion - index version
      * @param dims - dims of the sparse array
+     * @param values - values of the sparse array
      * @param dimCount - number of the dimensions, necessary as values and dims are dynamically created arrays,
      *          and may be over-allocated
      * @return BytesRef
      */
-    public static BytesRef encodeSparseVector(int[] dims, float[] values, int dimCount) {
+    public static BytesRef encodeSparseVector(Version indexVersion, int[] dims, float[] values, int dimCount) {
         // 1. Sort dims and values
         sortSparseDimsValues(dims, values, dimCount);
-        byte[] buf = new byte[dimCount * (INT_BYTES + SHORT_BYTES)];
 
         // 2. Encode dimensions
         // as each dimension is a positive value that doesn't exceed 65535, 2 bytes is enough for encoding it
-        int offset = 0;
+        byte[] bytes = indexVersion.onOrAfter(Version.V_7_5_0) ? new byte[dimCount * (INT_BYTES + SHORT_BYTES) + INT_BYTES] :
+            new byte[dimCount * (INT_BYTES + SHORT_BYTES)];
+        ByteBuffer byteBuffer = ByteBuffer.wrap(bytes);
+
         for (int dim = 0; dim < dimCount; dim++) {
-            buf[offset] = (byte) (dims[dim] >>  8);
-            buf[offset+1] = (byte) dims[dim];
-            offset += SHORT_BYTES;
+            int dimValue = dims[dim];
+            byteBuffer.put((byte) (dimValue >> 8));
+            byteBuffer.put((byte) dimValue);
         }
 
         // 3. Encode values
+        double dotProduct = 0.0f;
         for (int dim = 0; dim < dimCount; dim++) {
-            int intValue = Float.floatToIntBits(values[dim]);
-            buf[offset] =  (byte) (intValue >> 24);
-            buf[offset+1] = (byte) (intValue >> 16);
-            buf[offset+2] = (byte) (intValue >>  8);
-            buf[offset+3] = (byte) intValue;
-            offset += INT_BYTES;
+            float value = values[dim];
+            byteBuffer.putFloat(value);
+            dotProduct += value * value;
         }
 
-        return new BytesRef(buf);
+        // 4. Encode vector magnitude at the end
+        if (indexVersion.onOrAfter(Version.V_7_5_0)) {
+            float vectorMagnitude = (float) Math.sqrt(dotProduct);
+            byteBuffer.putFloat(vectorMagnitude);
+        }
+
+        return new BytesRef(bytes);
     }
 
     /**
      * Decodes the first part of BytesRef into sparse vector dimensions
+     * @param indexVersion - index version
      * @param vectorBR - sparse vector encoded in BytesRef
      */
-    public static int[] decodeSparseVectorDims(BytesRef vectorBR) {
-        if (vectorBR == null) {
-            throw new IllegalArgumentException("A document doesn't have a value for a vector field!");
-        }
-        int dimCount = vectorBR.length / (INT_BYTES + SHORT_BYTES);
+    public static int[] decodeSparseVectorDims(Version indexVersion, BytesRef vectorBR) {
+        int dimCount = indexVersion.onOrAfter(Version.V_7_5_0)
+            ? (vectorBR.length - INT_BYTES) / (INT_BYTES + SHORT_BYTES)
+            : vectorBR.length / (INT_BYTES + SHORT_BYTES);
+        ByteBuffer byteBuffer = ByteBuffer.wrap(vectorBR.bytes, vectorBR.offset, dimCount * SHORT_BYTES);
+
         int[] dims = new int[dimCount];
-        int offset = vectorBR.offset;
         for (int dim = 0; dim < dimCount; dim++) {
-            dims[dim] = ((vectorBR.bytes[offset] & 0xFF) << 8) | (vectorBR.bytes[offset+1] & 0xFF);
-            offset += SHORT_BYTES;
+            dims[dim] = ((byteBuffer.get() & 0xFF) << 8) | (byteBuffer.get() & 0xFF);
         }
         return dims;
     }
 
     /**
      * Decodes the second part of the BytesRef into sparse vector values
+     * @param indexVersion - index version
      * @param vectorBR - sparse vector encoded in BytesRef
      */
-    public static float[] decodeSparseVector(BytesRef vectorBR) {
-        if (vectorBR == null) {
-            throw new IllegalArgumentException("A document doesn't have a value for a vector field!");
-        }
-        int dimCount = vectorBR.length / (INT_BYTES + SHORT_BYTES);
-        int offset =  vectorBR.offset + SHORT_BYTES * dimCount; //calculate the offset from where values are encoded
+    public static float[] decodeSparseVector(Version indexVersion, BytesRef vectorBR) {
+        int dimCount = indexVersion.onOrAfter(Version.V_7_5_0)
+            ? (vectorBR.length - INT_BYTES) / (INT_BYTES + SHORT_BYTES)
+            : vectorBR.length / (INT_BYTES + SHORT_BYTES);
+        int offset =  vectorBR.offset + SHORT_BYTES * dimCount;
         float[] vector = new float[dimCount];
+
+        ByteBuffer byteBuffer = ByteBuffer.wrap(vectorBR.bytes, offset, dimCount * INT_BYTES);
         for (int dim = 0; dim < dimCount; dim++) {
-            int intValue = ((vectorBR.bytes[offset] & 0xFF) << 24)   |
-                ((vectorBR.bytes[offset+1] & 0xFF) << 16) |
-                ((vectorBR.bytes[offset+2] & 0xFF) <<  8) |
-                (vectorBR.bytes[offset+3] & 0xFF);
-            vector[dim] = Float.intBitsToFloat(intValue);
-            offset = offset + INT_BYTES;
+            vector[dim] = byteBuffer.getFloat();
         }
         return vector;
     }
 
-
     /**
      * Sorts dimensions in the ascending order and
      * sorts values in the same order as their corresponding dimensions
@@ -150,24 +156,20 @@ public void swap(int i, int j) {
         }.sort(0, n);
     }
 
+    public static int denseVectorLength(Version indexVersion, BytesRef vectorBR) {
+        return indexVersion.onOrAfter(Version.V_7_5_0)
+            ? (vectorBR.length - INT_BYTES) / INT_BYTES
+            : vectorBR.length / INT_BYTES;
+    }
+
     /**
-     * Decodes a BytesRef into an array of floats
-     * @param vectorBR - dense vector encoded in BytesRef
+     * Decodes the last 4 bytes of the encoded vector, which contains the vector magnitude.
+     * NOTE: this function can only be called on vectors from an index version greater than or
+     * equal to 7.5.0, since vectors created prior to that do not store the magnitude.
      */
-    public static float[] decodeDenseVector(BytesRef vectorBR) {
-        if (vectorBR == null) {
-            throw new IllegalArgumentException("A document doesn't have a value for a vector field!");
-        }
-        int dimCount = vectorBR.length / INT_BYTES;
-        float[] vector = new float[dimCount];
-        int offset = vectorBR.offset;
-        for (int dim = 0; dim < dimCount; dim++) {
-            int intValue = ((vectorBR.bytes[offset++] & 0xFF) << 24)   |
-                ((vectorBR.bytes[offset++] & 0xFF) << 16) |
-                ((vectorBR.bytes[offset++] & 0xFF) <<  8) |
-                (vectorBR.bytes[offset++] & 0xFF);
-            vector[dim] = Float.intBitsToFloat(intValue);
-        }
-        return vector;
+    public static float decodeVectorMagnitude(Version indexVersion, BytesRef vectorBR) {
+        assert indexVersion.onOrAfter(Version.V_7_5_0);
+        ByteBuffer byteBuffer = ByteBuffer.wrap(vectorBR.bytes, vectorBR.offset, vectorBR.length);
+        return byteBuffer.getFloat(vectorBR.offset + vectorBR.length - 4);
     }
 }