Merge branch 'main' into remove-unused-field

Author: Yuristry

benchmarks/src/main/java/org/elasticsearch/benchmark/vector/OSQScorerBenchmark.java

@@ -0,0 +1,204 @@
+/*
+ * Copyright Elasticsearch B.V. and/or licensed to Elasticsearch B.V. under one
+ * or more contributor license agreements. Licensed under the "Elastic License
+ * 2.0", the "GNU Affero General Public License v3.0 only", and the "Server Side
+ * Public License v 1"; you may not use this file except in compliance with, at
+ * your election, the "Elastic License 2.0", the "GNU Affero General Public
+ * License v3.0 only", or the "Server Side Public License, v 1".
+ */
+package org.elasticsearch.benchmark.vector;
+
+import org.apache.lucene.index.VectorSimilarityFunction;
+import org.apache.lucene.store.Directory;
+import org.apache.lucene.store.IOContext;
+import org.apache.lucene.store.IndexInput;
+import org.apache.lucene.store.IndexOutput;
+import org.apache.lucene.store.MMapDirectory;
+import org.apache.lucene.util.VectorUtil;
+import org.apache.lucene.util.quantization.OptimizedScalarQuantizer;
+import org.elasticsearch.common.logging.LogConfigurator;
+import org.elasticsearch.simdvec.internal.vectorization.ES91OSQVectorsScorer;
+import org.elasticsearch.simdvec.internal.vectorization.ESVectorizationProvider;
+import org.openjdk.jmh.annotations.Benchmark;
+import org.openjdk.jmh.annotations.BenchmarkMode;
+import org.openjdk.jmh.annotations.Fork;
+import org.openjdk.jmh.annotations.Measurement;
+import org.openjdk.jmh.annotations.Mode;
+import org.openjdk.jmh.annotations.OutputTimeUnit;
+import org.openjdk.jmh.annotations.Param;
+import org.openjdk.jmh.annotations.Scope;
+import org.openjdk.jmh.annotations.Setup;
+import org.openjdk.jmh.annotations.State;
+import org.openjdk.jmh.annotations.Warmup;
+import org.openjdk.jmh.infra.Blackhole;
+
+import java.io.IOException;
+import java.nio.file.Files;
+import java.util.Random;
+import java.util.concurrent.TimeUnit;
+
+@BenchmarkMode(Mode.Throughput)
+@OutputTimeUnit(TimeUnit.MILLISECONDS)
+@State(Scope.Benchmark)
+// first iteration is complete garbage, so make sure we really warmup
+@Warmup(iterations = 4, time = 1)
+// real iterations. not useful to spend tons of time here, better to fork more
+@Measurement(iterations = 5, time = 1)
+// engage some noise reduction
+@Fork(value = 1)
+public class OSQScorerBenchmark {
+
+    static {
+        LogConfigurator.configureESLogging(); // native access requires logging to be initialized
+    }
+
+    @Param({ "1024" })
+    int dims;
+
+    int length;
+
+    int numVectors = ES91OSQVectorsScorer.BULK_SIZE * 10;
+    int numQueries = 10;
+
+    byte[][] binaryVectors;
+    byte[][] binaryQueries;
+    OptimizedScalarQuantizer.QuantizationResult result;
+    float centroidDp;
+
+    byte[] scratch;
+    ES91OSQVectorsScorer scorer;
+
+    IndexInput in;
+
+    float[] scratchScores;
+    float[] corrections;
+
+    @Setup
+    public void setup() throws IOException {
+        Random random = new Random(123);
+
+        this.length = OptimizedScalarQuantizer.discretize(dims, 64) / 8;
+
+        binaryVectors = new byte[numVectors][length];
+        for (byte[] binaryVector : binaryVectors) {
+            random.nextBytes(binaryVector);
+        }
+
+        Directory dir = new MMapDirectory(Files.createTempDirectory("vectorData"));
+        IndexOutput out = dir.createOutput("vectors", IOContext.DEFAULT);
+        byte[] correctionBytes = new byte[14 * ES91OSQVectorsScorer.BULK_SIZE];
+        for (int i = 0; i < numVectors; i += ES91OSQVectorsScorer.BULK_SIZE) {
+            for (int j = 0; j < ES91OSQVectorsScorer.BULK_SIZE; j++) {
+                out.writeBytes(binaryVectors[i + j], 0, binaryVectors[i + j].length);
+            }
+            random.nextBytes(correctionBytes);
+            out.writeBytes(correctionBytes, 0, correctionBytes.length);
+        }
+        out.close();
+        in = dir.openInput("vectors", IOContext.DEFAULT);
+
+        binaryQueries = new byte[numVectors][4 * length];
+        for (byte[] binaryVector : binaryVectors) {
+            random.nextBytes(binaryVector);
+        }
+        result = new OptimizedScalarQuantizer.QuantizationResult(
+            random.nextFloat(),
+            random.nextFloat(),
+            random.nextFloat(),
+            Short.toUnsignedInt((short) random.nextInt())
+        );
+        centroidDp = random.nextFloat();
+
+        scratch = new byte[length];
+        scorer = ESVectorizationProvider.getInstance().newES91OSQVectorsScorer(in, dims);
+        scratchScores = new float[16];
+        corrections = new float[3];
+    }
+
+    @Benchmark
+    @Fork(jvmArgsPrepend = { "--add-modules=jdk.incubator.vector" })
+    public void scoreFromArray(Blackhole bh) throws IOException {
+        for (int j = 0; j < numQueries; j++) {
+            in.seek(0);
+            for (int i = 0; i < numVectors; i++) {
+                in.readBytes(scratch, 0, length);
+                float qDist = VectorUtil.int4BitDotProduct(binaryQueries[j], scratch);
+                in.readFloats(corrections, 0, corrections.length);
+                int addition = Short.toUnsignedInt(in.readShort());
+                float score = scorer.score(
+                    result,
+                    VectorSimilarityFunction.EUCLIDEAN,
+                    centroidDp,
+                    corrections[0],
+                    corrections[1],
+                    addition,
+                    corrections[2],
+                    qDist
+                );
+                bh.consume(score);
+            }
+        }
+    }
+
+    @Benchmark
+    @Fork(jvmArgsPrepend = { "--add-modules=jdk.incubator.vector" })
+    public void scoreFromMemorySegmentOnlyVector(Blackhole bh) throws IOException {
+        for (int j = 0; j < numQueries; j++) {
+            in.seek(0);
+            for (int i = 0; i < numVectors; i++) {
+                float qDist = scorer.quantizeScore(binaryQueries[j]);
+                in.readFloats(corrections, 0, corrections.length);
+                int addition = Short.toUnsignedInt(in.readShort());
+                float score = scorer.score(
+                    result,
+                    VectorSimilarityFunction.EUCLIDEAN,
+                    centroidDp,
+                    corrections[0],
+                    corrections[1],
+                    addition,
+                    corrections[2],
+                    qDist
+                );
+                bh.consume(score);
+            }
+        }
+    }
+
+    @Benchmark
+    @Fork(jvmArgsPrepend = { "--add-modules=jdk.incubator.vector" })
+    public void scoreFromMemorySegmentOnlyVectorBulk(Blackhole bh) throws IOException {
+        for (int j = 0; j < numQueries; j++) {
+            in.seek(0);
+            for (int i = 0; i < numVectors; i += 16) {
+                scorer.quantizeScoreBulk(binaryQueries[j], ES91OSQVectorsScorer.BULK_SIZE, scratchScores);
+                for (int k = 0; k < ES91OSQVectorsScorer.BULK_SIZE; k++) {
+                    in.readFloats(corrections, 0, corrections.length);
+                    int addition = Short.toUnsignedInt(in.readShort());
+                    float score = scorer.score(
+                        result,
+                        VectorSimilarityFunction.EUCLIDEAN,
+                        centroidDp,
+                        corrections[0],
+                        corrections[1],
+                        addition,
+                        corrections[2],
+                        scratchScores[k]
+                    );
+                    bh.consume(score);
+                }
+            }
+        }
+    }
+
+    @Benchmark
+    @Fork(jvmArgsPrepend = { "--add-modules=jdk.incubator.vector" })
+    public void scoreFromMemorySegmentAllBulk(Blackhole bh) throws IOException {
+        for (int j = 0; j < numQueries; j++) {
+            in.seek(0);
+            for (int i = 0; i < numVectors; i += 16) {
+                scorer.scoreBulk(binaryQueries[j], result, VectorSimilarityFunction.EUCLIDEAN, centroidDp, scratchScores);
+                bh.consume(scratchScores);
+            }
+        }
+    }
+}

docs/changelog/125922.yaml

@@ -0,0 +1,5 @@
+pr: 125922 
+summary: Fix text structure NPE when fields in list have null value
+area: Machine Learning
+type: bug
+issues: []

docs/changelog/126629.yaml

@@ -0,0 +1,5 @@
+pr: 126629
+summary: Default new `semantic_text` fields to use BBQ when models are compatible
+area: Relevance
+type: enhancement
+issues: []

docs/changelog/127285.yaml

@@ -0,0 +1,5 @@
+pr: 127285
+summary: Restore model registry validation for the semantic text field
+area: Search
+type: enhancement
+issues: []

docs/reference/enrich-processor/date-processor.md

@@ -6,7 +6,6 @@ mapped_pages:
 
 # Date processor [date-processor]
 
-
 Parses dates from fields, and then uses the date or timestamp as the timestamp for the document. By default, the date processor adds the parsed date as a new field called `@timestamp`. You can specify a different field by setting the `target_field` configuration parameter. Multiple date formats are supported as part of the same date processor definition. They will be used sequentially to attempt parsing the date field, in the same order they were defined as part of the processor definition.
 
 $$$date-options$$$
@@ -16,7 +15,7 @@ $$$date-options$$$
 | `field` | yes | - | The field to get the date from. |
 | `target_field` | no | @timestamp | The field that will hold the parsed date. |
 | `formats` | yes | - | An array of the expected date formats. Can be a [java time pattern](/reference/elasticsearch/mapping-reference/mapping-date-format.md) or one of the following formats: ISO8601, UNIX, UNIX_MS, or TAI64N. |
-| `timezone` | no | UTC | The timezone to use when parsing the date. Supports [template snippets](docs-content://manage-data/ingest/transform-enrich/ingest-pipelines.md#template-snippets). |
+| `timezone` | no | UTC | The default [timezone](#date-processor-timezones) used by the processor. Supports [template snippets](docs-content://manage-data/ingest/transform-enrich/ingest-pipelines.md#template-snippets). |
 | `locale` | no | ENGLISH | The locale to use when parsing the date, relevant when parsing month names or week days. Supports [template snippets](docs-content://manage-data/ingest/transform-enrich/ingest-pipelines.md#template-snippets). |
 | `output_format` | no | `yyyy-MM-dd'T'HH:mm:ss.SSSXXX` | The format to use when writing the date to `target_field`. Must be a valid [java time pattern](/reference/elasticsearch/mapping-reference/mapping-date-format.md). |
 | `description` | no | - | Description of the processor. Useful for describing the purpose of the processor or its configuration. |
@@ -25,6 +24,20 @@ $$$date-options$$$
 | `on_failure` | no | - | Handle failures for the processor. See [Handling pipeline failures](docs-content://manage-data/ingest/transform-enrich/ingest-pipelines.md#handling-pipeline-failures). |
 | `tag` | no | - | Identifier for the processor. Useful for debugging and metrics. |
 
+## Timezones [date-processor-timezones]
+
+The `timezone` option may have two effects on the behavior of the processor:
+ - If the string being parsed matches a format representing a local date-time, such as `yyyy-MM-dd HH:mm:ss`, it will be assumed to be in the timezone specified by this option. This is not applicable if the string matches a format representing a zoned date-time, such as `yyyy-MM-dd HH:mm:ss zzz`: in that case, the timezone parsed from the string will be used. It is also not applicable if the string matches an absolute time format, such as `epoch_millis`.
+ - The date-time will be converted into the timezone given by this option before it is formatted and written into the target field. This is not applicable if the `output_format` is an absolute time format such as `epoch_millis`.
+
+::::{warning}
+We recommend avoiding the use of short abbreviations for timezone names, since they can be ambiguous. For example, one JDK might interpret `PST` as `America/Tijuana`, i.e. Pacific (Standard) Time, while another JDK might interpret it as `Asia/Manila`, i.e. Philippine Standard Time. If your input data contains such abbreviations, you should convert them into either standard full names or UTC offsets before parsing them, using your own knowledge of what each abbreviation means in your data. See [below](#date-processor-short-timezone-example) for an example. (This does not apply to `UTC`, which is safe.)
+::::
+
+## Examples [date-processor-examples]
+
+### Simple example [date-processor-simple-example]
+
 Here is an example that adds the parsed date to the `timestamp` field based on the `initial_date` field:
 
 ```js
@@ -43,6 +56,8 @@ Here is an example that adds the parsed date to the `timestamp` field based on t
 }
 ```
 
+### Example using templated parameters [date-processor-templated-example]
+
 The `timezone` and `locale` processor parameters are templated. This means that their values can be extracted from fields within documents. The example below shows how to extract the locale/timezone details from existing fields, `my_timezone` and `my_locale`, in the ingested document that contain the timezone and locale values.
 
 ```js
@@ -62,3 +77,49 @@ The `timezone` and `locale` processor parameters are templated. This means that
 }
 ```
 
+### Example dealing with short timezone abbreviations safely [date-processor-short-timezone-example]
+
+In the example below, the `message` field in the input is expected to be a string formed of a local date-time in `yyyyMMddHHmmss` format, a timezone abbreviated to one of `PST`, `CET`, or `JST` representing Pacific, Central European, or Japan time, and a payload. This field is split up using a `grok` processor, then the timezones are converted into full names using a `script` processor, then the date-time is parsed using a `date` processor, and finally the unwanted fields are discarded using a `drop` processor.
+
+```js
+{
+  "description" : "...",
+  "processors": [
+    {
+      "grok": {
+        "field": "message",
+        "patterns": ["%{DATESTAMP_EVENTLOG:local_date_time} %{TZ:short_tz} %{GREEDYDATA:payload}"],
+        "pattern_definitions": {
+          "TZ": "[A-Z]{3}"
+        }
+      }
+    },
+    {
+      "script": {
+        "source": "ctx['full_tz'] = params['tz_map'][ctx['short_tz']]",
+        "params": {
+          "tz_map": {
+            "PST": "America/Los_Angeles",
+            "CET": "Europe/Amsterdam",
+            "JST": "Asia/Tokyo"
+          }
+        }
+      }
+    },
+    {
+      "date": {
+        "field": "local_date_time",
+        "formats": ["yyyyMMddHHmmss"],
+        "timezone": "{{{full_tz}}}"
+      }
+    },
+    {
+      "remove": {
+        "field": ["message", "local_date_time", "short_tz", "full_tz"]
+      }
+    }
+  ]
+}
+```
+
+With this pipeline, a `message` field with the value `20250102123456 PST Hello world` will result in a `@timestamp` field with the value `2025-01-02T12:34:56.000-08:00` and a `payload` field with the value `Hello world`. (Note: A `@timestamp` field will normally be mapped to a `date` type, and therefore it will be indexed as an integer representing milliseconds since the epoch, although the original format and timezone may be preserved in the `_source`.)

libs/simdvec/src/main/java/org/elasticsearch/simdvec/internal/vectorization/DefaultESVectorizationProvider.java

@@ -9,6 +9,10 @@
 
 package org.elasticsearch.simdvec.internal.vectorization;
 
+import org.apache.lucene.store.IndexInput;
+
+import java.io.IOException;
+
 final class DefaultESVectorizationProvider extends ESVectorizationProvider {
     private final ESVectorUtilSupport vectorUtilSupport;
 
@@ -20,4 +24,9 @@ final class DefaultESVectorizationProvider extends ESVectorizationProvider {
     public ESVectorUtilSupport getVectorUtilSupport() {
         return vectorUtilSupport;
     }
+
+    @Override
+    public ES91OSQVectorsScorer newES91OSQVectorsScorer(IndexInput input, int dimension) throws IOException {
+        return new ES91OSQVectorsScorer(input, dimension);
+    }
 }