Merge remote-tracking branch 'origin/main' into lucene_snapshot

Author: elasticsearchmachine

.buildkite/pipelines/periodic-fwc.template.yml

@@ -1,6 +1,6 @@
 steps:
-  - label: $FWC_VERSION / fwc
-    command: .ci/scripts/run-gradle.sh -Dbwc.checkout.align=true v$FWC_VERSION#fwcTest -Dtests.bwc.snapshot=false
+  - label: "{{matrix.FWC_VERSION}}" / fwc
+    command: .ci/scripts/run-gradle.sh -Dbwc.checkout.align=true v$$FWC_VERSION#fwcTest -Dtests.bwc.snapshot=false
     timeout_in_minutes: 300
     agents:
       provider: gcp
@@ -11,4 +11,4 @@ steps:
       setup:
         FWC_VERSION: $FWC_LIST
     env:
-      FWC_VERSION: $FWC_VERSION
+      FWC_VERSION: "{{matrix.FWC_VERSION}}"

.buildkite/pipelines/periodic-fwc.yml

@@ -1,7 +1,7 @@
 # This file is auto-generated. See .buildkite/pipelines/periodic-fwc.template.yml
 steps:
-  - label: $FWC_VERSION / fwc
-    command: .ci/scripts/run-gradle.sh -Dbwc.checkout.align=true v$FWC_VERSION#fwcTest -Dtests.bwc.snapshot=false
+  - label: "{{matrix.FWC_VERSION}}" / fwc
+    command: .ci/scripts/run-gradle.sh -Dbwc.checkout.align=true v$$FWC_VERSION#fwcTest -Dtests.bwc.snapshot=false
     timeout_in_minutes: 300
     agents:
       provider: gcp
@@ -12,4 +12,4 @@ steps:
       setup:
         FWC_VERSION: []
     env:
-      FWC_VERSION: $FWC_VERSION
+      FWC_VERSION: "{{matrix.FWC_VERSION}}"

.buildkite/scripts/fwc-branches.sh

@@ -0,0 +1,24 @@
+#!/bin/bash
+
+# Configure FwC test branches
+# We do not want 7.x branch and only to run for branches that:
+# - have released at least one minor version (not main)
+# - have previous minor unreleased (not the oldest development branch)
+FWC_BRANCHES=()
+for branch in "${BRANCHES[@]}"; do
+  if [[ ! "$branch" =~ ^7\..* ]]; then
+    FWC_BRANCHES+=("$branch")
+  fi
+done
+# Remove first and last element
+FWC_BRANCHES=("${FWC_BRANCHES[@]:1:${#FWC_BRANCHES[@]}-2}")
+
+shouldRunFwcFor() {
+  local branch=$1
+  for fwc_branch in "${FWC_BRANCHES[@]}"; do
+    if [[ "$fwc_branch" == "$branch" ]]; then
+      return 0
+    fi
+  done
+  return 1
+}

.buildkite/scripts/generate-pr-performance-benchmark.sh

@@ -2,39 +2,69 @@
 
 set -euo pipefail
 
+# uncomment for tests
+#function buildkite-agent {
+#  local command=$1
+#  echo "$@"
+#  if [ "$command" == "annotate" ]; then
+#    while read -r line; do
+#      echo "  read: $line";
+#    done
+#  fi
+#}
+
 env_id_baseline=$(python3 -c 'import uuid; print(uuid.uuid4())')
 env_id_contender=$(python3 -c 'import uuid; print(uuid.uuid4())')
 merge_base=$(git merge-base "${GITHUB_PR_TARGET_BRANCH}" HEAD)
 
+# PR comment
+buildkite-agent meta-data set pr_comment:early_comment_job_id "$BUILDKITE_JOB_ID"
 buildkite-agent meta-data set pr_comment:custom-body:body \
-  "This build attempted two ${GITHUB_PR_COMMENT_VAR_BENCHMARK} benchmarks to evaluate performance impact of this PR."
+  "This build attempts two ${GITHUB_PR_COMMENT_VAR_BENCHMARK} benchmarks to evaluate performance impact of this PR. \
+To estimate benchmark completion time inspect previous nightly runs [here](https://buildkite.com/elastic/elasticsearch-performance-esbench-nightly/builds?branch=master)."
 buildkite-agent meta-data set pr_comment:custom-baseline:head \
   "* Baseline: ${merge_base} (env ID ${env_id_baseline})"
 buildkite-agent meta-data set pr_comment:custom-contender:head \
   "* Contender: ${GITHUB_PR_TRIGGERED_SHA} (env ID ${env_id_contender})"
 
-cat << _EOF_
+# Buildkite annotation
+cat << _EOF1_ | buildkite-agent annotate --context "pr-benchmark-notification"
+  This build attempts two ${GITHUB_PR_COMMENT_VAR_BENCHMARK} benchmarks to evaluate performance impact of PR [${GITHUB_PR_NUMBER}](https://github.com/elastic/elasticsearch/pull/${GITHUB_PR_NUMBER}).
+  To estimate benchmark completion time inspect previous nightly runs [here](https://buildkite.com/elastic/elasticsearch-performance-esbench-nightly/builds?branch=master).
+  * Baseline: [${merge_base:0:7}](https://github.com/elastic/elasticsearch/commit/${merge_base}) (env ID ${env_id_baseline})
+  * Contender: [${GITHUB_PR_TRIGGERED_SHA:0:7}](https://github.com/elastic/elasticsearch/commit/${GITHUB_PR_TRIGGERED_SHA}) (env ID ${env_id_contender})
+_EOF1_
+
+cat << _EOF2_
 steps:
-  - label: Trigger baseline benchmark
+  - label: Trigger baseline benchmark with ${merge_base:0:7}
     trigger: elasticsearch-performance-esbench-pr
     build:
-      message: Baseline benchmark for PR${GITHUB_PR_NUMBER}
+      message: Baseline benchmark for PR ${GITHUB_PR_NUMBER} with ${merge_base:0:7}
       branch: master
       env:
         CONFIGURATION_NAME: ${GITHUB_PR_COMMENT_VAR_BENCHMARK}
         ENV_ID: ${env_id_baseline}
         REVISION: ${merge_base}
-  - label: Trigger contender benchmark
+  - label: Trigger contender benchmark with ${GITHUB_PR_TRIGGERED_SHA:0:7}
     trigger: elasticsearch-performance-esbench-pr
     build:
-      message: Contender benchmark for PR${GITHUB_PR_NUMBER}
+      message: Contender benchmark for PR ${GITHUB_PR_NUMBER} with ${GITHUB_PR_TRIGGERED_SHA:0:7}
       branch: master
       env:
         CONFIGURATION_NAME: ${GITHUB_PR_COMMENT_VAR_BENCHMARK}
         ENV_ID: ${env_id_contender}
         ES_REPO_URL: https://github.com/${GITHUB_PR_OWNER}/${GITHUB_PR_REPO}.git
         REVISION: ${GITHUB_PR_TRIGGERED_SHA}
   - wait: ~
-  - label: Modify PR comment
-    command: buildkite-agent meta-data set pr_comment:custom-comparison:head "* [Benchmark results](<https://esbench-metrics.kb.us-east-2.aws.elastic-cloud.com:9243/app/dashboards#/view/d9079962-5866-49ef-b9f5-145f2141cd31?_a=(query:(language:kuery,query:'user-tags.env-id:${env_id_baseline} or user-tags.env-id:${env_id_contender}'))>)"
-_EOF_
+  - label: Update PR comment and Buildkite annotation
+    command: |
+      buildkite-agent meta-data set pr_comment:custom-body:body "This build ran two ${GITHUB_PR_COMMENT_VAR_BENCHMARK} benchmarks to evaluate performance impact of this PR."
+      buildkite-agent meta-data set pr_comment:custom-comparison:head "* [Benchmark results](<https://esbench-metrics.kb.us-east-2.aws.elastic-cloud.com:9243/app/dashboards#/view/d9079962-5866-49ef-b9f5-145f2141cd31?_a=(query:(language:kuery,query:'user-tags.env-id:${env_id_baseline} or user-tags.env-id:${env_id_contender}'))>)"
+      cat << _EOF3_ | buildkite-agent annotate --context "pr-benchmark-notification"
+        This build ran two ${GITHUB_PR_COMMENT_VAR_BENCHMARK} benchmarks to evaluate performance impact of PR [${GITHUB_PR_NUMBER}](https://github.com/elastic/elasticsearch/pull/${GITHUB_PR_NUMBER}).
+        * Baseline: [${merge_base:0:7}](https://github.com/elastic/elasticsearch/commit/${merge_base}) (env ID ${env_id_baseline})
+        * Contender: [${GITHUB_PR_TRIGGERED_SHA:0:7}](https://github.com/elastic/elasticsearch/commit/${GITHUB_PR_TRIGGERED_SHA}) (env ID ${env_id_contender})
+        * [Benchmark results](<https://esbench-metrics.kb.us-east-2.aws.elastic-cloud.com:9243/app/dashboards#/view/d9079962-5866-49ef-b9f5-145f2141cd31?_a=(query:(language:kuery,query:'user-tags.env-id:${env_id_baseline} or user-tags.env-id:${env_id_contender}'))>)
+      _EOF3_
+_EOF2_

.buildkite/scripts/periodic.trigger.sh

@@ -5,6 +5,7 @@ set -euo pipefail
 echo "steps:"
 
 source .buildkite/scripts/branches.sh
+source .buildkite/scripts/fwc-branches.sh
 
 IS_FIRST=true
 SKIP_DELAY="${SKIP_DELAY:-false}"
@@ -46,8 +47,7 @@ EOF
       branch: "$BRANCH"
       commit: "$LAST_GOOD_COMMIT"
 EOF
-# Include forward compatibility tests only for the bugfix branch
-if [[ "${BRANCH}" == "${BRANCHES[2]}" ]]; then
+if shouldRunFwcFor "$BRANCH"; then
   cat <<EOF
   - trigger: elasticsearch-periodic-fwc
     label: Trigger periodic-fwc pipeline for $BRANCH

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

@@ -14,9 +14,10 @@
 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.store.NIOFSDirectory;
 import org.apache.lucene.util.quantization.OptimizedScalarQuantizer;
 import org.elasticsearch.common.logging.LogConfigurator;
+import org.elasticsearch.core.IOUtils;
 import org.elasticsearch.simdvec.ES91OSQVectorsScorer;
 import org.elasticsearch.simdvec.internal.vectorization.ESVectorizationProvider;
 import org.openjdk.jmh.annotations.Benchmark;
@@ -29,6 +30,7 @@
 import org.openjdk.jmh.annotations.Scope;
 import org.openjdk.jmh.annotations.Setup;
 import org.openjdk.jmh.annotations.State;
+import org.openjdk.jmh.annotations.TearDown;
 import org.openjdk.jmh.annotations.Warmup;
 import org.openjdk.jmh.infra.Blackhole;
 
@@ -66,9 +68,14 @@ public class OSQScorerBenchmark {
     float centroidDp;
 
     byte[] scratch;
-    ES91OSQVectorsScorer scorer;
+    ES91OSQVectorsScorer scorerMmap;
+    ES91OSQVectorsScorer scorerNfios;
 
-    IndexInput in;
+    Directory dirMmap;
+    IndexInput inMmap;
+
+    Directory dirNiofs;
+    IndexInput inNiofs;
 
     float[] scratchScores;
     float[] corrections;
@@ -84,18 +91,24 @@ public void setup() throws IOException {
             random.nextBytes(binaryVector);
         }
 
-        Directory dir = new MMapDirectory(Files.createTempDirectory("vectorData"));
-        IndexOutput out = dir.createOutput("vectors", IOContext.DEFAULT);
+        dirMmap = new MMapDirectory(Files.createTempDirectory("vectorDataMmap"));
+        dirNiofs = new NIOFSDirectory(Files.createTempDirectory("vectorDataNFIOS"));
+        IndexOutput outMmap = dirMmap.createOutput("vectors", IOContext.DEFAULT);
+        IndexOutput outNfios = dirNiofs.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);
+                outMmap.writeBytes(binaryVectors[i + j], 0, binaryVectors[i + j].length);
+                outNfios.writeBytes(binaryVectors[i + j], 0, binaryVectors[i + j].length);
             }
             random.nextBytes(correctionBytes);
-            out.writeBytes(correctionBytes, 0, correctionBytes.length);
+            outMmap.writeBytes(correctionBytes, 0, correctionBytes.length);
+            outNfios.writeBytes(correctionBytes, 0, correctionBytes.length);
         }
-        out.close();
-        in = dir.openInput("vectors", IOContext.DEFAULT);
+        outMmap.close();
+        outNfios.close();
+        inMmap = dirMmap.openInput("vectors", IOContext.DEFAULT);
+        inNiofs = dirNiofs.openInput("vectors", IOContext.DEFAULT);
 
         binaryQueries = new byte[numVectors][4 * length];
         for (byte[] binaryVector : binaryVectors) {
@@ -110,42 +123,40 @@ public void setup() throws IOException {
         centroidDp = random.nextFloat();
 
         scratch = new byte[length];
-        scorer = ESVectorizationProvider.getInstance().newES91OSQVectorsScorer(in, dims);
+        scorerMmap = ESVectorizationProvider.getInstance().newES91OSQVectorsScorer(inMmap, dims);
+        scorerNfios = ESVectorizationProvider.getInstance().newES91OSQVectorsScorer(inNiofs, dims);
         scratchScores = new float[16];
         corrections = new float[3];
     }
 
+    @TearDown
+    public void teardown() throws IOException {
+        IOUtils.close(dirMmap, inMmap, dirNiofs, inNiofs);
+    }
+
+    @Benchmark
+    public void scoreFromMemorySegmentOnlyVectorMmapScalar(Blackhole bh) throws IOException {
+        scoreFromMemorySegmentOnlyVector(bh, inMmap, scorerMmap);
+    }
+
     @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.lowerInterval(),
-                    result.upperInterval(),
-                    result.quantizedComponentSum(),
-                    result.additionalCorrection(),
-                    VectorSimilarityFunction.EUCLIDEAN,
-                    centroidDp,
-                    corrections[0],
-                    corrections[1],
-                    addition,
-                    corrections[2],
-                    qDist
-                );
-                bh.consume(score);
-            }
-        }
+    public void scoreFromMemorySegmentOnlyVectorMmapVect(Blackhole bh) throws IOException {
+        scoreFromMemorySegmentOnlyVector(bh, inMmap, scorerMmap);
+    }
+
+    @Benchmark
+    public void scoreFromMemorySegmentOnlyVectorNiofsScalar(Blackhole bh) throws IOException {
+        scoreFromMemorySegmentOnlyVector(bh, inNiofs, scorerNfios);
     }
 
     @Benchmark
     @Fork(jvmArgsPrepend = { "--add-modules=jdk.incubator.vector" })
-    public void scoreFromMemorySegmentOnlyVector(Blackhole bh) throws IOException {
+    public void scoreFromMemorySegmentOnlyVectorNiofsVect(Blackhole bh) throws IOException {
+        scoreFromMemorySegmentOnlyVector(bh, inNiofs, scorerNfios);
+    }
+
+    private void scoreFromMemorySegmentOnlyVector(Blackhole bh, IndexInput in, ES91OSQVectorsScorer scorer) throws IOException {
         for (int j = 0; j < numQueries; j++) {
             in.seek(0);
             for (int i = 0; i < numVectors; i++) {
@@ -170,9 +181,29 @@ public void scoreFromMemorySegmentOnlyVector(Blackhole bh) throws IOException {
         }
     }
 
+    @Benchmark
+    public void scoreFromMemorySegmentOnlyVectorBulkMmapScalar(Blackhole bh) throws IOException {
+        scoreFromMemorySegmentOnlyVectorBulk(bh, inMmap, scorerMmap);
+    }
+
+    @Benchmark
+    @Fork(jvmArgsPrepend = { "--add-modules=jdk.incubator.vector" })
+    public void scoreFromMemorySegmentOnlyVectorBulkMmapVect(Blackhole bh) throws IOException {
+        scoreFromMemorySegmentOnlyVectorBulk(bh, inMmap, scorerMmap);
+    }
+
+    @Benchmark
+    public void scoreFromMemorySegmentOnlyVectorBulkNiofsScalar(Blackhole bh) throws IOException {
+        scoreFromMemorySegmentOnlyVectorBulk(bh, inNiofs, scorerNfios);
+    }
+
     @Benchmark
     @Fork(jvmArgsPrepend = { "--add-modules=jdk.incubator.vector" })
-    public void scoreFromMemorySegmentOnlyVectorBulk(Blackhole bh) throws IOException {
+    public void scoreFromMemorySegmentOnlyVectorBulkNiofsVect(Blackhole bh) throws IOException {
+        scoreFromMemorySegmentOnlyVectorBulk(bh, inNiofs, scorerNfios);
+    }
+
+    private void scoreFromMemorySegmentOnlyVectorBulk(Blackhole bh, IndexInput in, ES91OSQVectorsScorer scorer) throws IOException {
         for (int j = 0; j < numQueries; j++) {
             in.seek(0);
             for (int i = 0; i < numVectors; i += 16) {
@@ -199,9 +230,29 @@ public void scoreFromMemorySegmentOnlyVectorBulk(Blackhole bh) throws IOExceptio
         }
     }
 
+    @Benchmark
+    public void scoreFromMemorySegmentAllBulkMmapScalar(Blackhole bh) throws IOException {
+        scoreFromMemorySegmentAllBulk(bh, inMmap, scorerMmap);
+    }
+
+    @Benchmark
+    @Fork(jvmArgsPrepend = { "--add-modules=jdk.incubator.vector" })
+    public void scoreFromMemorySegmentAllBulkMmapVect(Blackhole bh) throws IOException {
+        scoreFromMemorySegmentAllBulk(bh, inMmap, scorerMmap);
+    }
+
+    @Benchmark
+    public void scoreFromMemorySegmentAllBulkNiofsScalar(Blackhole bh) throws IOException {
+        scoreFromMemorySegmentAllBulk(bh, inNiofs, scorerNfios);
+    }
+
     @Benchmark
     @Fork(jvmArgsPrepend = { "--add-modules=jdk.incubator.vector" })
-    public void scoreFromMemorySegmentAllBulk(Blackhole bh) throws IOException {
+    public void scoreFromMemorySegmentAllBulkNiofsVect(Blackhole bh) throws IOException {
+        scoreFromMemorySegmentAllBulk(bh, inNiofs, scorerNfios);
+    }
+
+    private void scoreFromMemorySegmentAllBulk(Blackhole bh, IndexInput in, ES91OSQVectorsScorer scorer) throws IOException {
         for (int j = 0; j < numQueries; j++) {
             in.seek(0);
             for (int i = 0; i < numVectors; i += 16) {

branches.json

@@ -1,23 +1,29 @@
 {
-  "notice": "This file is not maintained outside of the main branch and should only be used for tooling.",
-  "branches": [
+  "notice" : "This file is not maintained outside of the main branch and should only be used for tooling.",
+  "branches" : [
     {
-      "branch": "main"
+      "branch" : "main",
+      "version" : "9.2.0"
     },
     {
-      "branch": "9.1"
+      "branch" : "9.1",
+      "version" : "9.1.4"
     },
     {
-      "branch": "9.0"
+      "branch" : "9.0",
+      "version" : "9.0.7"
     },
     {
-      "branch": "8.19"
+      "branch" : "8.19",
+      "version" : "8.19.4"
     },
     {
-      "branch": "8.18"
+      "branch" : "8.18",
+      "version" : "8.18.7"
     },
     {
-      "branch": "7.17"
+      "branch" : "7.17",
+      "version" : "7.17.30"
     }
   ]
 }