Add support for aggregation profiler with concurrent aggregation (#8801)

* Add support for aggregation profiler with concurrent aggregation (#8801) Signed-off-by: Ticheng Lin <ticheng@amazon.com> * Address review comments for support for aggregation profiler with concurrent aggregation (#8801) Signed-off-by: Ticheng Lin <ticheng@amazon.com> * Refactor ProfileResult class and add more tests Signed-off-by: Ticheng Lin <ticheng@amazon.com> * Fix flaky QueryProfilePhaseTests.testCollapseQuerySearchResults test Signed-off-by: Ticheng Lin <ticheng@amazon.com> --------- Signed-off-by: Ticheng Lin <ticheng@amazon.com>
opensearch-project · Aug 1, 2023 · a9af5fe · a9af5fe
1 parent 6e47095
commit a9af5fe
Show file tree

Hide file tree

Showing 18 changed files with 800 additions and 68 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -40,6 +40,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 - Add safeguard limits for file cache during node level allocation ([#8208](https://github.com/opensearch-project/OpenSearch/pull/8208))
 - Add support for aggregation profiler with concurrent aggregation ([#8801](https://github.com/opensearch-project/OpenSearch/pull/8801))
 - [Remove] Deprecated Fractional ByteSizeValue support #9005 ([#9005](https://github.com/opensearch-project/OpenSearch/pull/9005))
+- Add support for aggregation profiler with concurrent aggregation ([#8801](https://github.com/opensearch-project/OpenSearch/pull/8801))
 
 ### Deprecated
 

diff --git a/...rnalClusterTest/java/org/opensearch/search/profile/aggregation/AggregationProfilerIT.java b/...rnalClusterTest/java/org/opensearch/search/profile/aggregation/AggregationProfilerIT.java
diff --git a/server/src/main/java/org/opensearch/search/SearchService.java b/server/src/main/java/org/opensearch/search/SearchService.java
@@ -1271,7 +1271,7 @@ private void parseSource(DefaultSearchContext context, SearchSourceBuilder sourc
             context.minimumScore(source.minScore());
         }
         if (source.profile()) {
-            context.setProfilers(new Profilers(context.searcher()));
+            context.setProfilers(new Profilers(context.searcher(), context.isConcurrentSegmentSearchEnabled()));
         }
         if (source.timeout() != null) {
             context.timeout(source.timeout());

diff --git a/server/src/main/java/org/opensearch/search/profile/AbstractProfileBreakdown.java b/server/src/main/java/org/opensearch/search/profile/AbstractProfileBreakdown.java
@@ -50,8 +50,10 @@ public abstract class AbstractProfileBreakdown<T extends Enum<T>> {
     /**
      * The accumulated timings for this query node
      */
-    private final Timer[] timings;
-    private final T[] timingTypes;
+    protected final Timer[] timings;
+    protected final T[] timingTypes;
+    public static final String TIMING_TYPE_COUNT_SUFFIX = "_count";
+    public static final String TIMING_TYPE_START_TIME_SUFFIX = "_start_time";
 
     /** Sole constructor. */
     public AbstractProfileBreakdown(Class<T> clazz) {
@@ -74,17 +76,10 @@ public void setTimer(T timing, Timer timer) {
      * Build a timing count breakdown for current instance
      */
     public Map<String, Long> toBreakdownMap() {
-        return buildBreakdownMap(this);
-    }
-
-    /**
-     * Build a timing count breakdown for arbitrary instance
-     */
-    protected final Map<String, Long> buildBreakdownMap(AbstractProfileBreakdown<T> breakdown) {
-        Map<String, Long> map = new HashMap<>(breakdown.timings.length * 2);
-        for (T timingType : breakdown.timingTypes) {
-            map.put(timingType.toString(), breakdown.timings[timingType.ordinal()].getApproximateTiming());
-            map.put(timingType.toString() + "_count", breakdown.timings[timingType.ordinal()].getCount());
+        Map<String, Long> map = new HashMap<>(this.timings.length * 3);
+        for (T timingType : this.timingTypes) {
+            map.put(timingType.toString(), this.timings[timingType.ordinal()].getApproximateTiming());
+            map.put(timingType + TIMING_TYPE_COUNT_SUFFIX, this.timings[timingType.ordinal()].getCount());
         }
         return Collections.unmodifiableMap(map);
     }

diff --git a/server/src/main/java/org/opensearch/search/profile/ProfileResult.java b/server/src/main/java/org/opensearch/search/profile/ProfileResult.java
@@ -33,6 +33,7 @@
 package org.opensearch.search.profile;
 
 import org.opensearch.LegacyESVersion;
+import org.opensearch.Version;
 import org.opensearch.core.ParseField;
 import org.opensearch.core.common.io.stream.StreamInput;
 import org.opensearch.core.common.io.stream.StreamOutput;
@@ -45,8 +46,10 @@
 
 import java.io.IOException;
 import java.util.Collections;
+import java.util.Iterator;
 import java.util.List;
 import java.util.Map;
+import java.util.LinkedHashMap;
 import java.util.Objects;
 import java.util.concurrent.TimeUnit;
 
@@ -70,14 +73,23 @@ public final class ProfileResult implements Writeable, ToXContentObject {
     static final ParseField BREAKDOWN = new ParseField("breakdown");
     static final ParseField DEBUG = new ParseField("debug");
     static final ParseField NODE_TIME = new ParseField("time");
+    static final ParseField MAX_SLICE_NODE_TIME = new ParseField("max_slice_time");
+    static final ParseField MIN_SLICE_NODE_TIME = new ParseField("min_slice_time");
+    static final ParseField AVG_SLICE_NODE_TIME = new ParseField("avg_slice_time");
     static final ParseField NODE_TIME_RAW = new ParseField("time_in_nanos");
+    static final ParseField MAX_SLICE_NODE_TIME_RAW = new ParseField("max_slice_time_in_nanos");
+    static final ParseField MIN_SLICE_NODE_TIME_RAW = new ParseField("min_slice_time_in_nanos");
+    static final ParseField AVG_SLICE_NODE_TIME_RAW = new ParseField("avg_slice_time_in_nanos");
     static final ParseField CHILDREN = new ParseField("children");
 
     private final String type;
     private final String description;
     private final Map<String, Long> breakdown;
     private final Map<String, Object> debug;
     private final long nodeTime;
+    private Long maxSliceNodeTime;
+    private Long minSliceNodeTime;
+    private Long avgSliceNodeTime;
     private final List<ProfileResult> children;
 
     public ProfileResult(
@@ -87,13 +99,30 @@ public ProfileResult(
         Map<String, Object> debug,
         long nodeTime,
         List<ProfileResult> children
+    ) {
+        this(type, description, breakdown, debug, nodeTime, children, null, null, null);
+    }
+
+    public ProfileResult(
+        String type,
+        String description,
+        Map<String, Long> breakdown,
+        Map<String, Object> debug,
+        long nodeTime,
+        List<ProfileResult> children,
+        Long maxSliceNodeTime,
+        Long minSliceNodeTime,
+        Long avgSliceNodeTime
     ) {
         this.type = type;
         this.description = description;
         this.breakdown = Objects.requireNonNull(breakdown, "required breakdown argument missing");
         this.debug = debug == null ? Map.of() : debug;
         this.children = children == null ? List.of() : children;
         this.nodeTime = nodeTime;
+        this.maxSliceNodeTime = maxSliceNodeTime;
+        this.minSliceNodeTime = minSliceNodeTime;
+        this.avgSliceNodeTime = avgSliceNodeTime;
     }
 
     /**
@@ -110,6 +139,15 @@ public ProfileResult(StreamInput in) throws IOException {
             debug = Map.of();
         }
         children = in.readList(ProfileResult::new);
+        if (in.getVersion().onOrAfter(Version.V_2_10_0)) {
+            this.maxSliceNodeTime = in.readOptionalLong();
+            this.minSliceNodeTime = in.readOptionalLong();
+            this.avgSliceNodeTime = in.readOptionalLong();
+        } else {
+            this.maxSliceNodeTime = null;
+            this.minSliceNodeTime = null;
+            this.avgSliceNodeTime = null;
+        }
     }
 
     @Override
@@ -122,6 +160,11 @@ public void writeTo(StreamOutput out) throws IOException {
             out.writeMap(debug, StreamOutput::writeString, StreamOutput::writeGenericValue);
         }
         out.writeList(children);
+        if (out.getVersion().onOrAfter(Version.V_2_10_0)) {
+            out.writeOptionalLong(maxSliceNodeTime);
+            out.writeOptionalLong(minSliceNodeTime);
+            out.writeOptionalLong(avgSliceNodeTime);
+        }
     }
 
     /**
@@ -161,6 +204,18 @@ public long getTime() {
         return nodeTime;
     }
 
+    public Long getMaxSliceTime() {
+        return maxSliceNodeTime;
+    }
+
+    public Long getMinSliceTime() {
+        return minSliceNodeTime;
+    }
+
+    public Long getAvgSliceTime() {
+        return avgSliceNodeTime;
+    }
+
     /**
      * Returns a list of all profiled children queries
      */
@@ -175,9 +230,27 @@ public XContentBuilder toXContent(XContentBuilder builder, Params params) throws
         builder.field(DESCRIPTION.getPreferredName(), description);
         if (builder.humanReadable()) {
             builder.field(NODE_TIME.getPreferredName(), new TimeValue(getTime(), TimeUnit.NANOSECONDS).toString());
+            if (getMaxSliceTime() != null) {
+                builder.field(MAX_SLICE_NODE_TIME.getPreferredName(), new TimeValue(getMaxSliceTime(), TimeUnit.NANOSECONDS).toString());
+            }
+            if (getMinSliceTime() != null) {
+                builder.field(MIN_SLICE_NODE_TIME.getPreferredName(), new TimeValue(getMinSliceTime(), TimeUnit.NANOSECONDS).toString());
+            }
+            if (getAvgSliceTime() != null) {
+                builder.field(AVG_SLICE_NODE_TIME.getPreferredName(), new TimeValue(getAvgSliceTime(), TimeUnit.NANOSECONDS).toString());
+            }
         }
         builder.field(NODE_TIME_RAW.getPreferredName(), getTime());
-        builder.field(BREAKDOWN.getPreferredName(), breakdown);
+        if (getMaxSliceTime() != null) {
+            builder.field(MAX_SLICE_NODE_TIME_RAW.getPreferredName(), getMaxSliceTime());
+        }
+        if (getMinSliceTime() != null) {
+            builder.field(MIN_SLICE_NODE_TIME_RAW.getPreferredName(), getMinSliceTime());
+        }
+        if (getAvgSliceTime() != null) {
+            builder.field(AVG_SLICE_NODE_TIME_RAW.getPreferredName(), getAvgSliceTime());
+        }
+        createBreakdownView(builder);
         if (false == debug.isEmpty()) {
             builder.field(DEBUG.getPreferredName(), debug);
         }
@@ -193,6 +266,22 @@ public XContentBuilder toXContent(XContentBuilder builder, Params params) throws
         return builder.endObject();
     }
 
+    private void createBreakdownView(XContentBuilder builder) throws IOException {
+        Map<String, Long> modifiedBreakdown = new LinkedHashMap<>(breakdown);
+        removeStartTimeFields(modifiedBreakdown);
+        builder.field(BREAKDOWN.getPreferredName(), modifiedBreakdown);
+    }
+
+    static void removeStartTimeFields(Map<String, Long> modifiedBreakdown) {
+        Iterator<Map.Entry<String, Long>> iterator = modifiedBreakdown.entrySet().iterator();
+        while (iterator.hasNext()) {
+            Map.Entry<String, Long> entry = iterator.next();
+            if (entry.getKey().endsWith(AbstractProfileBreakdown.TIMING_TYPE_START_TIME_SUFFIX)) {
+                iterator.remove();
+            }
+        }
+    }
+
     private static final InstantiatingObjectParser<ProfileResult, Void> PARSER;
     static {
         InstantiatingObjectParser.Builder<ProfileResult, Void> parser = InstantiatingObjectParser.builder(
@@ -206,6 +295,9 @@ public XContentBuilder toXContent(XContentBuilder builder, Params params) throws
         parser.declareObject(optionalConstructorArg(), (p, c) -> p.map(), DEBUG);
         parser.declareLong(constructorArg(), NODE_TIME_RAW);
         parser.declareObjectArray(optionalConstructorArg(), (p, c) -> fromXContent(p), CHILDREN);
+        parser.declareLong(optionalConstructorArg(), MAX_SLICE_NODE_TIME_RAW);
+        parser.declareLong(optionalConstructorArg(), MIN_SLICE_NODE_TIME_RAW);
+        parser.declareLong(optionalConstructorArg(), AVG_SLICE_NODE_TIME_RAW);
         PARSER = parser.build();
     }
 

diff --git a/server/src/main/java/org/opensearch/search/profile/Profilers.java b/server/src/main/java/org/opensearch/search/profile/Profilers.java
@@ -34,6 +34,7 @@
 
 import org.opensearch.search.internal.ContextIndexSearcher;
 import org.opensearch.search.profile.aggregation.AggregationProfiler;
+import org.opensearch.search.profile.aggregation.ConcurrentAggregationProfiler;
 import org.opensearch.search.profile.query.QueryProfiler;
 
 import java.util.ArrayList;
@@ -50,18 +51,20 @@ public final class Profilers {
     private final ContextIndexSearcher searcher;
     private final List<QueryProfiler> queryProfilers;
     private final AggregationProfiler aggProfiler;
+    private final boolean isConcurrentSegmentSearchEnabled;
 
     /** Sole constructor. This {@link Profilers} instance will initially wrap one {@link QueryProfiler}. */
-    public Profilers(ContextIndexSearcher searcher) {
+    public Profilers(ContextIndexSearcher searcher, boolean isConcurrentSegmentSearchEnabled) {
         this.searcher = searcher;
+        this.isConcurrentSegmentSearchEnabled = isConcurrentSegmentSearchEnabled;
         this.queryProfilers = new ArrayList<>();
-        this.aggProfiler = new AggregationProfiler();
+        this.aggProfiler = isConcurrentSegmentSearchEnabled ? new ConcurrentAggregationProfiler() : new AggregationProfiler();
         addQueryProfiler();
     }
 
     /** Switch to a new profile. */
     public QueryProfiler addQueryProfiler() {
-        QueryProfiler profiler = new QueryProfiler(searcher.getExecutor() != null);
+        QueryProfiler profiler = new QueryProfiler(isConcurrentSegmentSearchEnabled);
         searcher.setProfiler(profiler);
         queryProfilers.add(profiler);
         return profiler;

diff --git a/server/src/main/java/org/opensearch/search/profile/Timer.java b/server/src/main/java/org/opensearch/search/profile/Timer.java
@@ -51,7 +51,7 @@
 public class Timer {
 
     private boolean doTiming;
-    private long timing, count, lastCount, start;
+    private long timing, count, lastCount, start, earliestTimerStartTime;
 
     /** pkg-private for testing */
     long nanoTime() {
@@ -71,6 +71,9 @@ public final void start() {
         doTiming = (count - lastCount) >= Math.min(lastCount >>> 8, 1024);
         if (doTiming) {
             start = nanoTime();
+            if (count == 0) {
+                earliestTimerStartTime = start;
+            }
         }
         count++;
     }
@@ -92,6 +95,14 @@ public final long getCount() {
         return count;
     }
 
+    /** Return the timer start time in nanoseconds.*/
+    public final long getEarliestTimerStartTime() {
+        if (start != 0) {
+            throw new IllegalStateException("#start call misses a matching #stop call");
+        }
+        return earliestTimerStartTime;
+    }
+
     /** Return an approximation of the total time spent between consecutive calls of #start and #stop. */
     public final long getApproximateTiming() {
         if (start != 0) {

diff --git a/.../src/main/java/org/opensearch/search/profile/aggregation/AggregationProfileBreakdown.java b/.../src/main/java/org/opensearch/search/profile/aggregation/AggregationProfileBreakdown.java
@@ -34,6 +34,7 @@
 
 import org.opensearch.search.profile.AbstractProfileBreakdown;
 
+import java.util.Collections;
 import java.util.HashMap;
 import java.util.Map;
 
@@ -62,4 +63,18 @@ public void addDebugInfo(String key, Object value) {
     protected Map<String, Object> toDebugMap() {
         return unmodifiableMap(extra);
     }
+
+    /**
+     * Build a timing count startTime breakdown for aggregation timing types
+     */
+    @Override
+    public Map<String, Long> toBreakdownMap() {
+        Map<String, Long> map = new HashMap<>(timings.length * 3);
+        for (AggregationTimingType timingType : timingTypes) {
+            map.put(timingType.toString(), timings[timingType.ordinal()].getApproximateTiming());
+            map.put(timingType + TIMING_TYPE_COUNT_SUFFIX, timings[timingType.ordinal()].getCount());
+            map.put(timingType + TIMING_TYPE_START_TIME_SUFFIX, timings[timingType.ordinal()].getEarliestTimerStartTime());
+        }
+        return Collections.unmodifiableMap(map);
+    }
 }
diff --git a/server/src/main/java/org/opensearch/search/profile/aggregation/AggregationProfiler.java b/server/src/main/java/org/opensearch/search/profile/aggregation/AggregationProfiler.java
@@ -36,8 +36,6 @@
 import org.opensearch.search.profile.AbstractProfiler;
 
 import java.util.HashMap;
-import java.util.LinkedList;
-import java.util.List;
 import java.util.Map;
 
 /**
@@ -47,29 +45,25 @@
  */
 public class AggregationProfiler extends AbstractProfiler<AggregationProfileBreakdown, Aggregator> {
 
-    private final Map<List<String>, AggregationProfileBreakdown> profileBreakdownLookup = new HashMap<>();
+    private final Map<Aggregator, AggregationProfileBreakdown> profileBreakdownLookup = new HashMap<>();
 
     public AggregationProfiler() {
         super(new InternalAggregationProfileTree());
     }
 
+    /**
+     * This method does not need to be thread safe for concurrent search use case as well.
+     * The {@link AggregationProfileBreakdown} for each Aggregation operator is created in sync path when
+     * {@link org.opensearch.search.aggregations.BucketCollector#preCollection()} is called
+     * on the Aggregation collector instances during construction.
+     */
     @Override
     public AggregationProfileBreakdown getQueryBreakdown(Aggregator agg) {
-        List<String> path = getAggregatorPath(agg);
-        AggregationProfileBreakdown aggregationProfileBreakdown = profileBreakdownLookup.get(path);
+        AggregationProfileBreakdown aggregationProfileBreakdown = profileBreakdownLookup.get(agg);
         if (aggregationProfileBreakdown == null) {
             aggregationProfileBreakdown = super.getQueryBreakdown(agg);
-            profileBreakdownLookup.put(path, aggregationProfileBreakdown);
+            profileBreakdownLookup.put(agg, aggregationProfileBreakdown);
         }
         return aggregationProfileBreakdown;
     }
-
-    public static List<String> getAggregatorPath(Aggregator agg) {
-        LinkedList<String> path = new LinkedList<>();
-        while (agg != null) {
-            path.addFirst(agg.name());
-            agg = agg.parent();
-        }
-        return path;
-    }
 }