Tracing for deep search path

Signed-off-by: David Zane <[email protected]>
opensearch-project · Feb 6, 2024 · 21ab882 · 21ab882
1 parent 52b27f4
commit 21ab882
Show file tree

Hide file tree

Showing 18 changed files with 273 additions and 42 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -16,6 +16,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 - Allow to pass the list settings through environment variables (like [], ["a", "b", "c"], ...) ([#10625](https://github.com/opensearch-project/OpenSearch/pull/10625))
 - [Admission Control] Integrate CPU AC with ResourceUsageCollector and add CPU AC stats to nodes/stats ([#10887](https://github.com/opensearch-project/OpenSearch/pull/10887))
 - [S3 Repository] Add setting to control connection count for sync client ([#12028](https://github.com/opensearch-project/OpenSearch/pull/12028))
+- Tracing for deep search path ([#12103](https://github.com/opensearch-project/OpenSearch/pull/12103))
 
 ### Dependencies
 - Bump `log4j-core` from 2.18.0 to 2.19.0

diff --git a/...hts/src/main/java/org/opensearch/plugin/insights/core/listener/QueryInsightsListener.java b/...hts/src/main/java/org/opensearch/plugin/insights/core/listener/QueryInsightsListener.java
@@ -110,13 +110,13 @@ public boolean isEnabled() {
     }
 
     @Override
-    public void onPhaseStart(SearchPhaseContext context) {}
+    public void onPhaseStart(SearchPhaseContext context, SearchRequestContext searchRequestContext) {}
 
     @Override
     public void onPhaseEnd(SearchPhaseContext context, SearchRequestContext searchRequestContext) {}
 
     @Override
-    public void onPhaseFailure(SearchPhaseContext context) {}
+    public void onPhaseFailure(SearchPhaseContext context, SearchRequestContext searchRequestContext) {}
 
     @Override
     public void onRequestStart(SearchRequestContext searchRequestContext) {}

diff --git a/server/src/main/java/org/opensearch/action/search/AbstractSearchAsyncAction.java b/server/src/main/java/org/opensearch/action/search/AbstractSearchAsyncAction.java
@@ -220,6 +220,7 @@ public final void start() {
                     null
                 )
             );
+            onRequestEnd(searchRequestContext);
             return;
         }
         executePhase(this);
@@ -439,10 +440,10 @@ private void onPhaseEnd(SearchRequestContext searchRequestContext) {
         }
     }
 
-    void onPhaseStart(SearchPhase phase) {
+    void onPhaseStart(SearchPhase phase, SearchRequestContext searchRequestContext) {
         setCurrentPhase(phase);
         if (SearchPhaseName.isValidName(phase.getName())) {
-            this.searchRequestContext.getSearchRequestOperationsListener().onPhaseStart(this);
+            this.searchRequestContext.getSearchRequestOperationsListener().onPhaseStart(this, searchRequestContext);
         }
     }
 
@@ -452,7 +453,7 @@ private void onRequestEnd(SearchRequestContext searchRequestContext) {
 
     private void executePhase(SearchPhase phase) {
         try {
-            onPhaseStart(phase);
+            onPhaseStart(phase, searchRequestContext);
             phase.recordAndRun();
         } catch (Exception e) {
             if (logger.isDebugEnabled()) {
@@ -717,7 +718,7 @@ public void sendSearchResponse(InternalSearchResponse internalSearchResponse, At
     @Override
     public final void onPhaseFailure(SearchPhase phase, String msg, Throwable cause) {
         if (SearchPhaseName.isValidName(phase.getName())) {
-            this.searchRequestContext.getSearchRequestOperationsListener().onPhaseFailure(this);
+            this.searchRequestContext.getSearchRequestOperationsListener().onPhaseFailure(this, searchRequestContext);
         }
         raisePhaseFailure(new SearchPhaseExecutionException(phase.getName(), msg, cause, buildShardFailures()));
     }

diff --git a/server/src/main/java/org/opensearch/action/search/SearchRequestContext.java b/server/src/main/java/org/opensearch/action/search/SearchRequestContext.java
@@ -23,20 +23,19 @@
  */
 @InternalApi
 public class SearchRequestContext {
+    private final SearchRequest searchRequest;
     private final SearchRequestOperationsListener searchRequestOperationsListener;
     private long absoluteStartNanos;
     private final Map<String, Long> phaseTookMap;
     private TotalHits totalHits;
     private final EnumMap<ShardStatsFieldNames, Integer> shardStats;
 
-    private final SearchRequest searchRequest;
-
-    SearchRequestContext(final SearchRequestOperationsListener searchRequestOperationsListener, final SearchRequest searchRequest) {
+    SearchRequestContext(SearchRequestOperationsListener searchRequestOperationsListener, SearchRequest searchRequest) {
         this.searchRequestOperationsListener = searchRequestOperationsListener;
+        this.searchRequest = searchRequest;
         this.absoluteStartNanos = System.nanoTime();
         this.phaseTookMap = new HashMap<>();
         this.shardStats = new EnumMap<>(ShardStatsFieldNames.class);
-        this.searchRequest = searchRequest;
     }
 
     SearchRequestOperationsListener getSearchRequestOperationsListener() {
@@ -78,7 +77,7 @@ void setTotalHits(TotalHits totalHits) {
         this.totalHits = totalHits;
     }
 
-    TotalHits totalHits() {
+    public TotalHits totalHits() {
         return totalHits;
     }
 
@@ -89,7 +88,7 @@ void setShardStats(int total, int successful, int skipped, int failed) {
         this.shardStats.put(ShardStatsFieldNames.SEARCH_REQUEST_SLOWLOG_SHARD_FAILED, failed);
     }
 
-    String formattedShardStats() {
+    public String formattedShardStats() {
         if (shardStats.isEmpty()) {
             return "";
         } else {

diff --git a/server/src/main/java/org/opensearch/action/search/SearchRequestOperationsListener.java b/server/src/main/java/org/opensearch/action/search/SearchRequestOperationsListener.java
@@ -31,11 +31,11 @@ protected SearchRequestOperationsListener(final boolean enabled) {
         this.enabled = enabled;
     }
 
-    protected abstract void onPhaseStart(SearchPhaseContext context);
+    protected abstract void onPhaseStart(SearchPhaseContext context, SearchRequestContext searchRequestContext);
 
     protected abstract void onPhaseEnd(SearchPhaseContext context, SearchRequestContext searchRequestContext);
 
-    protected abstract void onPhaseFailure(SearchPhaseContext context);
+    protected abstract void onPhaseFailure(SearchPhaseContext context, SearchRequestContext searchRequestContext);
 
     protected void onRequestStart(SearchRequestContext searchRequestContext) {}
 
@@ -69,10 +69,10 @@ static final class CompositeListener extends SearchRequestOperationsListener {
         }
 
         @Override
-        protected void onPhaseStart(SearchPhaseContext context) {
+        protected void onPhaseStart(SearchPhaseContext context, SearchRequestContext searchRequestContext) {
             for (SearchRequestOperationsListener listener : listeners) {
                 try {
-                    listener.onPhaseStart(context);
+                    listener.onPhaseStart(context, searchRequestContext);
                 } catch (Exception e) {
                     logger.warn(() -> new ParameterizedMessage("onPhaseStart listener [{}] failed", listener), e);
                 }
@@ -91,10 +91,10 @@ protected void onPhaseEnd(SearchPhaseContext context, SearchRequestContext searc
         }
 
         @Override
-        protected void onPhaseFailure(SearchPhaseContext context) {
+        protected void onPhaseFailure(SearchPhaseContext context, SearchRequestContext searchRequestContext) {
             for (SearchRequestOperationsListener listener : listeners) {
                 try {
-                    listener.onPhaseFailure(context);
+                    listener.onPhaseFailure(context, searchRequestContext);
                 } catch (Exception e) {
                     logger.warn(() -> new ParameterizedMessage("onPhaseFailure listener [{}] failed", listener), e);
                 }

diff --git a/server/src/main/java/org/opensearch/action/search/SearchRequestSlowLog.java b/server/src/main/java/org/opensearch/action/search/SearchRequestSlowLog.java
@@ -134,13 +134,13 @@ public SearchRequestSlowLog(ClusterService clusterService) {
     }
 
     @Override
-    protected void onPhaseStart(SearchPhaseContext context) {}
+    protected void onPhaseStart(SearchPhaseContext context, SearchRequestContext searchRequestContext) {}
 
     @Override
     protected void onPhaseEnd(SearchPhaseContext context, SearchRequestContext searchRequestContext) {}
 
     @Override
-    protected void onPhaseFailure(SearchPhaseContext context) {}
+    protected void onPhaseFailure(SearchPhaseContext context, SearchRequestContext searchRequestContext) {}
 
     @Override
     protected void onRequestStart(SearchRequestContext searchRequestContext) {}

diff --git a/server/src/main/java/org/opensearch/action/search/SearchRequestStats.java b/server/src/main/java/org/opensearch/action/search/SearchRequestStats.java
@@ -58,7 +58,7 @@ public long getPhaseMetric(SearchPhaseName searchPhaseName) {
     }
 
     @Override
-    protected void onPhaseStart(SearchPhaseContext context) {
+    protected void onPhaseStart(SearchPhaseContext context, SearchRequestContext searchRequestContext) {
         phaseStatsMap.get(context.getCurrentPhase().getSearchPhaseName()).current.inc();
     }
 
@@ -71,7 +71,7 @@ protected void onPhaseEnd(SearchPhaseContext context, SearchRequestContext searc
     }
 
     @Override
-    protected void onPhaseFailure(SearchPhaseContext context) {
+    protected void onPhaseFailure(SearchPhaseContext context, SearchRequestContext searchRequestContext) {
         phaseStatsMap.get(context.getCurrentPhase().getSearchPhaseName()).current.dec();
     }
 

diff --git a/server/src/main/java/org/opensearch/action/search/TransportSearchAction.java b/server/src/main/java/org/opensearch/action/search/TransportSearchAction.java
@@ -88,6 +88,10 @@
 import org.opensearch.tasks.CancellableTask;
 import org.opensearch.tasks.Task;
 import org.opensearch.telemetry.metrics.MetricsRegistry;
+import org.opensearch.telemetry.tracing.SpanScope;
+import org.opensearch.telemetry.tracing.Tracer;
+import org.opensearch.telemetry.tracing.listener.TraceableActionListener;
+import org.opensearch.telemetry.tracing.listener.TraceableSearchRequestOperationsListener;
 import org.opensearch.threadpool.ThreadPool;
 import org.opensearch.transport.RemoteClusterAware;
 import org.opensearch.transport.RemoteClusterService;
@@ -173,6 +177,7 @@ public class TransportSearchAction extends HandledTransportAction<SearchRequest,
     private final CircuitBreaker circuitBreaker;
     private final SearchPipelineService searchPipelineService;
     private final SearchRequestOperationsCompositeListenerFactory searchRequestOperationsCompositeListenerFactory;
+    private final Tracer tracer;
 
     private volatile boolean searchQueryMetricsEnabled;
 
@@ -195,7 +200,8 @@ public TransportSearchAction(
         NamedWriteableRegistry namedWriteableRegistry,
         SearchPipelineService searchPipelineService,
         MetricsRegistry metricsRegistry,
-        SearchRequestOperationsCompositeListenerFactory searchRequestOperationsCompositeListenerFactory
+        SearchRequestOperationsCompositeListenerFactory searchRequestOperationsCompositeListenerFactory,
+        Tracer tracer
     ) {
         super(SearchAction.NAME, transportService, actionFilters, (Writeable.Reader<SearchRequest>) SearchRequest::new);
         this.client = client;
@@ -215,6 +221,7 @@ public TransportSearchAction(
         this.searchRequestOperationsCompositeListenerFactory = searchRequestOperationsCompositeListenerFactory;
         clusterService.getClusterSettings()
             .addSettingsUpdateConsumer(SEARCH_QUERY_METRICS_ENABLED_SETTING, this::setSearchQueryMetricsEnabled);
+        this.tracer = tracer;
     }
 
     private void setSearchQueryMetricsEnabled(boolean searchQueryMetricsEnabled) {
@@ -431,13 +438,35 @@ private void executeRequest(
         if (originalSearchRequest.isPhaseTook() == null) {
             originalSearchRequest.setPhaseTook(clusterService.getClusterSettings().get(SEARCH_PHASE_TOOK_ENABLED));
         }
-        SearchRequestOperationsListener.CompositeListener requestOperationsListeners = searchRequestOperationsCompositeListenerFactory
-            .buildCompositeListener(originalSearchRequest, logger);
+        SearchRequestOperationsListener.CompositeListener requestOperationsListeners;
+        TraceableSearchRequestOperationsListener traceableSearchRequestOperationsListener = null;
+        if (tracer.isRecording()) {
+            traceableSearchRequestOperationsListener = new TraceableSearchRequestOperationsListener(tracer);
+            requestOperationsListeners = searchRequestOperationsCompositeListenerFactory.buildCompositeListener(
+                originalSearchRequest,
+                logger,
+                traceableSearchRequestOperationsListener
+            );
+        } else {
+            requestOperationsListeners = searchRequestOperationsCompositeListenerFactory.buildCompositeListener(
+                originalSearchRequest,
+                logger
+            );
+        }
         SearchRequestContext searchRequestContext = new SearchRequestContext(requestOperationsListeners, originalSearchRequest);
         searchRequestContext.getSearchRequestOperationsListener().onRequestStart(searchRequestContext);
 
         PipelinedRequest searchRequest;
         ActionListener<SearchResponse> listener;
+
+        if (tracer.isRecording() && traceableSearchRequestOperationsListener != null) {
+            originalListener = TraceableActionListener.create(
+                originalListener,
+                traceableSearchRequestOperationsListener.getRequestSpan(),
+                tracer
+            );
+        }
+
         try {
             searchRequest = searchPipelineService.resolvePipeline(originalSearchRequest);
             listener = searchRequest.transformResponseListener(originalListener);
@@ -473,7 +502,9 @@ private void executeRequest(
                 );
             }
         }, listener::onFailure);
-        searchRequest.transformRequest(requestTransformListener);
+        try (SpanScope spanScope = tracer.withSpanInScope(traceableSearchRequestOperationsListener.getRequestSpan())) {
+            searchRequest.transformRequest(requestTransformListener);
+        }
     }
 
     private ActionListener<SearchSourceBuilder> buildRewriteListener(

diff --git a/server/src/main/java/org/opensearch/telemetry/tracing/AttributeNames.java b/server/src/main/java/org/opensearch/telemetry/tracing/AttributeNames.java
@@ -94,4 +94,19 @@ private AttributeNames() {
      * Refresh Policy
      */
     public static final String REFRESH_POLICY = "refresh_policy";
+
+    /**
+     * Search Request Source
+     */
+    public static final String SOURCE = "source";
+
+    /**
+     * Search Response Shard Stats
+     */
+    public static final String SHARDS = "shards";
+
+    /**
+     * Search Response Total Hits
+     */
+    public static final String TOTAL_HITS = "total_hits";
 }
diff --git a/server/src/main/java/org/opensearch/telemetry/tracing/SpanBuilder.java b/server/src/main/java/org/opensearch/telemetry/tracing/SpanBuilder.java
@@ -42,6 +42,15 @@ private SpanBuilder() {
 
     }
 
+    /**
+     * Creates {@link SpanCreationContext} from String
+     * @param spanName String.
+     * @return context.
+     */
+    public static SpanCreationContext from(String spanName) {
+        return SpanCreationContext.server().name(spanName);
+    }
+
     /**
      * Creates {@link SpanCreationContext} from the {@link HttpRequest}
      * @param request Http request.
@@ -170,4 +179,13 @@ private static Attributes buildSpanAttributes(String nodeId, ReplicatedWriteRequ
         return attributes;
     }
 
+    /**
+     * Creates {@link SpanCreationContext} with parent set to specified SpanContext.
+     * @param spanName name of span.
+     * @param parentSpan target parent span.
+     * @return context
+     */
+    public static SpanCreationContext from(String spanName, SpanContext parentSpan) {
+        return SpanCreationContext.server().name(spanName).parent(parentSpan);
+    }
 }
diff --git a/...a/org/opensearch/telemetry/tracing/listener/TraceableSearchRequestOperationsListener.java b/...a/org/opensearch/telemetry/tracing/listener/TraceableSearchRequestOperationsListener.java
@@ -0,0 +1,82 @@
+/*
+ * SPDX-License-Identifier: Apache-2.0
+ *
+ * The OpenSearch Contributors require contributions made to
+ * this file be licensed under the Apache-2.0 license or a
+ * compatible open source license.
+ */
+
+package org.opensearch.telemetry.tracing.listener;
+
+import org.opensearch.action.search.SearchPhaseContext;
+import org.opensearch.action.search.SearchRequestContext;
+import org.opensearch.action.search.SearchRequestOperationsListener;
+import org.opensearch.telemetry.tracing.AttributeNames;
+import org.opensearch.telemetry.tracing.Span;
+import org.opensearch.telemetry.tracing.SpanBuilder;
+import org.opensearch.telemetry.tracing.SpanContext;
+import org.opensearch.telemetry.tracing.SpanScope;
+import org.opensearch.telemetry.tracing.Tracer;
+import org.opensearch.telemetry.tracing.noop.NoopSpan;
+
+import static org.opensearch.core.common.Strings.capitalize;
+
+/**
+ * SearchRequestOperationsListener subscriber for search request tracing
+ *
+ * @opensearch.internal
+ */
+public final class TraceableSearchRequestOperationsListener extends SearchRequestOperationsListener {
+    private final Tracer tracer;
+    private Span requestSpan;
+    private Span phaseSpan;
+    private SpanScope phaseSpanScope;
+
+    public TraceableSearchRequestOperationsListener(Tracer tracer) {
+        this.tracer = tracer;
+        this.requestSpan = NoopSpan.INSTANCE;
+        this.phaseSpan = NoopSpan.INSTANCE;
+    }
+
+    @Override
+    protected void onPhaseStart(SearchPhaseContext context, SearchRequestContext searchRequestContext) {
+        phaseSpan = tracer.startSpan(
+            SpanBuilder.from("coord" + capitalize(context.getCurrentPhase().getName()), new SpanContext(requestSpan))
+        );
+        phaseSpanScope = tracer.withSpanInScope(phaseSpan);
+    }
+
+    @Override
+    protected void onPhaseEnd(SearchPhaseContext context, SearchRequestContext searchRequestContext) {
+        phaseSpan.endSpan();
+        phaseSpanScope.close();
+    }
+
+    @Override
+    protected void onPhaseFailure(SearchPhaseContext context, SearchRequestContext searchRequestContext) {
+        phaseSpan.endSpan();
+        phaseSpanScope.close();
+    }
+
+    @Override
+    public void onRequestStart(SearchRequestContext searchRequestContext) {
+        requestSpan = tracer.startSpan(SpanBuilder.from("coordReq"));
+    }
+
+    @Override
+    public void onRequestEnd(SearchPhaseContext context, SearchRequestContext searchRequestContext) {
+        // add response-related attributes on request end
+        requestSpan.addAttribute(
+            AttributeNames.TOTAL_HITS,
+            searchRequestContext.totalHits() == null ? 0 : searchRequestContext.totalHits().value
+        );
+        requestSpan.addAttribute(
+            AttributeNames.SHARDS,
+            searchRequestContext.formattedShardStats().isEmpty() ? "null" : searchRequestContext.formattedShardStats()
+        );
+    }
+
+    public Span getRequestSpan() {
+        return requestSpan;
+    }
+}