Tracing for deep search path

Signed-off-by: David Zane <[email protected]>
dzane17 · Feb 7, 2024 · 8a16fa1 · 8a16fa1
1 parent 52b27f4
commit 8a16fa1
Show file tree

Hide file tree

Showing 10 changed files with 246 additions and 41 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/libs/telemetry/src/main/java/org/opensearch/telemetry/tracing/SpanCreationContext.java b/libs/telemetry/src/main/java/org/opensearch/telemetry/tracing/SpanCreationContext.java
@@ -28,6 +28,11 @@ public final class SpanCreationContext {
      */
     private SpanCreationContext() {}
 
+    SpanCreationContext(String spanName, Attributes attributes) {
+        this.spanName = spanName;
+        this.attributes = attributes;
+    }
+
     /**
      * Sets the span type to server.
      * @return spanCreationContext

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);

diff --git a/server/src/main/java/org/opensearch/action/search/SearchRequestContext.java b/server/src/main/java/org/opensearch/action/search/SearchRequestContext.java
@@ -78,7 +78,7 @@ void setTotalHits(TotalHits totalHits) {
         this.totalHits = totalHits;
     }
 
-    TotalHits totalHits() {
+    public TotalHits totalHits() {
         return totalHits;
     }
 

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 +88,12 @@
 import org.opensearch.tasks.CancellableTask;
 import org.opensearch.tasks.Task;
 import org.opensearch.telemetry.metrics.MetricsRegistry;
+import org.opensearch.telemetry.tracing.Span;
+import org.opensearch.telemetry.tracing.SpanBuilder;
+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 +179,7 @@ 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 +202,8 @@ 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 +223,7 @@ public TransportSearchAction(
         this.searchRequestOperationsCompositeListenerFactory = searchRequestOperationsCompositeListenerFactory;
         clusterService.getClusterSettings()
             .addSettingsUpdateConsumer(SEARCH_QUERY_METRICS_ENABLED_SETTING, this::setSearchQueryMetricsEnabled);
+        this.tracer = tracer;
     }
 
     private void setSearchQueryMetricsEnabled(boolean searchQueryMetricsEnabled) {
@@ -431,49 +440,72 @@ private void executeRequest(
         if (originalSearchRequest.isPhaseTook() == null) {
             originalSearchRequest.setPhaseTook(clusterService.getClusterSettings().get(SEARCH_PHASE_TOOK_ENABLED));
         }
-        SearchRequestOperationsListener.CompositeListener requestOperationsListeners = searchRequestOperationsCompositeListenerFactory
-            .buildCompositeListener(originalSearchRequest, logger);
-        SearchRequestContext searchRequestContext = new SearchRequestContext(requestOperationsListeners, originalSearchRequest);
-        searchRequestContext.getSearchRequestOperationsListener().onRequestStart(searchRequestContext);
-
-        PipelinedRequest searchRequest;
-        ActionListener<SearchResponse> listener;
-        try {
-            searchRequest = searchPipelineService.resolvePipeline(originalSearchRequest);
-            listener = searchRequest.transformResponseListener(originalListener);
-        } catch (Exception e) {
-            originalListener.onFailure(e);
-            return;
-        }
-
-        ActionListener<SearchRequest> requestTransformListener = ActionListener.wrap(sr -> {
-            if (searchQueryMetricsEnabled) {
-                try {
-                    searchQueryCategorizer.categorize(sr.source());
-                } catch (Exception e) {
-                    logger.error("Error while trying to categorize the query.", e);
-                }
-            }
 
-            ActionListener<SearchSourceBuilder> rewriteListener = buildRewriteListener(
-                sr,
-                task,
-                timeProvider,
-                searchAsyncActionProvider,
-                listener,
-                searchRequestContext
+        final Span requestSpan = tracer.startSpan(SpanBuilder.from(task, actionName));
+        try (final SpanScope spanScope = tracer.withSpanInScope(requestSpan)) {
+            originalListener = TraceableActionListener.create(
+                originalListener,
+                requestSpan,
+                tracer
             );
-            if (sr.source() == null) {
-                rewriteListener.onResponse(sr.source());
+
+            SearchRequestOperationsListener.CompositeListener requestOperationsListeners;
+            if (tracer.isRecording()) {
+                TraceableSearchRequestOperationsListener traceableSearchRequestOperationsListener = new TraceableSearchRequestOperationsListener(tracer, requestSpan);
+                requestOperationsListeners = searchRequestOperationsCompositeListenerFactory.buildCompositeListener(
+                    originalSearchRequest,
+                    logger,
+                    traceableSearchRequestOperationsListener
+                );
             } else {
-                Rewriteable.rewriteAndFetch(
-                    sr.source(),
-                    searchService.getRewriteContext(timeProvider::getAbsoluteStartMillis),
-                    rewriteListener
+                requestOperationsListeners = searchRequestOperationsCompositeListenerFactory.buildCompositeListener(
+                    originalSearchRequest,
+                    logger
                 );
             }
-        }, listener::onFailure);
-        searchRequest.transformRequest(requestTransformListener);
+            SearchRequestContext searchRequestContext = new SearchRequestContext(requestOperationsListeners, originalSearchRequest);
+            searchRequestContext.getSearchRequestOperationsListener().onRequestStart(searchRequestContext);
+
+            PipelinedRequest searchRequest;
+            ActionListener<SearchResponse> listener;
+
+            try {
+                searchRequest = searchPipelineService.resolvePipeline(originalSearchRequest);
+                listener = searchRequest.transformResponseListener(originalListener);
+            } catch (Exception e) {
+                originalListener.onFailure(e);
+                return;
+            }
+
+            ActionListener<SearchRequest> requestTransformListener = ActionListener.wrap(sr -> {
+                if (searchQueryMetricsEnabled) {
+                    try {
+                        searchQueryCategorizer.categorize(sr.source());
+                    } catch (Exception e) {
+                        logger.error("Error while trying to categorize the query.", e);
+                    }
+                }
+
+                ActionListener<SearchSourceBuilder> rewriteListener = buildRewriteListener(
+                    sr,
+                    task,
+                    timeProvider,
+                    searchAsyncActionProvider,
+                    listener,
+                    searchRequestContext
+                );
+                if (sr.source() == null) {
+                    rewriteListener.onResponse(sr.source());
+                } else {
+                    Rewriteable.rewriteAndFetch(
+                        sr.source(),
+                        searchService.getRewriteContext(timeProvider::getAbsoluteStartMillis),
+                        rewriteListener
+                    );
+                }
+            }, listener::onFailure);
+            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
@@ -70,6 +70,16 @@ private AttributeNames() {
      */
     public static final String TRANSPORT_ACTION = "action";
 
+    /**
+     * Task id
+     */
+    public static final String TASK_ID = "task_id";
+
+    /**
+     * Parent task id
+     */
+    public static final String PARENT_TASK_ID = "p_task_id";
+
     /**
      * Index Name
      */
@@ -94,4 +104,19 @@ 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
@@ -14,6 +14,7 @@
 import org.opensearch.core.common.Strings;
 import org.opensearch.http.HttpRequest;
 import org.opensearch.rest.RestRequest;
+import org.opensearch.tasks.Task;
 import org.opensearch.telemetry.tracing.attributes.Attributes;
 import org.opensearch.transport.TcpChannel;
 import org.opensearch.transport.Transport;
@@ -42,6 +43,15 @@ 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 +180,43 @@ 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);
+    }
+
+    /**
+     * Creates {@link SpanCreationContext} with parent set to specified SpanContext.
+     * @param task search task.
+     * @param actionName action.
+     * @return context
+     */
+    public static SpanCreationContext from(Task task, String actionName) {
+        return new SpanCreationContext(createSpanName(task, actionName), buildSpanAttributes(task, actionName));
+    }
+
+    private static Attributes buildSpanAttributes(Task task, String actionName) {
+        Attributes attributes = Attributes.create().addAttribute(AttributeNames.TRANSPORT_ACTION, actionName);
+        if (task != null) {
+            attributes.addAttribute(AttributeNames.TASK_ID, task.getId());
+            if (task.getParentTaskId() != null && task.getParentTaskId().isSet()) {
+                attributes.addAttribute(AttributeNames.PARENT_TASK_ID, task.getParentTaskId().getId());
+            }
+        }
+        return attributes;
+
+    }
+
+    private static String createSpanName(Task task, String actionName) {
+        if (task != null) {
+            return task.getType() + SEPARATOR + task.getAction();
+        } else {
+            return actionName;
+        }
+    }
 }
diff --git a/...a/org/opensearch/telemetry/tracing/listener/TraceableSearchRequestOperationsListener.java b/...a/org/opensearch/telemetry/tracing/listener/TraceableSearchRequestOperationsListener.java
@@ -0,0 +1,72 @@
+/*
+ * 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 final Span requestSpan;
+    private Span phaseSpan;
+    private SpanScope phaseSpanScope;
+
+    public TraceableSearchRequestOperationsListener(final Tracer tracer, final Span requestSpan) {
+        this.tracer = tracer;
+        this.requestSpan = requestSpan;
+        this.phaseSpan = NoopSpan.INSTANCE;
+    }
+
+    @Override
+    protected void onPhaseStart(SearchPhaseContext context) {
+        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) {
+        phaseSpan.endSpan();
+        phaseSpanScope.close();
+    }
+
+    @Override
+    public void onRequestStart(SearchRequestContext searchRequestContext) {}
+
+    @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
+        );
+    }
+}
diff --git a/server/src/test/java/org/opensearch/snapshots/SnapshotResiliencyTests.java b/server/src/test/java/org/opensearch/snapshots/SnapshotResiliencyTests.java
@@ -2310,7 +2310,8 @@ public void onFailure(final Exception e) {
                             client
                         ),
                         NoopMetricsRegistry.INSTANCE,
-                        searchRequestOperationsCompositeListenerFactory
+                        searchRequestOperationsCompositeListenerFactory,
+                        NoopTracer.INSTANCE
                     )
                 );
                 actions.put(