Introduce BytesTransportResponse to enable zero-copy network writes

original-brownbear · original-brownbear · commit 87438aebe4e8 · 2025-04-21T14:35:01.000+02:00
This introduces a way of sending transport responses without materializing
an actual response object. This gets a little more complicated for responses
that can both go to the local direct response channel as well as to the wire,
but for the per-node query response that isn't sent locally we can use it
to save creating the response object.
The version here only saves a small amount of heap yet (though it makes the GC's
life a little easier) but this can be taken further step by step easily by
also e.g. streaming aggregation results into the output stream instead of ever
materializing them.
Another candidate that could win big from this change is the fetch phase where
we could effectively halve the peak heap use by serializing directly.
diff --git a/server/src/main/java/org/elasticsearch/action/search/SearchQueryThenFetchAsyncAction.java b/server/src/main/java/org/elasticsearch/action/search/SearchQueryThenFetchAsyncAction.java
@@ -23,7 +23,9 @@
 import org.elasticsearch.action.support.IndicesOptions;
 import org.elasticsearch.client.internal.Client;
 import org.elasticsearch.cluster.ClusterState;
+import org.elasticsearch.common.bytes.ReleasableBytesReference;
 import org.elasticsearch.common.io.stream.NamedWriteableRegistry;
+import org.elasticsearch.common.io.stream.RecyclerBytesStreamOutput;
 import org.elasticsearch.common.io.stream.StreamInput;
 import org.elasticsearch.common.io.stream.StreamOutput;
 import org.elasticsearch.common.io.stream.Writeable;
@@ -50,6 +52,7 @@
 import org.elasticsearch.tasks.TaskId;
 import org.elasticsearch.threadpool.ThreadPool;
 import org.elasticsearch.transport.AbstractTransportRequest;
+import org.elasticsearch.transport.BytesTransportResponse;
 import org.elasticsearch.transport.LeakTracker;
 import org.elasticsearch.transport.SendRequestTransportException;
 import org.elasticsearch.transport.Transport;
@@ -58,6 +61,7 @@
 import org.elasticsearch.transport.TransportException;
 import org.elasticsearch.transport.TransportResponse;
 import org.elasticsearch.transport.TransportResponseHandler;
+import org.elasticsearch.transport.TransportService;
 
 import java.io.IOException;
 import java.util.ArrayList;
@@ -215,22 +219,6 @@ public static final class NodeQueryResponse extends TransportResponse {
             this.topDocsStats = SearchPhaseController.TopDocsStats.readFrom(in);
         }
 
-        NodeQueryResponse(
-            QueryPhaseResultConsumer.MergeResult mergeResult,
-            Object[] results,
-            SearchPhaseController.TopDocsStats topDocsStats
-        ) {
-            this.results = results;
-            for (Object result : results) {
-                if (result instanceof QuerySearchResult r) {
-                    r.incRef();
-                }
-            }
-            this.mergeResult = mergeResult;
-            this.topDocsStats = topDocsStats;
-            assert Arrays.stream(results).noneMatch(Objects::isNull) : Arrays.toString(results);
-        }
-
         // public for tests
         public Object[] getResults() {
             return results;
@@ -552,7 +540,7 @@ static void registerNodeSearchAction(
     ) {
         var transportService = searchTransportService.transportService();
         var threadPool = transportService.getThreadPool();
-        final Dependencies dependencies = new Dependencies(searchService, threadPool.executor(ThreadPool.Names.SEARCH));
+        final Dependencies dependencies = new Dependencies(searchService, transportService, threadPool.executor(ThreadPool.Names.SEARCH));
         // Even though not all searches run on the search pool, we use the search pool size as the upper limit of shards to execute in
         // parallel to keep the implementation simple instead of working out the exact pool(s) a query will use up-front.
         final int searchPoolMax = threadPool.info(ThreadPool.Names.SEARCH).getMax();
@@ -715,7 +703,7 @@ public void onFailure(Exception e) {
         }
     }
 
-    private record Dependencies(SearchService searchService, Executor executor) {}
+    private record Dependencies(SearchService searchService, TransportService transportService, Executor executor) {}
 
     private static final class QueryPerNodeState {
 
@@ -760,6 +748,8 @@ void onShardDone() {
             if (countDown.countDown() == false) {
                 return;
             }
+            RecyclerBytesStreamOutput out = null;
+            boolean success = false;
             var channelListener = new ChannelActionListener<>(channel);
             try (queryPhaseResultConsumer) {
                 var failure = queryPhaseResultConsumer.failure.get();
@@ -788,34 +778,46 @@ void onShardDone() {
                         relevantShardIndices.set(localIndex);
                     }
                 }
-                final Object[] results = new Object[queryPhaseResultConsumer.getNumShards()];
-                for (int i = 0; i < results.length; i++) {
-                    var result = queryPhaseResultConsumer.results.get(i);
-                    if (result == null) {
-                        results[i] = failures.get(i);
-                    } else {
-                        // free context id and remove it from the result right away in case we don't need it anymore
-                        if (result instanceof QuerySearchResult q
-                            && q.getContextId() != null
-                            && relevantShardIndices.get(q.getShardIndex()) == false
-                            && q.hasSuggestHits() == false
-                            && q.getRankShardResult() == null
-                            && searchRequest.searchRequest.scroll() == null
-                            && isPartOfPIT(searchRequest.searchRequest, q.getContextId()) == false) {
-                            if (dependencies.searchService.freeReaderContext(q.getContextId())) {
-                                q.clearContextId();
+                final int resultCount = queryPhaseResultConsumer.getNumShards();
+                out = dependencies.transportService.newNetworkBytesStream();
+                out.setTransportVersion(channel.getVersion());
+                try {
+                    out.writeVInt(resultCount);
+                    for (int i = 0; i < resultCount; i++) {
+                        var result = queryPhaseResultConsumer.results.get(i);
+                        if (result == null) {
+                            out.writeBoolean(false);
+                            out.writeException(failures.remove(i));
+                        } else {
+                            // free context id and remove it from the result right away in case we don't need it anymore
+                            if (result instanceof QuerySearchResult q
+                                && q.getContextId() != null
+                                && relevantShardIndices.get(q.getShardIndex()) == false
+                                && q.hasSuggestHits() == false
+                                && q.getRankShardResult() == null
+                                && searchRequest.searchRequest.scroll() == null
+                                && isPartOfPIT(searchRequest.searchRequest, q.getContextId()) == false) {
+                                if (dependencies.searchService.freeReaderContext(q.getContextId())) {
+                                    q.clearContextId();
+                                }
                             }
+                            out.writeBoolean(true);
+                            result.writeTo(out);
                         }
-                        results[i] = result;
                     }
-                    assert results[i] != null;
+                    mergeResult.writeTo(out);
+                    queryPhaseResultConsumer.topDocsStats.writeTo(out);
+                    success = true;
+                } catch (IOException e) {
+                    handleMergeFailure(e, channelListener);
+                    return;
+                }
+            } finally {
+                if (success == false && out != null) {
+                    out.close();
                 }
-
-                ActionListener.respondAndRelease(
-                    channelListener,
-                    new NodeQueryResponse(mergeResult, results, queryPhaseResultConsumer.topDocsStats)
-                );
             }
+            ActionListener.respondAndRelease(channelListener, new BytesTransportResponse(new ReleasableBytesReference(out.bytes(), out)));
         }
 
         private void handleMergeFailure(Exception e, ChannelActionListener<TransportResponse> channelListener) {
diff --git a/server/src/main/java/org/elasticsearch/transport/BytesTransportMessage.java b/server/src/main/java/org/elasticsearch/transport/BytesTransportMessage.java
@@ -0,0 +1,26 @@
+/*
+ * 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.transport;
+
+import org.elasticsearch.common.bytes.ReleasableBytesReference;
+import org.elasticsearch.common.io.stream.StreamOutput;
+
+import java.io.IOException;
+
+public interface BytesTransportMessage {
+
+    ReleasableBytesReference bytes();
+
+    /**
+     * Writes the data in a "thin" manner, without the actual bytes, assumes
+     * the actual bytes will be appended right after this content.
+     */
+    void writeThin(StreamOutput out) throws IOException;
+}
diff --git a/server/src/main/java/org/elasticsearch/transport/BytesTransportRequest.java b/server/src/main/java/org/elasticsearch/transport/BytesTransportRequest.java
@@ -10,7 +10,6 @@
 package org.elasticsearch.transport;
 
 import org.elasticsearch.TransportVersion;
-import org.elasticsearch.common.bytes.BytesReference;
 import org.elasticsearch.common.bytes.ReleasableBytesReference;
 import org.elasticsearch.common.io.stream.StreamInput;
 import org.elasticsearch.common.io.stream.StreamOutput;
@@ -21,7 +20,7 @@
  * A specialized, bytes only request, that can potentially be optimized on the network
  * layer, specifically for the same large buffer send to several nodes.
  */
-public class BytesTransportRequest extends AbstractTransportRequest {
+public class BytesTransportRequest extends AbstractTransportRequest implements BytesTransportMessage {
 
     final ReleasableBytesReference bytes;
     private final TransportVersion version;
@@ -41,14 +40,12 @@ public TransportVersion version() {
         return this.version;
     }
 
-    public BytesReference bytes() {
+    @Override
+    public ReleasableBytesReference bytes() {
         return this.bytes;
     }
 
-    /**
-     * Writes the data in a "thin" manner, without the actual bytes, assumes
-     * the actual bytes will be appended right after this content.
-     */
+    @Override
     public void writeThin(StreamOutput out) throws IOException {
         super.writeTo(out);
         out.writeVInt(bytes.length());
diff --git a/server/src/main/java/org/elasticsearch/transport/BytesTransportResponse.java b/server/src/main/java/org/elasticsearch/transport/BytesTransportResponse.java
@@ -0,0 +1,60 @@
+/*
+ * 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.transport;
+
+import org.elasticsearch.common.bytes.ReleasableBytesReference;
+import org.elasticsearch.common.io.stream.StreamOutput;
+
+import java.io.IOException;
+
+/**
+ * A specialized, bytes only response, that can potentially be optimized on the network layer.
+ */
+public class BytesTransportResponse extends TransportResponse implements BytesTransportMessage {
+
+    private final ReleasableBytesReference bytes;
+
+    public BytesTransportResponse(ReleasableBytesReference bytes) {
+        this.bytes = bytes;
+    }
+
+    @Override
+    public ReleasableBytesReference bytes() {
+        return this.bytes;
+    }
+
+    @Override
+    public void writeThin(StreamOutput out) throws IOException {}
+
+    @Override
+    public void writeTo(StreamOutput out) throws IOException {
+        bytes.writeTo(out);
+    }
+
+    @Override
+    public void incRef() {
+        bytes.incRef();
+    }
+
+    @Override
+    public boolean tryIncRef() {
+        return bytes.tryIncRef();
+    }
+
+    @Override
+    public boolean decRef() {
+        return bytes.decRef();
+    }
+
+    @Override
+    public boolean hasReferences() {
+        return bytes.hasReferences();
+    }
+}
diff --git a/server/src/main/java/org/elasticsearch/transport/OutboundHandler.java b/server/src/main/java/org/elasticsearch/transport/OutboundHandler.java
@@ -227,7 +227,7 @@ private void sendMessage(
         Releasable onAfter
     ) throws IOException {
         assert action != null;
-        final var compressionScheme = writeable instanceof BytesTransportRequest ? null : possibleCompressionScheme;
+        final var compressionScheme = writeable instanceof BytesTransportMessage ? null : possibleCompressionScheme;
         final BytesReference message;
         boolean serializeSuccess = false;
         final RecyclerBytesStreamOutput byteStreamOutput = new RecyclerBytesStreamOutput(recycler);
@@ -334,11 +334,11 @@ private static BytesReference serializeMessageBody(
         final ReleasableBytesReference zeroCopyBuffer;
         try {
             stream.setTransportVersion(version);
-            if (writeable instanceof BytesTransportRequest bRequest) {
+            if (writeable instanceof BytesTransportMessage bRequest) {
                 assert stream == byteStreamOutput;
                 assert compressionScheme == null;
                 bRequest.writeThin(stream);
-                zeroCopyBuffer = bRequest.bytes;
+                zeroCopyBuffer = bRequest.bytes();
             } else if (writeable instanceof RemoteTransportException remoteTransportException) {
                 stream.writeException(remoteTransportException);
                 zeroCopyBuffer = ReleasableBytesReference.empty();
