Add streaming search with configurable scoring modes

modules/reindex/src/test/java/org/opensearch/index/reindex/DeleteByQueryBasicTests.java

@@ -54,6 +54,7 @@
 import java.util.Map;
 import java.util.stream.Collectors;
 
+import static org.opensearch.action.search.StreamSearchTransportService.STREAM_SEARCH_ENABLED;
 import static org.opensearch.cluster.metadata.IndexMetadata.SETTING_READ_ONLY;
 import static org.opensearch.cluster.metadata.IndexMetadata.SETTING_READ_ONLY_ALLOW_DELETE;
 import static org.opensearch.index.query.QueryBuilders.matchQuery;
@@ -383,6 +384,63 @@ public void testMissingSources() {
         assertThat(response, matcher().deleted(0).slices(hasSize(0)));
     }
 
+    /**
+     * Regression test to ensure delete-by-query works correctly even when streaming search is enabled.
+     * Since delete-by-query uses scroll searches, it should automatically opt out of the streaming pipeline.
+     */
+    public void testDeleteByQueryWithStreamingEnabled() throws Exception {
+        // Enable streaming search globally
+        Settings streamingSettings = Settings.builder().put(STREAM_SEARCH_ENABLED.getKey(), true).build();
+        assertAcked(client().admin().cluster().prepareUpdateSettings().setTransientSettings(streamingSettings).get());
+
+        try {
+            // Index some test documents
+            indexRandom(
+                true,
+                client().prepareIndex("test").setId("1").setSource("foo", "delete_me"),
+                client().prepareIndex("test").setId("2").setSource("foo", "delete_me"),
+                client().prepareIndex("test").setId("3").setSource("foo", "keep_me")
+            );
+
+            assertHitCount(client().prepareSearch("test").setSize(0).get(), 3);
+
+            // Perform delete-by-query - this should work without "topDocs already consumed" errors
+            BulkByScrollResponse response = deleteByQuery().source("test").filter(termQuery("foo", "delete_me")).refresh(true).get();
+
+            // Verify the delete operation succeeded
+            assertThat(response, matcher().deleted(2));
+            assertHitCount(client().prepareSearch("test").setSize(0).get(), 1);
+
+            // Verify only the expected document remains
+            assertHitCount(client().prepareSearch("test").setQuery(termQuery("foo", "keep_me")).get(), 1);
+        } finally {
+            // Comprehensive cleanup: reset both transient and persistent streaming settings
+            Settings disableStreamingSettings = Settings.builder().putNull(STREAM_SEARCH_ENABLED.getKey()).build();
+            try {
+                // Clear transient settings
+                assertAcked(client().admin().cluster().prepareUpdateSettings().setTransientSettings(disableStreamingSettings).get());
+                // Clear persistent settings (in case they were set)
+                assertAcked(client().admin().cluster().prepareUpdateSettings().setPersistentSettings(disableStreamingSettings).get());
+
+                // Assert the setting is cleared to prevent test pollution
+                assertFalse(
+                    "Stream search should be disabled after cleanup",
+                    client().admin()
+                        .cluster()
+                        .prepareState()
+                        .get()
+                        .getState()
+                        .getMetadata()
+                        .transientSettings()
+                        .getAsBoolean(STREAM_SEARCH_ENABLED.getKey(), false)
+                );
+            } catch (Exception cleanupException) {
+                // Log cleanup failures but don't fail the test
+                logger.warn("Failed to clean up streaming settings", cleanupException);
+            }
+        }
+    }
+
     /** Enables or disables the cluster disk allocation decider **/
     private void setDiskAllocationDeciderEnabled(boolean value) {
         Settings settings = value

plugins/arrow-flight-rpc/src/main/java/org/opensearch/arrow/flight/transport/FlightOutboundHandler.java

@@ -17,6 +17,8 @@
 package org.opensearch.arrow.flight.transport;
 
 import org.apache.arrow.flight.FlightRuntimeException;
+import org.apache.logging.log4j.LogManager;
+import org.apache.logging.log4j.Logger;
 import org.opensearch.Version;
 import org.opensearch.cluster.node.DiscoveryNode;
 import org.opensearch.common.io.stream.BytesStreamOutput;
@@ -40,10 +42,13 @@
 
 /**
  * Outbound handler for Arrow Flight streaming responses.
- * It must invoke messageListener and relay any exception back to the caller and not supress them
+ * It must invoke messageListener and relay any exception back to the caller and
+ * not supress them
+ *
  * @opensearch.internal
  */
 class FlightOutboundHandler extends ProtocolOutboundHandler {
+    private static final Logger logger = LogManager.getLogger(FlightOutboundHandler.class);
     private volatile TransportMessageListener messageListener = TransportMessageListener.NOOP_LISTENER;
     private final String nodeName;
     private final Version version;
@@ -157,6 +162,19 @@ private void processBatchTask(BatchTask task) {
         try {
             try (VectorStreamOutput out = new VectorStreamOutput(flightChannel.getAllocator(), flightChannel.getRoot())) {
                 task.response().writeTo(out);
+                if (task.response() instanceof org.opensearch.search.query.QuerySearchResult) {
+                    logger.info(
+                        "QuerySearchResult hasAggs: {}",
+                        ((org.opensearch.search.query.QuerySearchResult) task.response()).hasAggs()
+                    );
+                }
+                logger.info(
+                    "Sending batch for requestId [{}], action [{}], items [{}], rows [{}]",
+                    task.requestId(),
+                    task.action(),
+                    task.response(),
+                    out.getRoot().getRowCount()
+                );
                 flightChannel.sendBatch(getHeaderBuffer(task.requestId(), task.nodeVersion(), task.features()), out);
                 messageListener.onResponseSent(task.requestId(), task.action(), task.response());
             }
@@ -292,7 +310,8 @@ public void setMessageListener(TransportMessageListener listener) {
     }
 
     private ByteBuffer getHeaderBuffer(long requestId, Version nodeVersion, Set<String> features) throws IOException {
-        // Just a way( probably inefficient) to serialize header to reuse existing logic present in
+        // Just a way( probably inefficient) to serialize header to reuse existing logic
+        // present in
         // NativeOutboundMessage.Response#writeVariableHeader()
         NativeOutboundMessage.Response headerMessage = new NativeOutboundMessage.Response(
             threadPool.getThreadContext(),

plugins/arrow-flight-rpc/src/main/java/org/opensearch/arrow/flight/transport/FlightStreamPlugin.java

@@ -235,6 +235,21 @@ public Map<String, Supplier<Transport>> getTransports(
         return Collections.emptyMap();
     }
 
+    @Override
+    public Map<String, Supplier<Transport>> getStreamTransports(
+        Settings settings,
+        ThreadPool threadPool,
+        PageCacheRecycler pageCacheRecycler,
+        CircuitBreakerService circuitBreakerService,
+        NamedWriteableRegistry namedWriteableRegistry,
+        NetworkService networkService,
+        Tracer tracer
+    ) {
+        // Return empty map since getTransports() already registers the FLIGHT transport
+        // The NetworkModule will find it via getStreamTransportSupplier() which looks in transportFactories
+        return Collections.emptyMap();
+    }
+
     /**
      * Gets the auxiliary transports for the FlightStream plugin.
      * @param settings The settings for the plugin.

plugins/arrow-flight-rpc/src/main/java/org/opensearch/arrow/flight/transport/FlightTransportResponse.java

@@ -32,6 +32,14 @@
 import static org.opensearch.arrow.flight.transport.ClientHeaderMiddleware.CORRELATION_ID_KEY;
 
 /**
+ * Arrow Flight implementation of streaming transport responses.
+ *
+ * <p>
+ * Handles streaming responses from Arrow Flight servers with lazy batch
+ * processing.
+ * Headers are extracted when first accessed, and responses are deserialized on
+ * demand.
+
  * Streaming transport response implementation using Arrow Flight.
  * Manages Flight stream lifecycle with lazy initialization and prefetching support.
  */
@@ -164,9 +172,13 @@ public void close() {
         if (flightStream != null) {
             try {
                 flightStream.close();
-            } catch (IllegalStateException ignore) {} catch (Exception e) {
+            } catch (IllegalStateException ignore) {
+                // this is fine if the allocator is already closed
+            } catch (Exception e) {
                 throw new StreamException(StreamErrorCode.INTERNAL, "Error closing flight stream", e);
             }
         }
+        closed = true;
     }
+
 }

server/src/main/java/org/opensearch/action/ActionModule.java

@@ -289,14 +289,14 @@
 import org.opensearch.action.search.SearchAction;
 import org.opensearch.action.search.SearchScrollAction;
 import org.opensearch.action.search.StreamSearchAction;
-import org.opensearch.action.search.StreamTransportSearchAction;
 import org.opensearch.action.search.TransportClearScrollAction;
 import org.opensearch.action.search.TransportCreatePitAction;
 import org.opensearch.action.search.TransportDeletePitAction;
 import org.opensearch.action.search.TransportGetAllPitsAction;
 import org.opensearch.action.search.TransportMultiSearchAction;
 import org.opensearch.action.search.TransportSearchAction;
 import org.opensearch.action.search.TransportSearchScrollAction;
+import org.opensearch.action.search.TransportStreamSearchAction;
 import org.opensearch.action.support.ActionFilters;
 import org.opensearch.action.support.AutoCreateIndex;
 import org.opensearch.action.support.DestructiveOperations;
@@ -741,9 +741,7 @@ public <Request extends ActionRequest, Response extends ActionResponse> void reg
         actions.register(MultiGetAction.INSTANCE, TransportMultiGetAction.class, TransportShardMultiGetAction.class);
         actions.register(BulkAction.INSTANCE, TransportBulkAction.class, TransportShardBulkAction.class);
         actions.register(SearchAction.INSTANCE, TransportSearchAction.class);
-        if (FeatureFlags.isEnabled(FeatureFlags.STREAM_TRANSPORT)) {
-            actions.register(StreamSearchAction.INSTANCE, StreamTransportSearchAction.class);
-        }
+        actions.register(StreamSearchAction.INSTANCE, TransportStreamSearchAction.class);
         actions.register(SearchScrollAction.INSTANCE, TransportSearchScrollAction.class);
         actions.register(MultiSearchAction.INSTANCE, TransportMultiSearchAction.class);
         actions.register(ExplainAction.INSTANCE, TransportExplainAction.class);