[Streaming Indexing] Introduce bulk HTTP API streaming flavor

CHANGELOG.md

@@ -26,6 +26,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 - Add allowlist setting for ingest-geoip and ingest-useragent ([#15325](https://github.com/opensearch-project/OpenSearch/pull/15325))
 - Adding access to noSubMatches and noOverlappingMatches in Hyphenation ([#13895](https://github.com/opensearch-project/OpenSearch/pull/13895))
 - Add support for index level max slice count setting for concurrent segment search ([#15336](https://github.com/opensearch-project/OpenSearch/pull/15336))
+- [Streaming Indexing] Introduce bulk HTTP API streaming flavor ([#15381](https://github.com/opensearch-project/OpenSearch/pull/15381))
 - Add support for centralize snapshot creation with pinned timestamp ([#15124](https://github.com/opensearch-project/OpenSearch/pull/15124))
 - Add concurrent search support for Derived Fields ([#15326](https://github.com/opensearch-project/OpenSearch/pull/15326))
 

buildSrc/version.properties

@@ -37,7 +37,7 @@ reactor_netty     = 1.1.22
 reactor           = 3.5.20
 
 # client dependencies
-httpclient5       = 5.2.3
+httpclient5       = 5.3.1
 httpcore5         = 5.2.5
 httpclient        = 4.5.14
 httpcore          = 4.4.16

client/rest/licenses/httpclient5-5.3.1.jar.sha1

@@ -0,0 +1 @@
+56b53c8f4bcdaada801d311cf2ff8a24d6d96883

client/rest/src/main/java/org/opensearch/client/RestClient.java

@@ -114,6 +114,7 @@
 import java.util.zip.GZIPOutputStream;
 
 import org.reactivestreams.Publisher;
+import reactor.core.publisher.Flux;
 import reactor.core.publisher.Mono;
 import reactor.core.publisher.MonoSink;
 
@@ -416,7 +417,12 @@
                 try {
                     final ResponseOrResponseException responseOrResponseException = convertResponse(request, node, message);
                     if (responseOrResponseException.responseException == null) {
-                        return Mono.just(message);
+                        return Mono.just(
+                            new Message<>(
+                                message.getHead(),
+                                Flux.from(message.getBody()).flatMapSequential(b -> Flux.fromIterable(frame(b)))
+                            )
+                        );
                     } else {
                         if (nodeTuple.nodes.hasNext()) {
                             return Mono.from(streamRequest(nodeTuple, request));
@@ -431,6 +437,48 @@
         });
     }
 
+    /**
+     * Frame the {@link ByteBuffer} into individual chunks that are separated by '\r\n' sequence.
+     * @param b {@link ByteBuffer} to split
+     * @return individual chunks
+     */
+    private static Collection<ByteBuffer> frame(ByteBuffer b) {
+        final Collection<ByteBuffer> buffers = new ArrayList<>();
+
+        int position = b.position();
+        while (b.hasRemaining()) {
+            // Skip the chunk separator when it comes right at the beginning
+            if (b.get() == '\r' && b.hasRemaining() && b.position() > 1) {
+                if (b.get() == '\n') {
+                    final byte[] chunk = new byte[b.position() - position];
+
+                    b.position(position);
+                    b.get(chunk);
+
+                    // Do not copy the '\r\n' sequence
+                    buffers.add(ByteBuffer.wrap(chunk, 0, chunk.length - 2));
+                    position = b.position();
+                }
+            }
+        }
+
+        if (buffers.isEmpty()) {
+            return Collections.singleton(b);
+        }
+
+        // Copy last chunk
+        if (position != b.position()) {
+            final byte[] chunk = new byte[b.position() - position];
+
+            b.position(position);
+            b.get(chunk);
+
+            buffers.add(ByteBuffer.wrap(chunk, 0, chunk.length));
+        }
+
+        return buffers;
+    }
+
     private ResponseOrResponseException convertResponse(InternalRequest request, Node node, ClassicHttpResponse httpResponse)
         throws IOException {
         RequestLogger.logResponse(logger, request.httpRequest, node.getHost(), httpResponse);

client/sniffer/licenses/httpclient5-5.3.1.jar.sha1

@@ -0,0 +1 @@
+56b53c8f4bcdaada801d311cf2ff8a24d6d96883

plugins/transport-reactor-netty4/src/javaRestTest/java/org/opensearch/rest/ReactorNetty4StreamingIT.java

@@ -44,7 +44,7 @@ public void tearDown() throws Exception {
         super.tearDown();
     }
 
-    public void testStreamingRequest() throws IOException {
+    public void testStreamingRequestNoBatching() throws IOException {
         final VirtualTimeScheduler scheduler = VirtualTimeScheduler.create(true);
 
         final Stream<String> stream = IntStream.range(1, 6)
@@ -85,6 +85,167 @@ public void testStreamingRequest() throws IOException {
         assertThat(count, equalTo(5));
     }
 
+    public void testStreamingRequestOneBatchBySize() throws IOException, InterruptedException {
+        final Stream<String> stream = IntStream.range(1, 6)
+            .mapToObj(id -> "{ \"index\": { \"_index\": \"test-streaming\", \"_id\": \"" + id + "\" } }\n" + "{ \"name\": \"josh\" }\n");
+
+        final Duration delay = Duration.ofMillis(1);
+        final StreamingRequest<ByteBuffer> streamingRequest = new StreamingRequest<>(
+            "POST",
+            "/_bulk/stream",
+            Flux.fromStream(stream).delayElements(delay).map(s -> ByteBuffer.wrap(s.getBytes(StandardCharsets.UTF_8)))
+        );
+        streamingRequest.addParameter("refresh", "true");
+        streamingRequest.addParameter("batch_size", "5");
+
+        final StreamingResponse<ByteBuffer> streamingResponse = client().streamRequest(streamingRequest);
+
+        StepVerifier.create(Flux.from(streamingResponse.getBody()).map(b -> new String(b.array(), StandardCharsets.UTF_8)))
+            .expectNextMatches(
+                s -> s.contains("\"result\":\"created\"")
+                    && s.contains("\"_id\":\"1\"")
+                    && s.contains("\"result\":\"created\"")
+                    && s.contains("\"_id\":\"2\"")
+                    && s.contains("\"result\":\"created\"")
+                    && s.contains("\"_id\":\"3\"")
+                    && s.contains("\"result\":\"created\"")
+                    && s.contains("\"_id\":\"4\"")
+                    && s.contains("\"result\":\"created\"")
+                    && s.contains("\"_id\":\"5\"")
+            )
+            .expectComplete()
+            .verify();
+
+        assertThat(streamingResponse.getStatusLine().getStatusCode(), equalTo(200));
+        assertThat(streamingResponse.getWarnings(), empty());
+
+        final Request request = new Request("GET", "/test-streaming/_count");
+        final Response response = client().performRequest(request);
+        final ObjectPath objectPath = ObjectPath.createFromResponse(response);
+        final Integer count = objectPath.evaluate("count");
+        assertThat(count, equalTo(5));
+    }
+
+    public void testStreamingRequestManyBatchesBySize() throws IOException {
+        final Stream<String> stream = IntStream.range(1, 6)
+            .mapToObj(id -> "{ \"index\": { \"_index\": \"test-streaming\", \"_id\": \"" + id + "\" } }\n" + "{ \"name\": \"josh\" }\n");
+
+        final Duration delay = Duration.ofMillis(1);
+        final StreamingRequest<ByteBuffer> streamingRequest = new StreamingRequest<>(
+            "POST",
+            "/_bulk/stream",
+            Flux.fromStream(stream).delayElements(delay).map(s -> ByteBuffer.wrap(s.getBytes(StandardCharsets.UTF_8)))
+        );
+        streamingRequest.addParameter("refresh", "true");
+        streamingRequest.addParameter("batch_size", "3");
+
+        final StreamingResponse<ByteBuffer> streamingResponse = client().streamRequest(streamingRequest);
+
+        StepVerifier.create(Flux.from(streamingResponse.getBody()).map(b -> new String(b.array(), StandardCharsets.UTF_8)))
+            .expectNextMatches(
+                s -> s.contains("\"result\":\"created\"")
+                    && s.contains("\"_id\":\"1\"")
+                    && s.contains("\"result\":\"created\"")
+                    && s.contains("\"_id\":\"2\"")
+                    && s.contains("\"result\":\"created\"")
+                    && s.contains("\"_id\":\"3\"")
+            )
+            .expectNextMatches(
+                s -> s.contains("\"result\":\"created\"")
+                    && s.contains("\"_id\":\"4\"")
+                    && s.contains("\"result\":\"created\"")
+                    && s.contains("\"_id\":\"5\"")
+            )
+            .expectComplete()
+            .verify();
+
+        assertThat(streamingResponse.getStatusLine().getStatusCode(), equalTo(200));
+        assertThat(streamingResponse.getWarnings(), empty());
+
+        final Request request = new Request("GET", "/test-streaming/_count");
+        final Response response = client().performRequest(request);
+        final ObjectPath objectPath = ObjectPath.createFromResponse(response);
+        final Integer count = objectPath.evaluate("count");
+        assertThat(count, equalTo(5));
+    }
+
+    public void testStreamingRequestManyBatchesByInterval() throws IOException {
+        final Stream<String> stream = IntStream.range(1, 6)
+            .mapToObj(id -> "{ \"index\": { \"_index\": \"test-streaming\", \"_id\": \"" + id + "\" } }\n" + "{ \"name\": \"josh\" }\n");
+
+        final Duration delay = Duration.ofMillis(500);
+        final StreamingRequest<ByteBuffer> streamingRequest = new StreamingRequest<>(
+            "POST",
+            "/_bulk/stream",
+            Flux.fromStream(stream).delayElements(delay).map(s -> ByteBuffer.wrap(s.getBytes(StandardCharsets.UTF_8)))
+        );
+        streamingRequest.addParameter("refresh", "true");
+        streamingRequest.addParameter("batch_interval", "5s");
+
+        final StreamingResponse<ByteBuffer> streamingResponse = client().streamRequest(streamingRequest);
+
+        // We don't check for a other documents here since those may appear in any of the chunks (it is very
+        // difficult to get the timing right). But at the end, the total number of the documents is being checked.
+        StepVerifier.create(Flux.from(streamingResponse.getBody()).map(b -> new String(b.array(), StandardCharsets.UTF_8)))
+            .expectNextMatches(
+                s -> s.contains("\"result\":\"created\"")
+                    && s.contains("\"_id\":\"1\"")
+                    && s.contains("\"result\":\"created\"")
+                    && s.contains("\"_id\":\"2\"")
+                    && s.contains("\"result\":\"created\"")
+                    && s.contains("\"_id\":\"3\"")
+                    && s.contains("\"result\":\"created\"")
+                    && s.contains("\"_id\":\"4\"")
+                    && s.contains("\"result\":\"created\"")
+                    && s.contains("\"_id\":\"5\"")
+            )
+            .expectComplete()
+            .verify();
+
+        assertThat(streamingResponse.getStatusLine().getStatusCode(), equalTo(200));
+        assertThat(streamingResponse.getWarnings(), empty());
+
+        final Request request = new Request("GET", "/test-streaming/_count");
+        final Response response = client().performRequest(request);
+        final ObjectPath objectPath = ObjectPath.createFromResponse(response);
+        final Integer count = objectPath.evaluate("count");
+        assertThat(count, equalTo(5));
+    }
+
+    public void testStreamingRequestManyBatchesByIntervalAndSize() throws IOException {
+        final Stream<String> stream = IntStream.range(1, 6)
+            .mapToObj(id -> "{ \"index\": { \"_index\": \"test-streaming\", \"_id\": \"" + id + "\" } }\n" + "{ \"name\": \"josh\" }\n");
+
+        final Duration delay = Duration.ofSeconds(1);
+        final StreamingRequest<ByteBuffer> streamingRequest = new StreamingRequest<>(
+            "POST",
+            "/_bulk/stream",
+            Flux.fromStream(stream).delayElements(delay).map(s -> ByteBuffer.wrap(s.getBytes(StandardCharsets.UTF_8)))
+        );
+        streamingRequest.addParameter("refresh", "true");
+        streamingRequest.addParameter("batch_interval", "3s");
+        streamingRequest.addParameter("batch_size", "5");
+
+        final StreamingResponse<ByteBuffer> streamingResponse = client().streamRequest(streamingRequest);
+
+        // We don't check for a other documents here since those may appear in any of the chunks (it is very
+        // difficult to get the timing right). But at the end, the total number of the documents is being checked.
+        StepVerifier.create(Flux.from(streamingResponse.getBody()).map(b -> new String(b.array(), StandardCharsets.UTF_8)))
+            .expectNextMatches(s -> s.contains("\"result\":\"created\"") && s.contains("\"_id\":\"1\""))
+            .expectNextMatches(s -> s.contains("\"result\":\"created\"") && s.contains("\"_id\":\"5\""))
+            .expectComplete()
+            .verify();
+
+        assertThat(streamingResponse.getStatusLine().getStatusCode(), equalTo(200));
+        assertThat(streamingResponse.getWarnings(), empty());
+
+        final Request request = new Request("GET", "/test-streaming/_count");
+        final Response response = client().performRequest(request);
+        final ObjectPath objectPath = ObjectPath.createFromResponse(response);
+        final Integer count = objectPath.evaluate("count");
+        assertThat(count, equalTo(5));
+    }
+
     public void testStreamingBadRequest() throws IOException {
         final Stream<String> stream = Stream.of(
             "{ \"index\": { \"_index\": \"test-streaming\", \"_id\": \"1\" } }\n" + "{ \"name\": \"josh\"  }\n"

plugins/transport-reactor-netty4/src/main/java/org/opensearch/http/reactor/netty4/ReactorNetty4StreamingHttpChannel.java

@@ -103,7 +103,6 @@ public void receiveChunk(HttpChunk message) {
             }
         } catch (final Exception ex) {
             producer.error(ex);
-        } finally {
             message.close();
         }
     }

plugins/transport-reactor-netty4/src/main/java/org/opensearch/http/reactor/netty4/ReactorNetty4StreamingRequestConsumer.java

@@ -44,7 +44,7 @@ public void subscribe(Subscriber<? super HttpContent> s) {
     }
 
     HttpChunk createChunk(HttpContent chunk, boolean last) {
-        return new ReactorNetty4HttpChunk(chunk.copy().content(), last);
+        return new ReactorNetty4HttpChunk(chunk.copy().content().retain(), last);
     }
 
     StreamingHttpChannel httpChannel() {

plugins/transport-reactor-netty4/src/test/java/org/opensearch/http/reactor/netty4/ReactorNetty4HttpServerTransportStreamingTests.java

@@ -191,7 +191,7 @@ public void dispatchBadRequest(final RestChannel channel, final ThreadContext th
                         } catch (final IOException ex) {
                             throw new UncheckedIOException(ex);
                         }
-                    }).collect(Collectors.joining(""))));
+                    }).collect(Collectors.joining("\r\n", "", "\r\n"))));
                 } finally {
                     response.release();
                 }

server/src/main/java/org/opensearch/common/xcontent/support/XContentHttpChunk.java

@@ -12,13 +12,15 @@
 import org.opensearch.common.lease.Releasable;
 import org.opensearch.core.common.bytes.BytesArray;
 import org.opensearch.core.common.bytes.BytesReference;
+import org.opensearch.core.common.bytes.CompositeBytesReference;
 import org.opensearch.core.xcontent.XContentBuilder;
 import org.opensearch.http.HttpChunk;
 
 /**
  * Wraps the instance of the {@link XContentBuilder} into {@link HttpChunk}
  */
 public final class XContentHttpChunk implements HttpChunk {
+    private static final byte[] CHUNK_SEPARATOR = new byte[] { '\r', '\n' };
     private final BytesReference content;
 
     /**
@@ -42,7 +44,8 @@ private XContentHttpChunk(@Nullable final XContentBuilder builder) {
         if (builder == null /* no content */) {
             content = BytesArray.EMPTY;
         } else {
-            content = BytesReference.bytes(builder);
+            // Always finalize the output chunk with '\r\n' sequence
+            content = CompositeBytesReference.of(BytesReference.bytes(builder), new BytesArray(CHUNK_SEPARATOR));
         }
     }