[Segment Replication] Add check to cancel ongoing replication with old primary on onNewCheckpoint on replica

CHANGELOG.md

@@ -49,6 +49,7 @@ Inspired from [Keep a Changelog](https://keepachangelog.com/en/1.0.0/)
 - Fixed cancellation of segment replication events ([#4225](https://github.com/opensearch-project/OpenSearch/pull/4225))
 - [Segment Replication] Add check to cancel ongoing replication with old primary on onNewCheckpoint on replica ([#4363](https://github.com/opensearch-project/OpenSearch/pull/4363))
 - [Segment Replication] Bump segment infos counter before commit during replica promotion ([#4365](https://github.com/opensearch-project/OpenSearch/pull/4365))
+- [Segment Replication] Extend FileChunkWriter to allow cancel on transport client ([#4386](https://github.com/opensearch-project/OpenSearch/pull/4386))
 - [Segment Replication] Fix NoSuchFileExceptions with segment replication when computing primary metadata snapshots ([#4366](https://github.com/opensearch-project/OpenSearch/pull/4366))
 
 ### Security

server/src/internalClusterTest/java/org/opensearch/indices/replication/SegmentReplicationIT.java

@@ -30,17 +30,23 @@
 import org.opensearch.index.engine.Segment;
 import org.opensearch.index.shard.IndexShard;
 import org.opensearch.indices.IndicesService;
+import org.opensearch.indices.recovery.FileChunkRequest;
 import org.opensearch.indices.replication.common.ReplicationType;
+import org.opensearch.plugins.Plugin;
 import org.opensearch.test.BackgroundIndexer;
 import org.opensearch.test.InternalTestCluster;
 import org.opensearch.test.OpenSearchIntegTestCase;
+import org.opensearch.test.transport.MockTransportService;
+import org.opensearch.transport.TransportService;
 
 import java.io.IOException;
+import java.util.Collection;
 import java.util.Arrays;
 import java.util.List;
 import java.util.Map;
-import java.util.Optional;
 import java.util.Set;
+import java.util.Optional;
+import java.util.concurrent.CountDownLatch;
 import java.util.concurrent.TimeUnit;
 import java.util.function.Function;
 import java.util.stream.Collectors;
@@ -60,6 +66,11 @@ public static void assumeFeatureFlag() {
         assumeTrue("Segment replication Feature flag is enabled", Boolean.parseBoolean(System.getProperty(FeatureFlags.REPLICATION_TYPE)));
     }
 
+    @Override
+    protected Collection<Class<? extends Plugin>> nodePlugins() {
+        return Arrays.asList(MockTransportService.TestPlugin.class);
+    }
+
     @Override
     public Settings indexSettings() {
         return Settings.builder()
@@ -313,6 +324,65 @@ public void testReplicationAfterForceMerge() throws Exception {
         }
     }
 
+    public void testCancellation() throws Exception {
+        final String primaryNode = internalCluster().startNode();
+        createIndex(INDEX_NAME, Settings.builder().put(indexSettings()).put(IndexMetadata.SETTING_NUMBER_OF_REPLICAS, 1).build());
+        ensureYellow(INDEX_NAME);
+
+        final String replicaNode = internalCluster().startNode();
+
+        final SegmentReplicationSourceService segmentReplicationSourceService = internalCluster().getInstance(
+            SegmentReplicationSourceService.class,
+            primaryNode
+        );
+        final IndexShard primaryShard = getIndexShard(primaryNode);
+
+        CountDownLatch latch = new CountDownLatch(1);
+
+        MockTransportService mockTransportService = ((MockTransportService) internalCluster().getInstance(
+            TransportService.class,
+            primaryNode
+        ));
+        mockTransportService.addSendBehavior(
+            internalCluster().getInstance(TransportService.class, replicaNode),
+            (connection, requestId, action, request, options) -> {
+                if (action.equals(SegmentReplicationTargetService.Actions.FILE_CHUNK)) {
+                    FileChunkRequest req = (FileChunkRequest) request;
+                    logger.debug("file chunk [{}] lastChunk: {}", req, req.lastChunk());
+                    if (req.name().endsWith("cfs") && req.lastChunk()) {
+                        try {
+                            latch.await();
+                        } catch (InterruptedException e) {
+                            throw new RuntimeException(e);
+                        }
+                    }
+                }
+                connection.sendRequest(requestId, action, request, options);
+            }
+        );
+
+        final int docCount = scaledRandomIntBetween(0, 200);
+        try (
+            BackgroundIndexer indexer = new BackgroundIndexer(
+                INDEX_NAME,
+                "_doc",
+                client(),
+                -1,
+                RandomizedTest.scaledRandomIntBetween(2, 5),
+                false,
+                random()
+            )
+        ) {
+            indexer.start(docCount);
+            waitForDocs(docCount, indexer);
+
+            flush(INDEX_NAME);
+        }
+        segmentReplicationSourceService.beforeIndexShardClosed(primaryShard.shardId(), primaryShard, indexSettings());
+        latch.countDown();
+        assertDocCounts(docCount, primaryNode);
+    }
+
     public void testStartReplicaAfterPrimaryIndexesDocs() throws Exception {
         final String primaryNode = internalCluster().startNode();
         createIndex(INDEX_NAME, Settings.builder().put(indexSettings()).put(IndexMetadata.SETTING_NUMBER_OF_REPLICAS, 0).build());

server/src/main/java/org/opensearch/indices/recovery/FileChunkWriter.java

@@ -28,4 +28,6 @@ void writeFileChunk(
         int totalTranslogOps,
         ActionListener<Void> listener
     );
+
+    default void cancel() {}
 }

server/src/main/java/org/opensearch/indices/replication/OngoingSegmentReplications.java

@@ -126,7 +126,7 @@ void startSegmentCopy(GetSegmentFilesRequest request, ActionListener<GetSegmentF
 
     /**
      * Prepare for a Replication event. This method constructs a {@link CopyState} holding files to be sent off of the current
-     * nodes's store.  This state is intended to be sent back to Replicas before copy is initiated so the replica can perform a diff against its
+     * node's store.  This state is intended to be sent back to Replicas before copy is initiated so the replica can perform a diff against its
      * local store.  It will then build a handler to orchestrate the segment copy that will be stored locally and started on a subsequent request from replicas
      * with the list of required files.
      *

server/src/main/java/org/opensearch/indices/replication/RemoteSegmentFileChunkWriter.java

@@ -122,4 +122,9 @@ public void writeFileChunk(
             reader
         );
     }
+
+    @Override
+    public void cancel() {
+        retryableTransportClient.cancel();
+    }
 }

server/src/main/java/org/opensearch/indices/replication/SegmentReplicationSourceHandler.java

@@ -58,6 +58,8 @@ class SegmentReplicationSourceHandler {
     private final DiscoveryNode targetNode;
     private final String allocationId;
 
+    private final FileChunkWriter writer;
+
     /**
      * Constructor.
      *
@@ -96,6 +98,7 @@ class SegmentReplicationSourceHandler {
         );
         this.allocationId = allocationId;
         this.copyState = copyState;
+        this.writer = writer;
     }
 
     /**
@@ -186,9 +189,10 @@ public synchronized void sendFiles(GetSegmentFilesRequest request, ActionListene
     }
 
     /**
-     * Cancels the recovery and interrupts all eligible threads.
+     * Cancels the replication and interrupts all eligible threads.
      */
     public void cancel(String reason) {
+        writer.cancel();
         cancellableThreads.cancel(reason);
     }
 

server/src/test/java/org/opensearch/indices/replication/SegmentReplicationSourceHandlerTests.java

@@ -34,6 +34,8 @@
 import java.util.concurrent.TimeUnit;
 
 import static org.mockito.Mockito.mock;
+import static org.mockito.Mockito.verify;
+import static org.mockito.Mockito.times;
 
 public class SegmentReplicationSourceHandlerTests extends IndexShardTestCase {
 
@@ -245,6 +247,7 @@ public void onFailure(Exception e) {
             }
         });
         latch.await(2, TimeUnit.SECONDS);
+        verify(chunkWriter, times(1)).cancel();
         assertEquals("listener should have resolved with failure", 0, latch.getCount());
     }
 }