HBASE-26679 Wait on the future returned by FanOutOneBlockAsyncDFSOutput.flush would stuck (#4039)

Signed-off-by: Duo Zhang <zhangduo@apache.org>

hbase-asyncfs/src/main/java/org/apache/hadoop/hbase/io/asyncfs/FanOutOneBlockAsyncDFSOutput.java

@@ -26,6 +26,7 @@ import static org.apache.hbase.thirdparty.io.netty.handler.timeout.IdleState.REA
 import static org.apache.hbase.thirdparty.io.netty.handler.timeout.IdleState.WRITER_IDLE;
 import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_CLIENT_SOCKET_TIMEOUT_KEY;
 
+import com.google.errorprone.annotations.RestrictedApi;
 import java.io.IOException;
 import java.io.InterruptedIOException;
 import java.nio.ByteBuffer;
@@ -232,7 +233,11 @@ public class FanOutOneBlockAsyncDFSOutput implements AsyncFSOutput {
   // so that the implementation will not burn up our brain as there are multiple state changes and
   // checks.
   private synchronized void failed(Channel channel, Supplier<Throwable> errorSupplier) {
-    if (state == State.BROKEN || state == State.CLOSED) {
+    if (state == State.CLOSED) {
+      return;
+    }
+    if (state == State.BROKEN) {
+      failWaitingAckQueue(channel, errorSupplier);
       return;
     }
     if (state == State.CLOSING) {
@@ -244,6 +249,11 @@ public class FanOutOneBlockAsyncDFSOutput implements AsyncFSOutput {
     }
     // disable further write, and fail all pending ack.
     state = State.BROKEN;
+    failWaitingAckQueue(channel, errorSupplier);
+    datanodeInfoMap.keySet().forEach(ChannelOutboundInvoker::close);
+  }
+
+  private void failWaitingAckQueue(Channel channel, Supplier<Throwable> errorSupplier) {
     Throwable error = errorSupplier.get();
     for (Iterator<Callback> iter = waitingAckQueue.iterator(); iter.hasNext();) {
       Callback c = iter.next();
@@ -260,7 +270,6 @@ public class FanOutOneBlockAsyncDFSOutput implements AsyncFSOutput {
       }
       break;
     }
-    datanodeInfoMap.keySet().forEach(ChannelOutboundInvoker::close);
   }
 
   @Sharable
@@ -599,4 +608,10 @@ public class FanOutOneBlockAsyncDFSOutput implements AsyncFSOutput {
   public long getSyncedLength() {
     return this.ackedBlockLength;
   }
+
+  @RestrictedApi(explanation = "Should only be called in tests", link = "",
+      allowedOnPath = ".*/src/test/.*")
+  Map<Channel, DatanodeInfo> getDatanodeInfoMap() {
+    return this.datanodeInfoMap;
+  }
 }

hbase-asyncfs/src/test/java/org/apache/hadoop/hbase/io/asyncfs/TestFanOutOneBlockAsyncDFSOutputHang.java

@@ -0,0 +1,249 @@
+/**
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.hadoop.hbase.io.asyncfs;
+
+
+import static org.junit.Assert.assertTrue;
+import static org.junit.Assert.fail;
+
+import java.io.IOException;
+import java.util.ArrayList;
+import java.util.Iterator;
+import java.util.List;
+import java.util.Map;
+import java.util.concurrent.CompletableFuture;
+import java.util.concurrent.CyclicBarrier;
+import java.util.concurrent.ExecutionException;
+import java.util.concurrent.ThreadLocalRandom;
+import org.apache.hadoop.fs.Path;
+import org.apache.hadoop.hbase.HBaseClassTestRule;
+import org.apache.hadoop.hbase.io.asyncfs.monitor.StreamSlowMonitor;
+import org.apache.hadoop.hbase.testclassification.MediumTests;
+import org.apache.hadoop.hbase.testclassification.MiscTests;
+import org.apache.hadoop.hdfs.DistributedFileSystem;
+import org.apache.hadoop.hdfs.MiniDFSCluster.DataNodeProperties;
+import org.apache.hadoop.hdfs.protocol.DatanodeInfo;
+import org.apache.hadoop.hdfs.server.datanode.DataNode;
+import org.junit.AfterClass;
+import org.junit.BeforeClass;
+import org.junit.ClassRule;
+import org.junit.Rule;
+import org.junit.Test;
+import org.junit.experimental.categories.Category;
+import org.junit.rules.TestName;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+import org.apache.hbase.thirdparty.io.netty.buffer.ByteBuf;
+import org.apache.hbase.thirdparty.io.netty.channel.Channel;
+import org.apache.hbase.thirdparty.io.netty.channel.ChannelHandlerContext;
+import org.apache.hbase.thirdparty.io.netty.channel.ChannelInboundHandlerAdapter;
+import org.apache.hbase.thirdparty.io.netty.channel.EventLoop;
+import org.apache.hbase.thirdparty.io.netty.channel.EventLoopGroup;
+import org.apache.hbase.thirdparty.io.netty.channel.nio.NioEventLoopGroup;
+import org.apache.hbase.thirdparty.io.netty.channel.socket.nio.NioSocketChannel;
+
+
+/**
+ * Testcase for HBASE-26679, here we introduce a separate test class and not put the testcase in
+ * {@link TestFanOutOneBlockAsyncDFSOutput} because we will send heartbeat to DN when there is no
+ * out going packet, the timeout is controlled by
+ * {@link TestFanOutOneBlockAsyncDFSOutput#READ_TIMEOUT_MS},which is 2 seconds, it will keep sending
+ * package out and DN will respond immedately and then mess up the testing handler added by us. So
+ * in this test class we use the default value for timeout which is 60 seconds and it is enough for
+ * this test.
+ */
+@Category({ MiscTests.class, MediumTests.class })
+public class TestFanOutOneBlockAsyncDFSOutputHang extends AsyncFSTestBase {
+
+  @ClassRule
+  public static final HBaseClassTestRule CLASS_RULE =
+      HBaseClassTestRule.forClass(TestFanOutOneBlockAsyncDFSOutputHang.class);
+
+  private static final Logger LOG =
+      LoggerFactory.getLogger(TestFanOutOneBlockAsyncDFSOutputHang.class);
+
+  private static DistributedFileSystem FS;
+
+  private static EventLoopGroup EVENT_LOOP_GROUP;
+
+  private static Class<? extends Channel> CHANNEL_CLASS;
+
+  private static StreamSlowMonitor MONITOR;
+
+  private static FanOutOneBlockAsyncDFSOutput OUT;
+
+  @Rule
+  public TestName name = new TestName();
+
+  @BeforeClass
+  public static void setUp() throws Exception {
+    startMiniDFSCluster(2);
+    FS = CLUSTER.getFileSystem();
+    EVENT_LOOP_GROUP = new NioEventLoopGroup();
+    CHANNEL_CLASS = NioSocketChannel.class;
+    MONITOR = StreamSlowMonitor.create(UTIL.getConfiguration(), "testMonitor");
+    Path f = new Path("/testHang");
+    EventLoop eventLoop = EVENT_LOOP_GROUP.next();
+    OUT = FanOutOneBlockAsyncDFSOutputHelper.createOutput(FS, f, true, false, (short) 2,
+      FS.getDefaultBlockSize(), eventLoop, CHANNEL_CLASS, MONITOR);
+  }
+
+  @AfterClass
+  public static void tearDown() throws IOException, InterruptedException {
+    if (OUT != null) {
+      OUT.recoverAndClose(null);
+    }
+    if (EVENT_LOOP_GROUP != null) {
+      EVENT_LOOP_GROUP.shutdownGracefully().sync();
+    }
+    shutdownMiniDFSCluster();
+  }
+
+  /**
+   * <pre>
+   * This test is for HBASE-26679. Consider there are two dataNodes: dn1 and dn2,dn2 is a slow DN.
+   * The threads sequence before HBASE-26679 is:
+   * 1.We write some data to {@link FanOutOneBlockAsyncDFSOutput} and then flush it, there are one
+   *   {@link FanOutOneBlockAsyncDFSOutput.Callback} in
+   *   {@link FanOutOneBlockAsyncDFSOutput#waitingAckQueue}.
+   * 2.The ack from dn1 arrives firstly and triggers Netty to invoke
+   *   {@link FanOutOneBlockAsyncDFSOutput#completed} with dn1's channel, then in
+   *   {@link FanOutOneBlockAsyncDFSOutput#completed}, dn1's channel is removed from
+   *   {@link FanOutOneBlockAsyncDFSOutput.Callback#unfinishedReplicas}.
+   * 3.But dn2 responds slowly, before dn2 sending ack,dn1 is shut down or have a exception,
+   *   so {@link FanOutOneBlockAsyncDFSOutput#failed} is triggered by Netty with dn1's channel,
+   *   and because the {@link FanOutOneBlockAsyncDFSOutput.Callback#unfinishedReplicas} does not
+   *   contain dn1's channel,the {@link FanOutOneBlockAsyncDFSOutput.Callback} is skipped in
+   *   {@link FanOutOneBlockAsyncDFSOutput#failed} method,and
+   *   {@link FanOutOneBlockAsyncDFSOutput#state} is set to
+   *   {@link FanOutOneBlockAsyncDFSOutput.State#BROKEN},and dn1,dn2 are all closed at the end of
+   *   {@link FanOutOneBlockAsyncDFSOutput#failed}.
+   * 4.{@link FanOutOneBlockAsyncDFSOutput#failed} is triggered again by dn2 because it is closed,
+   *   but because {@link FanOutOneBlockAsyncDFSOutput#state} is already
+   *   {@link FanOutOneBlockAsyncDFSOutput.State#BROKEN},the whole
+   *   {@link FanOutOneBlockAsyncDFSOutput#failed} is skipped. So wait on the future
+   *   returned by {@link FanOutOneBlockAsyncDFSOutput#flush} would be stuck for ever.
+   * After HBASE-26679, for above step 4,even if the {@link FanOutOneBlockAsyncDFSOutput#state}
+   * is already {@link FanOutOneBlockAsyncDFSOutput.State#BROKEN}, we would still try to trigger
+   * {@link FanOutOneBlockAsyncDFSOutput.Callback#future}.
+   * </pre>
+   */
+  @Test
+  public void testFlushHangWhenOneDataNodeFailedBeforeOtherDataNodeAck() throws Exception {
+
+    DataNodeProperties firstDataNodeProperties = null;
+    try {
+
+      final CyclicBarrier dn1AckReceivedCyclicBarrier = new CyclicBarrier(2);
+      Map<Channel, DatanodeInfo> datanodeInfoMap = OUT.getDatanodeInfoMap();
+      Iterator<Map.Entry<Channel, DatanodeInfo>> iterator = datanodeInfoMap.entrySet().iterator();
+      assertTrue(iterator.hasNext());
+      Map.Entry<Channel, DatanodeInfo> dn1Entry = iterator.next();
+      Channel dn1Channel = dn1Entry.getKey();
+      DatanodeInfo dn1DatanodeInfo = dn1Entry.getValue();
+      final List<String> protobufDecoderNames = new ArrayList<String>();
+      dn1Channel.pipeline().forEach((entry) -> {
+        if (ProtobufDecoder.class.isInstance(entry.getValue())) {
+          protobufDecoderNames.add(entry.getKey());
+        }
+      });
+      assertTrue(protobufDecoderNames.size() == 1);
+      dn1Channel.pipeline().addAfter(protobufDecoderNames.get(0), "dn1AckReceivedHandler",
+        new ChannelInboundHandlerAdapter() {
+          @Override
+          public void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {
+            super.channelRead(ctx, msg);
+            dn1AckReceivedCyclicBarrier.await();
+          }
+        });
+
+      assertTrue(iterator.hasNext());
+      Map.Entry<Channel, DatanodeInfo> dn2Entry = iterator.next();
+      Channel dn2Channel = dn2Entry.getKey();
+
+      /**
+       * Here we add a {@link ChannelInboundHandlerAdapter} to eat all the responses to simulate a
+       * slow dn2.
+       */
+      dn2Channel.pipeline().addFirst(new ChannelInboundHandlerAdapter() {
+
+        @Override
+        public void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {
+          if (!(msg instanceof ByteBuf)) {
+            ctx.fireChannelRead(msg);
+          }
+        }
+      });
+
+      byte[] b = new byte[10];
+      ThreadLocalRandom.current().nextBytes(b);
+      OUT.write(b, 0, b.length);
+      CompletableFuture<Long> future = OUT.flush(false);
+      /**
+       * Wait for ack from dn1.
+       */
+      dn1AckReceivedCyclicBarrier.await();
+      /**
+       * First ack is received from dn1,we could stop dn1 now.
+       */
+      firstDataNodeProperties = findAndKillFirstDataNode(dn1DatanodeInfo);
+      assertTrue(firstDataNodeProperties != null);
+      try {
+        /**
+         * Before HBASE-26679,here we should be stuck, after HBASE-26679,we would fail soon with
+         * {@link ExecutionException}.
+         */
+        future.get();
+        fail();
+      } catch (ExecutionException e) {
+        assertTrue(e != null);
+        LOG.info("expected exception caught when get future", e);
+      }
+      /**
+       * Make sure all the data node channel are closed.
+       */
+      datanodeInfoMap.keySet().forEach(ch -> {
+        try {
+          ch.closeFuture().get();
+        } catch (InterruptedException | ExecutionException e) {
+          throw new RuntimeException(e);
+        }
+      });
+    } finally {
+      if (firstDataNodeProperties != null) {
+        CLUSTER.restartDataNode(firstDataNodeProperties);
+      }
+    }
+  }
+
+  private static DataNodeProperties findAndKillFirstDataNode(DatanodeInfo firstDatanodeInfo) {
+    assertTrue(firstDatanodeInfo != null);
+    ArrayList<DataNode> dataNodes = CLUSTER.getDataNodes();
+    ArrayList<Integer> foundIndexes = new ArrayList<Integer>();
+    int index = 0;
+    for (DataNode dataNode : dataNodes) {
+      if (firstDatanodeInfo.getXferAddr().equals(dataNode.getDatanodeId().getXferAddr())) {
+        foundIndexes.add(index);
+      }
+      index++;
+    }
+    assertTrue(foundIndexes.size() == 1);
+    return CLUSTER.stopDataNode(foundIndexes.get(0));
+  }
+
+}