You are viewing a plain text version of this content. The canonical link for it is here.
Posted to common-commits@hadoop.apache.org by we...@apache.org on 2020/12/15 04:42:05 UTC
[hadoop] branch branch-3.3 updated: HDFS-15170. EC: Block gets
marked as CORRUPT in case of failover and pipeline recovery. Contributed by
Ayush Saxena.
This is an automated email from the ASF dual-hosted git repository.
weichiu pushed a commit to branch branch-3.3
in repository https://gitbox.apache.org/repos/asf/hadoop.git
The following commit(s) were added to refs/heads/branch-3.3 by this push:
new faf26f7 HDFS-15170. EC: Block gets marked as CORRUPT in case of failover and pipeline recovery. Contributed by Ayush Saxena.
faf26f7 is described below
commit faf26f70aaa628b750ea5ad2eb0d68e0fdbc55d5
Author: Ayush Saxena <ay...@apache.org>
AuthorDate: Mon Dec 14 20:40:00 2020 -0800
HDFS-15170. EC: Block gets marked as CORRUPT in case of failover and pipeline recovery. Contributed by Ayush Saxena.
Signed-off-by: Wei-Chiu Chuang <we...@apache.org>
(cherry picked from commit 5e5c53070f1946b86e49c039e1f5ae42e5d022cb)
---
.../hdfs/server/blockmanagement/BlockManager.java | 10 +++
.../TestErasureCodingCorruption.java | 91 ++++++++++++++++++++++
2 files changed, 101 insertions(+)
diff --git a/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/BlockManager.java b/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/BlockManager.java
index 6264516..933f821 100644
--- a/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/BlockManager.java
+++ b/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/BlockManager.java
@@ -1856,6 +1856,16 @@ public class BlockManager implements BlockStatsMXBean {
// In case of 3, rbw block will be deleted and valid block can be replicated
if (hasEnoughLiveReplicas || hasMoreCorruptReplicas
|| corruptedDuringWrite) {
+ if (b.getStored().isStriped()) {
+ // If the block is an EC block, the whole block group is marked
+ // corrupted, so if this block is getting deleted, remove the block
+ // from corrupt replica map explicitly, since removal of the
+ // block from corrupt replicas may be delayed if the blocks are on
+ // stale storage due to failover or any other reason.
+ corruptReplicas.removeFromCorruptReplicasMap(b.getStored(), node);
+ BlockInfoStriped blk = (BlockInfoStriped) getStoredBlock(b.getStored());
+ blk.removeStorage(storageInfo);
+ }
// the block is over-replicated so invalidate the replicas immediately
invalidateBlock(b, node, numberOfReplicas);
} else if (isPopulatingReplQueues()) {
diff --git a/hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/blockmanagement/TestErasureCodingCorruption.java b/hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/blockmanagement/TestErasureCodingCorruption.java
new file mode 100644
index 0000000..cb1ef2b
--- /dev/null
+++ b/hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/blockmanagement/TestErasureCodingCorruption.java
@@ -0,0 +1,91 @@
+/**
+ * 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.hdfs.server.blockmanagement;
+
+import org.apache.hadoop.conf.Configuration;
+import org.apache.hadoop.fs.FSDataOutputStream;
+import org.apache.hadoop.fs.Path;
+import org.apache.hadoop.hdfs.DistributedFileSystem;
+import org.apache.hadoop.hdfs.MiniDFSCluster;
+import org.apache.hadoop.hdfs.MiniDFSNNTopology;
+import org.apache.hadoop.test.GenericTestUtils;
+import org.junit.Test;
+
+import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_NAMENODE_CORRUPT_BLOCK_DELETE_IMMEDIATELY_ENABLED;
+
+/**
+ * In case of Erasure Coding the entire block group is marked corrupted, in
+ * case there is a failure during write, the replica when reported to
+ * Namenode marks the block group as corrupt, but since the file is complete
+ * and the replica has a lower gen stamp, the replica is marked to be
+ * deleted, but in case the replica is on stale storage(Namenode marks
+ * datanodes as stale in case of failover, etc), then the deletion of block
+ * is postponed and hence the block group isn't removed from the corrupt
+ * replica map. So, in case of striped blocks, if the replica is suppose to
+ * get deleted, explicitly remove the block group from corrupt-replica map.
+ */
+public class TestErasureCodingCorruption {
+
+ @Test
+ public void testCorruptionDuringFailover() throws Exception {
+ Configuration conf = new Configuration();
+ // Set removal of corrupt replicas immediately as false, to trigger this
+ // case.
+ conf.setBoolean(DFS_NAMENODE_CORRUPT_BLOCK_DELETE_IMMEDIATELY_ENABLED,
+ false);
+ try (MiniDFSCluster cluster = new MiniDFSCluster
+ .Builder(conf)
+ .nnTopology(MiniDFSNNTopology.simpleHATopology())
+ .numDataNodes(8)
+ .build()) {
+ cluster.transitionToActive(0);
+ cluster.waitActive();
+
+ DistributedFileSystem dfs = cluster.getFileSystem(0);
+ dfs.mkdirs(new Path("/dir"));
+ dfs.setErasureCodingPolicy(new Path("/dir"), "RS-6-3-1024k");
+
+ FSDataOutputStream out = dfs.create(new Path("/dir/file"));
+ // Write more than one stripe, so that data can get flushed to all
+ // datanodes.
+ for (int i = 0; i < 15 * 1024 * 1024; i++) {
+ out.write(i);
+ }
+
+ // Stop one datanode, so as to trigger update pipeline.
+ MiniDFSCluster.DataNodeProperties dn = cluster.stopDataNode(0);
+ // Write some more data and close the file.
+ for (int i = 0; i < 7 * 1024 * 1024; i++) {
+ out.write(i);
+ }
+ out.close();
+
+ BlockManager bm = cluster.getNamesystem(0).getBlockManager();
+
+ // Transition to standby and then to active.
+ cluster.transitionToStandby(0);
+ cluster.transitionToActive(0);
+
+ // Restart the stopped Datanode, this datanode would report a replica
+ // that failed during write.
+ cluster.restartDataNode(dn);
+ GenericTestUtils
+ .waitFor(() -> bm.getCorruptECBlockGroups() == 0, 100, 10000);
+ }
+ }
+}
\ No newline at end of file
---------------------------------------------------------------------
To unsubscribe, e-mail: common-commits-unsubscribe@hadoop.apache.org
For additional commands, e-mail: common-commits-help@hadoop.apache.org