Apache
/
hadoop
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

HDFS-11486. Client close() should not fail fast if the last block is being decommissioned. Contributed by Wei-Chiu Chuang and Yiqun Lin. 

(cherry picked from commit ca3ede5fda)
This commit is contained in:
Masatake Iwasaki 2017-03-28 18:15:09 +09:00
parent 295aab8e3a
commit 0ece3c8dd9
2 changed files with 54 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt
View File

@ -232,6 +232,9 @@ Release 2.7.4 - UNRELEASED
HDFS-11499. Decommissioning stuck because of failing recovery. HDFS-11499. Decommissioning stuck because of failing recovery.
Contributed by Lukas Majercak and Manoj Govindassamy. Contributed by Lukas Majercak and Manoj Govindassamy.
HDFS-11486. Client close() should not fail fast if the last block is being
decommissioned. Contributed by Wei-Chiu Chuang and Yiqun Lin.
Release 2.7.3 - 2016-08-25 Release 2.7.3 - 2016-08-25
INCOMPATIBLE CHANGES INCOMPATIBLE CHANGES

51
hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestDecommission.java
View File

@ -917,6 +917,57 @@ public class TestDecommission {
assertEquals(dfs.getFileStatus(file).getLen(), writtenBytes); assertEquals(dfs.getFileStatus(file).getLen(), writtenBytes);
} }
@Test(timeout=120000)
public void testCloseWhileDecommission() throws IOException,
ExecutionException, InterruptedException {
LOG.info("Starting test testCloseWhileDecommission");
// min replication = 2
Configuration hdfsConf = new HdfsConfiguration(conf);
hdfsConf.setInt(DFSConfigKeys.DFS_NAMENODE_REPLICATION_MIN_KEY, 2);
startCluster(1, 3, hdfsConf);
FileSystem fileSys = cluster.getFileSystem(0);
FSNamesystem ns = cluster.getNamesystem(0);
String openFile = "/testDecommissionWithOpenfile.dat";
writeFile(fileSys, new Path(openFile), (short)3);
// make sure the file was open for write
FSDataOutputStream fdos = fileSys.append(new Path(openFile));
byte[] bytes = new byte[1];
fdos.write(bytes);
fdos.hsync();
LocatedBlocks lbs = NameNodeAdapter.getBlockLocations(
cluster.getNameNode(0), openFile, 0, fileSize);
DatanodeInfo[] dnInfos4LastBlock = lbs.getLastLocatedBlock().getLocations();
ArrayList<String> nodes = new ArrayList<String>();
ArrayList<DatanodeInfo> dnInfos = new ArrayList<DatanodeInfo>();
DatanodeManager dm = ns.getBlockManager().getDatanodeManager();
//decommission 2 of the 3 nodes which have last block
nodes.add(dnInfos4LastBlock[0].getXferAddr());
dnInfos.add(dm.getDatanode(dnInfos4LastBlock[0]));
nodes.add(dnInfos4LastBlock[1].getXferAddr());
dnInfos.add(dm.getDatanode(dnInfos4LastBlock[1]));
writeConfigFile(excludeFile, nodes);
// because the cluster has only 3 nodes, and 2 of which are decomm'ed,
// the last block file will remain under replicated.
refreshNodes(cluster.getNamesystem(0), cluster.getConfiguration(0));
// the close() should not fail despite the number of live replicas of
// the last block becomes one.
fdos.close();
// make sure the two datanodes remain in decomm in progress state
BlockManagerTestUtil.recheckDecommissionState(dm);
assertTrackedAndPending(dm.getDecomManager(), 2, 0);
}
/** /**
* Tests restart of namenode while datanode hosts are added to exclude file * Tests restart of namenode while datanode hosts are added to exclude file