/// <summary>
/// Test when a block's replica is removed from RBW folder in one of the
/// datanode, namenode should ask to invalidate that corrupted block and
/// schedule replication for one more replica for that under replicated block.
/// </summary>
/// <exception cref="System.IO.IOException"/>
/// <exception cref="System.Exception"/>
public virtual void TestBlockInvalidationWhenRBWReplicaMissedInDN()
{
// This test cannot pass on Windows due to file locking enforcement. It will
// reject the attempt to delete the block file from the RBW folder.
Assume.AssumeTrue(!Path.Windows);
Configuration conf = new HdfsConfiguration();
conf.SetInt(DFSConfigKeys.DfsReplicationKey, 2);
conf.SetLong(DFSConfigKeys.DfsBlockreportIntervalMsecKey, 300);
conf.SetLong(DFSConfigKeys.DfsDatanodeDirectoryscanIntervalKey, 1);
conf.SetLong(DFSConfigKeys.DfsHeartbeatIntervalKey, 1);
MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(2).Build();
FSDataOutputStream @out = null;
try
{
FSNamesystem namesystem = cluster.GetNamesystem();
FileSystem fs = cluster.GetFileSystem();
Path testPath = new Path("/tmp/TestRBWBlockInvalidation", "foo1");
@out = fs.Create(testPath, (short)2);
@out.WriteBytes("HDFS-3157: " + testPath);
@out.Hsync();
cluster.StartDataNodes(conf, 1, true, null, null, null);
string bpid = namesystem.GetBlockPoolId();
ExtendedBlock blk = DFSTestUtil.GetFirstBlock(fs, testPath);
Block block = blk.GetLocalBlock();
DataNode dn = cluster.GetDataNodes()[0];
// Delete partial block and its meta information from the RBW folder
// of first datanode.
FilePath blockFile = DataNodeTestUtils.GetBlockFile(dn, bpid, block);
FilePath metaFile = DataNodeTestUtils.GetMetaFile(dn, bpid, block);
NUnit.Framework.Assert.IsTrue("Could not delete the block file from the RBW folder"
, blockFile.Delete());
NUnit.Framework.Assert.IsTrue("Could not delete the block meta file from the RBW folder"
, metaFile.Delete());
@out.Close();
int liveReplicas = 0;
while (true)
{
if ((liveReplicas = CountReplicas(namesystem, blk).LiveReplicas()) < 2)
{
// This confirms we have a corrupt replica
Log.Info("Live Replicas after corruption: " + liveReplicas);
break;
}
Sharpen.Thread.Sleep(100);
}
NUnit.Framework.Assert.AreEqual("There should be less than 2 replicas in the " +
"liveReplicasMap", 1, liveReplicas);
while (true)
{
if ((liveReplicas = CountReplicas(namesystem, blk).LiveReplicas()) > 1)
{
//Wait till the live replica count becomes equal to Replication Factor
Log.Info("Live Replicas after Rereplication: " + liveReplicas);
break;
}
Sharpen.Thread.Sleep(100);
}
NUnit.Framework.Assert.AreEqual("There should be two live replicas", 2, liveReplicas
);
while (true)
{
Sharpen.Thread.Sleep(100);
if (CountReplicas(namesystem, blk).CorruptReplicas() == 0)
{
Log.Info("Corrupt Replicas becomes 0");
break;
}
}
}
finally
{
if (@out != null)
{
@out.Close();
}
cluster.Shutdown();
}
}