public virtual void TestReduceReplFactorRespectsRackPolicy()
{
Configuration conf = GetConf();
short ReplicationFactor = 3;
Path filePath = new Path("/testFile");
string[] racks = new string[] { "/rack1", "/rack1", "/rack2", "/rack2" };
MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(racks.Length
).Racks(racks).Build();
FSNamesystem ns = cluster.GetNameNode().GetNamesystem();
try
{
// Create a file with one block
FileSystem fs = cluster.GetFileSystem();
DFSTestUtil.CreateFile(fs, filePath, 1L, ReplicationFactor, 1L);
ExtendedBlock b = DFSTestUtil.GetFirstBlock(fs, filePath);
DFSTestUtil.WaitForReplication(cluster, b, 2, ReplicationFactor, 0);
// Decrease the replication factor, make sure the deleted replica
// was not the one that lived on the rack with only one replica,
// ie we should still have 2 racks after reducing the repl factor.
ReplicationFactor = 2;
NameNodeAdapter.SetReplication(ns, "/testFile", ReplicationFactor);
DFSTestUtil.WaitForReplication(cluster, b, 2, ReplicationFactor, 0);
}
finally
{
cluster.Shutdown();
}
}
public virtual void TestSufficientlySingleReplBlockUsesNewRack()
{
Configuration conf = GetConf();
short ReplicationFactor = 1;
Path filePath = new Path("/testFile");
string[] racks = new string[] { "/rack1", "/rack1", "/rack1", "/rack2" };
MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(racks.Length
).Racks(racks).Build();
FSNamesystem ns = cluster.GetNameNode().GetNamesystem();
try
{
// Create a file with one block with a replication factor of 1
FileSystem fs = cluster.GetFileSystem();
DFSTestUtil.CreateFile(fs, filePath, 1L, ReplicationFactor, 1L);
ExtendedBlock b = DFSTestUtil.GetFirstBlock(fs, filePath);
DFSTestUtil.WaitForReplication(cluster, b, 1, ReplicationFactor, 0);
ReplicationFactor = 2;
NameNodeAdapter.SetReplication(ns, "/testFile", ReplicationFactor);
DFSTestUtil.WaitForReplication(cluster, b, 2, ReplicationFactor, 0);
}
finally
{
cluster.Shutdown();
}
}
public virtual void TestUnderReplicatedUsesNewRacks()
{
Configuration conf = GetConf();
short ReplicationFactor = 3;
Path filePath = new Path("/testFile");
// All datanodes are on the same rack
string[] racks = new string[] { "/rack1", "/rack1", "/rack1", "/rack1", "/rack1" };
MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(racks.Length
).Racks(racks).Build();
FSNamesystem ns = cluster.GetNameNode().GetNamesystem();
try
{
// Create a file with one block
FileSystem fs = cluster.GetFileSystem();
DFSTestUtil.CreateFile(fs, filePath, 1L, ReplicationFactor, 1L);
ExtendedBlock b = DFSTestUtil.GetFirstBlock(fs, filePath);
DFSTestUtil.WaitForReplication(cluster, b, 1, ReplicationFactor, 0);
// Add new datanodes on a different rack and increase the
// replication factor so the block is underreplicated and make
// sure at least one of the hosts on the new rack is used.
string[] newRacks = new string[] { "/rack2", "/rack2" };
cluster.StartDataNodes(conf, 2, true, null, newRacks);
ReplicationFactor = 5;
NameNodeAdapter.SetReplication(ns, "/testFile", ReplicationFactor);
DFSTestUtil.WaitForReplication(cluster, b, 2, ReplicationFactor, 0);
}
finally
{
cluster.Shutdown();
}
}
public virtual void TestExcessBlocks()
{
Path file = GetTestPath("testExcessBlocks");
CreateFile(file, 100, (short)2);
NameNodeAdapter.SetReplication(namesystem, file.ToString(), (short)1);
UpdateMetrics();
MetricsRecordBuilder rb = MetricsAsserts.GetMetrics(NsMetrics);
MetricsAsserts.AssertGauge("ExcessBlocks", 1L, rb);
// verify ExcessBlocks metric is decremented and
// excessReplicateMap is cleared after deleting a file
fs.Delete(file, true);
rb = MetricsAsserts.GetMetrics(NsMetrics);
MetricsAsserts.AssertGauge("ExcessBlocks", 0L, rb);
NUnit.Framework.Assert.IsTrue(bm.excessReplicateMap.IsEmpty());
}
public virtual void TestProcesOverReplicateBlock()
{
Configuration conf = new HdfsConfiguration();
conf.SetLong(DFSConfigKeys.DfsDatanodeScanPeriodHoursKey, 100L);
conf.SetLong(DFSConfigKeys.DfsBlockreportIntervalMsecKey, 1000L);
conf.Set(DFSConfigKeys.DfsNamenodeReplicationPendingTimeoutSecKey, Sharpen.Extensions.ToString
(2));
MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(3).Build();
FileSystem fs = cluster.GetFileSystem();
try
{
Path fileName = new Path("/foo1");
DFSTestUtil.CreateFile(fs, fileName, 2, (short)3, 0L);
DFSTestUtil.WaitReplication(fs, fileName, (short)3);
// corrupt the block on datanode 0
ExtendedBlock block = DFSTestUtil.GetFirstBlock(fs, fileName);
NUnit.Framework.Assert.IsTrue(cluster.CorruptReplica(0, block));
MiniDFSCluster.DataNodeProperties dnProps = cluster.StopDataNode(0);
// remove block scanner log to trigger block scanning
FilePath scanCursor = new FilePath(new FilePath(MiniDFSCluster.GetFinalizedDir(cluster
.GetInstanceStorageDir(0, 0), cluster.GetNamesystem().GetBlockPoolId()).GetParent
()).GetParent(), "scanner.cursor");
//wait for one minute for deletion to succeed;
for (int i = 0; !scanCursor.Delete(); i++)
{
NUnit.Framework.Assert.IsTrue("Could not delete " + scanCursor.GetAbsolutePath()
+ " in one minute", i < 60);
try
{
Sharpen.Thread.Sleep(1000);
}
catch (Exception)
{
}
}
// restart the datanode so the corrupt replica will be detected
cluster.RestartDataNode(dnProps);
DFSTestUtil.WaitReplication(fs, fileName, (short)2);
string blockPoolId = cluster.GetNamesystem().GetBlockPoolId();
DatanodeID corruptDataNode = DataNodeTestUtils.GetDNRegistrationForBP(cluster.GetDataNodes
()[2], blockPoolId);
FSNamesystem namesystem = cluster.GetNamesystem();
BlockManager bm = namesystem.GetBlockManager();
HeartbeatManager hm = bm.GetDatanodeManager().GetHeartbeatManager();
try
{
namesystem.WriteLock();
lock (hm)
{
// set live datanode's remaining space to be 0
// so they will be chosen to be deleted when over-replication occurs
string corruptMachineName = corruptDataNode.GetXferAddr();
foreach (DatanodeDescriptor datanode in hm.GetDatanodes())
{
if (!corruptMachineName.Equals(datanode.GetXferAddr()))
{
datanode.GetStorageInfos()[0].SetUtilizationForTesting(100L, 100L, 0, 100L);
datanode.UpdateHeartbeat(BlockManagerTestUtil.GetStorageReportsForDatanode(datanode
), 0L, 0L, 0, 0, null);
}
}
// decrease the replication factor to 1;
NameNodeAdapter.SetReplication(namesystem, fileName.ToString(), (short)1);
// corrupt one won't be chosen to be excess one
// without 4910 the number of live replicas would be 0: block gets lost
NUnit.Framework.Assert.AreEqual(1, bm.CountNodes(block.GetLocalBlock()).LiveReplicas
());
}
}
finally
{
namesystem.WriteUnlock();
}
}
finally
{
cluster.Shutdown();
}
}
