public virtual void TestChooseTargetWithDecomNodes()
{
namenode.GetNamesystem().WriteLock();
try
{
string blockPoolId = namenode.GetNamesystem().GetBlockPoolId();
dnManager.HandleHeartbeat(dnrList[3], BlockManagerTestUtil.GetStorageReportsForDatanode
(dataNodes[3]), blockPoolId, dataNodes[3].GetCacheCapacity(), dataNodes[3].GetCacheRemaining
(), 2, 0, 0, null);
dnManager.HandleHeartbeat(dnrList[4], BlockManagerTestUtil.GetStorageReportsForDatanode
(dataNodes[4]), blockPoolId, dataNodes[4].GetCacheCapacity(), dataNodes[4].GetCacheRemaining
(), 4, 0, 0, null);
dnManager.HandleHeartbeat(dnrList[5], BlockManagerTestUtil.GetStorageReportsForDatanode
(dataNodes[5]), blockPoolId, dataNodes[5].GetCacheCapacity(), dataNodes[5].GetCacheRemaining
(), 4, 0, 0, null);
// value in the above heartbeats
int load = 2 + 4 + 4;
FSNamesystem fsn = namenode.GetNamesystem();
NUnit.Framework.Assert.AreEqual((double)load / 6, dnManager.GetFSClusterStats().GetInServiceXceiverAverage
(), Epsilon);
// Decommission DNs so BlockPlacementPolicyDefault.isGoodTarget()
// returns false
for (int i = 0; i < 3; i++)
{
DatanodeDescriptor d = dnManager.GetDatanode(dnrList[i]);
dnManager.GetDecomManager().StartDecommission(d);
d.SetDecommissioned();
}
NUnit.Framework.Assert.AreEqual((double)load / 3, dnManager.GetFSClusterStats().GetInServiceXceiverAverage
(), Epsilon);
// update references of writer DN to update the de-commissioned state
IList <DatanodeDescriptor> liveNodes = new AList <DatanodeDescriptor>();
dnManager.FetchDatanodes(liveNodes, null, false);
DatanodeDescriptor writerDn = null;
if (liveNodes.Contains(dataNodes[0]))
{
writerDn = liveNodes[liveNodes.IndexOf(dataNodes[0])];
}
// Call chooseTarget()
DatanodeStorageInfo[] targets = namenode.GetNamesystem().GetBlockManager().GetBlockPlacementPolicy
().ChooseTarget("testFile.txt", 3, writerDn, new AList <DatanodeStorageInfo>(), false
, null, 1024, TestBlockStoragePolicy.DefaultStoragePolicy);
NUnit.Framework.Assert.AreEqual(3, targets.Length);
ICollection <DatanodeStorageInfo> targetSet = new HashSet <DatanodeStorageInfo>(Arrays
.AsList(targets));
for (int i_1 = 3; i_1 < storages.Length; i_1++)
{
NUnit.Framework.Assert.IsTrue(targetSet.Contains(storages[i_1]));
}
}
finally
{
dataNodes[0].StopDecommission();
dataNodes[1].StopDecommission();
dataNodes[2].StopDecommission();
namenode.GetNamesystem().WriteUnlock();
}
}
/// <summary>Have DatanodeManager check decommission state.</summary>
/// <param name="dm">the DatanodeManager to manipulate</param>
/// <exception cref="Sharpen.ExecutionException"/>
/// <exception cref="System.Exception"/>
public static void RecheckDecommissionState(DatanodeManager dm)
{
dm.GetDecomManager().RunMonitor();
}
public virtual void TestXceiverCount()
{
Configuration conf = new HdfsConfiguration();
// retry one time, if close fails
conf.SetInt(DFSConfigKeys.DfsClientBlockWriteLocatefollowingblockRetriesKey, 1);
MiniDFSCluster cluster = null;
int nodes = 8;
int fileCount = 5;
short fileRepl = 3;
try
{
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(nodes).Build();
cluster.WaitActive();
FSNamesystem namesystem = cluster.GetNamesystem();
DatanodeManager dnm = namesystem.GetBlockManager().GetDatanodeManager();
IList <DataNode> datanodes = cluster.GetDataNodes();
DistributedFileSystem fs = cluster.GetFileSystem();
// trigger heartbeats in case not already sent
TriggerHeartbeats(datanodes);
// check that all nodes are live and in service
int expectedTotalLoad = nodes;
// xceiver server adds 1 to load
int expectedInServiceNodes = nodes;
int expectedInServiceLoad = nodes;
CheckClusterHealth(nodes, namesystem, expectedTotalLoad, expectedInServiceNodes,
expectedInServiceLoad);
// shutdown half the nodes and force a heartbeat check to ensure
// counts are accurate
for (int i = 0; i < nodes / 2; i++)
{
DataNode dn = datanodes[i];
DatanodeDescriptor dnd = dnm.GetDatanode(dn.GetDatanodeId());
dn.Shutdown();
DFSTestUtil.SetDatanodeDead(dnd);
BlockManagerTestUtil.CheckHeartbeat(namesystem.GetBlockManager());
//Verify decommission of dead node won't impact nodesInService metrics.
dnm.GetDecomManager().StartDecommission(dnd);
expectedInServiceNodes--;
NUnit.Framework.Assert.AreEqual(expectedInServiceNodes, namesystem.GetNumLiveDataNodes
());
NUnit.Framework.Assert.AreEqual(expectedInServiceNodes, GetNumDNInService(namesystem
));
//Verify recommission of dead node won't impact nodesInService metrics.
dnm.GetDecomManager().StopDecommission(dnd);
NUnit.Framework.Assert.AreEqual(expectedInServiceNodes, GetNumDNInService(namesystem
));
}
// restart the nodes to verify that counts are correct after
// node re-registration
cluster.RestartDataNodes();
cluster.WaitActive();
datanodes = cluster.GetDataNodes();
expectedInServiceNodes = nodes;
NUnit.Framework.Assert.AreEqual(nodes, datanodes.Count);
CheckClusterHealth(nodes, namesystem, expectedTotalLoad, expectedInServiceNodes,
expectedInServiceLoad);
// create streams and hsync to force datastreamers to start
DFSOutputStream[] streams = new DFSOutputStream[fileCount];
for (int i_1 = 0; i_1 < fileCount; i_1++)
{
streams[i_1] = (DFSOutputStream)fs.Create(new Path("/f" + i_1), fileRepl).GetWrappedStream
();
streams[i_1].Write(Sharpen.Runtime.GetBytesForString("1"));
streams[i_1].Hsync();
// the load for writers is 2 because both the write xceiver & packet
// responder threads are counted in the load
expectedTotalLoad += 2 * fileRepl;
expectedInServiceLoad += 2 * fileRepl;
}
// force nodes to send load update
TriggerHeartbeats(datanodes);
CheckClusterHealth(nodes, namesystem, expectedTotalLoad, expectedInServiceNodes,
expectedInServiceLoad);
// decomm a few nodes, substract their load from the expected load,
// trigger heartbeat to force load update
for (int i_2 = 0; i_2 < fileRepl; i_2++)
{
expectedInServiceNodes--;
DatanodeDescriptor dnd = dnm.GetDatanode(datanodes[i_2].GetDatanodeId());
expectedInServiceLoad -= dnd.GetXceiverCount();
dnm.GetDecomManager().StartDecommission(dnd);
DataNodeTestUtils.TriggerHeartbeat(datanodes[i_2]);
Sharpen.Thread.Sleep(100);
CheckClusterHealth(nodes, namesystem, expectedTotalLoad, expectedInServiceNodes,
expectedInServiceLoad);
}
// check expected load while closing each stream. recalc expected
// load based on whether the nodes in the pipeline are decomm
for (int i_3 = 0; i_3 < fileCount; i_3++)
{
int decomm = 0;
foreach (DatanodeInfo dni in streams[i_3].GetPipeline())
{
DatanodeDescriptor dnd = dnm.GetDatanode(dni);
expectedTotalLoad -= 2;
if (dnd.IsDecommissionInProgress() || dnd.IsDecommissioned())
{
decomm++;
}
else
{
expectedInServiceLoad -= 2;
}
}
try
{
streams[i_3].Close();
}
catch (IOException ioe)
{
// nodes will go decommissioned even if there's a UC block whose
// other locations are decommissioned too. we'll ignore that
// bug for now
if (decomm < fileRepl)
{
throw;
}
}
TriggerHeartbeats(datanodes);
// verify node count and loads
CheckClusterHealth(nodes, namesystem, expectedTotalLoad, expectedInServiceNodes,
expectedInServiceLoad);
}
// shutdown each node, verify node counts based on decomm state
for (int i_4 = 0; i_4 < nodes; i_4++)
{
DataNode dn = datanodes[i_4];
dn.Shutdown();
// force it to appear dead so live count decreases
DatanodeDescriptor dnDesc = dnm.GetDatanode(dn.GetDatanodeId());
DFSTestUtil.SetDatanodeDead(dnDesc);
BlockManagerTestUtil.CheckHeartbeat(namesystem.GetBlockManager());
NUnit.Framework.Assert.AreEqual(nodes - 1 - i_4, namesystem.GetNumLiveDataNodes()
);
// first few nodes are already out of service
if (i_4 >= fileRepl)
{
expectedInServiceNodes--;
}
NUnit.Framework.Assert.AreEqual(expectedInServiceNodes, GetNumDNInService(namesystem
));
// live nodes always report load of 1. no nodes is load 0
double expectedXceiverAvg = (i_4 == nodes - 1) ? 0.0 : 1.0;
NUnit.Framework.Assert.AreEqual((double)expectedXceiverAvg, GetInServiceXceiverAverage
(namesystem), Epsilon);
}
// final sanity check
CheckClusterHealth(0, namesystem, 0.0, 0, 0.0);
}
finally
{
if (cluster != null)
{
cluster.Shutdown();
}
}
}