public virtual void TestFinalizeWithJournalNodes()
{
MiniQJMHACluster qjCluster = null;
FileSystem fs = null;
try
{
MiniQJMHACluster.Builder builder = new MiniQJMHACluster.Builder(conf);
builder.GetDfsBuilder().NumDataNodes(0);
qjCluster = builder.Build();
MiniDFSCluster cluster = qjCluster.GetDfsCluster();
// No upgrade is in progress at the moment.
CheckJnPreviousDirExistence(qjCluster, false);
CheckClusterPreviousDirExistence(cluster, false);
AssertCTimesEqual(cluster);
// Transition NN0 to active and do some FS ops.
cluster.TransitionToActive(0);
fs = HATestUtil.ConfigureFailoverFs(cluster, conf);
NUnit.Framework.Assert.IsTrue(fs.Mkdirs(new Path("/foo1")));
long cidBeforeUpgrade = GetCommittedTxnIdValue(qjCluster);
// Do the upgrade. Shut down NN1 and then restart NN0 with the upgrade
// flag.
cluster.ShutdownNameNode(1);
cluster.GetNameNodeInfos()[0].SetStartOpt(HdfsServerConstants.StartupOption.Upgrade
);
cluster.RestartNameNode(0, false);
NUnit.Framework.Assert.IsTrue(cidBeforeUpgrade <= GetCommittedTxnIdValue(qjCluster
));
NUnit.Framework.Assert.IsTrue(fs.Mkdirs(new Path("/foo2")));
CheckNnPreviousDirExistence(cluster, 0, true);
CheckNnPreviousDirExistence(cluster, 1, false);
CheckJnPreviousDirExistence(qjCluster, true);
// Now bootstrap the standby with the upgraded info.
int rc = BootstrapStandby.Run(new string[] { "-force" }, cluster.GetConfiguration
(1));
NUnit.Framework.Assert.AreEqual(0, rc);
cluster.RestartNameNode(1);
long cidDuringUpgrade = GetCommittedTxnIdValue(qjCluster);
NUnit.Framework.Assert.IsTrue(cidDuringUpgrade > cidBeforeUpgrade);
RunFinalizeCommand(cluster);
NUnit.Framework.Assert.AreEqual(cidDuringUpgrade, GetCommittedTxnIdValue(qjCluster
));
CheckClusterPreviousDirExistence(cluster, false);
CheckJnPreviousDirExistence(qjCluster, false);
AssertCTimesEqual(cluster);
}
finally
{
if (fs != null)
{
fs.Close();
}
if (qjCluster != null)
{
qjCluster.Shutdown();
}
}
}
/// <exception cref="System.IO.IOException"/>
private MiniQJMHACluster(MiniQJMHACluster.Builder builder)
{
this.conf = builder.conf;
int retryCount = 0;
while (true)
{
try
{
basePort = 10000 + Random.Next(1000) * 4;
// start 3 journal nodes
journalCluster = new MiniJournalCluster.Builder(conf).Format(true).Build();
URI journalURI = journalCluster.GetQuorumJournalURI(Nameservice);
// start cluster with 2 NameNodes
MiniDFSNNTopology topology = CreateDefaultTopology(basePort);
InitHAConf(journalURI, builder.conf);
// First start up the NNs just to format the namespace. The MinIDFSCluster
// has no way to just format the NameNodes without also starting them.
cluster = builder.dfsBuilder.NnTopology(topology).ManageNameDfsSharedDirs(false).
Build();
cluster.WaitActive();
cluster.ShutdownNameNodes();
// initialize the journal nodes
Configuration confNN0 = cluster.GetConfiguration(0);
NameNode.InitializeSharedEdits(confNN0, true);
cluster.GetNameNodeInfos()[0].SetStartOpt(builder.startOpt);
cluster.GetNameNodeInfos()[1].SetStartOpt(builder.startOpt);
// restart the cluster
cluster.RestartNameNodes();
++retryCount;
break;
}
catch (BindException)
{
Log.Info("MiniQJMHACluster port conflicts, retried " + retryCount + " times");
}
}
}
public virtual void TestRollbackWithJournalNodes()
{
MiniQJMHACluster qjCluster = null;
FileSystem fs = null;
try
{
MiniQJMHACluster.Builder builder = new MiniQJMHACluster.Builder(conf);
builder.GetDfsBuilder().NumDataNodes(0);
qjCluster = builder.Build();
MiniDFSCluster cluster = qjCluster.GetDfsCluster();
// No upgrade is in progress at the moment.
CheckClusterPreviousDirExistence(cluster, false);
AssertCTimesEqual(cluster);
CheckJnPreviousDirExistence(qjCluster, false);
// Transition NN0 to active and do some FS ops.
cluster.TransitionToActive(0);
fs = HATestUtil.ConfigureFailoverFs(cluster, conf);
NUnit.Framework.Assert.IsTrue(fs.Mkdirs(new Path("/foo1")));
long cidBeforeUpgrade = GetCommittedTxnIdValue(qjCluster);
// Do the upgrade. Shut down NN1 and then restart NN0 with the upgrade
// flag.
cluster.ShutdownNameNode(1);
cluster.GetNameNodeInfos()[0].SetStartOpt(HdfsServerConstants.StartupOption.Upgrade
);
cluster.RestartNameNode(0, false);
CheckNnPreviousDirExistence(cluster, 0, true);
CheckNnPreviousDirExistence(cluster, 1, false);
CheckJnPreviousDirExistence(qjCluster, true);
// NN0 should come up in the active state when given the -upgrade option,
// so no need to transition it to active.
NUnit.Framework.Assert.IsTrue(fs.Mkdirs(new Path("/foo2")));
long cidDuringUpgrade = GetCommittedTxnIdValue(qjCluster);
NUnit.Framework.Assert.IsTrue(cidDuringUpgrade > cidBeforeUpgrade);
// Now bootstrap the standby with the upgraded info.
int rc = BootstrapStandby.Run(new string[] { "-force" }, cluster.GetConfiguration
(1));
NUnit.Framework.Assert.AreEqual(0, rc);
cluster.RestartNameNode(1);
CheckNnPreviousDirExistence(cluster, 0, true);
CheckNnPreviousDirExistence(cluster, 1, true);
CheckJnPreviousDirExistence(qjCluster, true);
AssertCTimesEqual(cluster);
// Shut down the NNs, but deliberately leave the JNs up and running.
ICollection <URI> nn1NameDirs = cluster.GetNameDirs(0);
cluster.Shutdown();
conf.SetStrings(DFSConfigKeys.DfsNamenodeNameDirKey, Joiner.On(",").Join(nn1NameDirs
));
NameNode.DoRollback(conf, false);
long cidAfterRollback = GetCommittedTxnIdValue(qjCluster);
NUnit.Framework.Assert.IsTrue(cidBeforeUpgrade < cidAfterRollback);
// make sure the committedTxnId has been reset correctly after rollback
NUnit.Framework.Assert.IsTrue(cidDuringUpgrade > cidAfterRollback);
// The rollback operation should have rolled back the first NN's local
// dirs, and the shared dir, but not the other NN's dirs. Those have to be
// done by bootstrapping the standby.
CheckNnPreviousDirExistence(cluster, 0, false);
CheckJnPreviousDirExistence(qjCluster, false);
}
finally
{
if (fs != null)
{
fs.Close();
}
if (qjCluster != null)
{
qjCluster.Shutdown();
}
}
}
public virtual void TestUpgradeWithJournalNodes()
{
MiniQJMHACluster qjCluster = null;
FileSystem fs = null;
try
{
MiniQJMHACluster.Builder builder = new MiniQJMHACluster.Builder(conf);
builder.GetDfsBuilder().NumDataNodes(0);
qjCluster = builder.Build();
MiniDFSCluster cluster = qjCluster.GetDfsCluster();
// No upgrade is in progress at the moment.
CheckJnPreviousDirExistence(qjCluster, false);
CheckClusterPreviousDirExistence(cluster, false);
AssertCTimesEqual(cluster);
// Transition NN0 to active and do some FS ops.
cluster.TransitionToActive(0);
fs = HATestUtil.ConfigureFailoverFs(cluster, conf);
NUnit.Framework.Assert.IsTrue(fs.Mkdirs(new Path("/foo1")));
// get the value of the committedTxnId in journal nodes
long cidBeforeUpgrade = GetCommittedTxnIdValue(qjCluster);
// Do the upgrade. Shut down NN1 and then restart NN0 with the upgrade
// flag.
cluster.ShutdownNameNode(1);
cluster.GetNameNodeInfos()[0].SetStartOpt(HdfsServerConstants.StartupOption.Upgrade
);
cluster.RestartNameNode(0, false);
CheckNnPreviousDirExistence(cluster, 0, true);
CheckNnPreviousDirExistence(cluster, 1, false);
CheckJnPreviousDirExistence(qjCluster, true);
NUnit.Framework.Assert.IsTrue(cidBeforeUpgrade <= GetCommittedTxnIdValue(qjCluster
));
// NN0 should come up in the active state when given the -upgrade option,
// so no need to transition it to active.
NUnit.Framework.Assert.IsTrue(fs.Mkdirs(new Path("/foo2")));
// Restart NN0 without the -upgrade flag, to make sure that works.
cluster.GetNameNodeInfos()[0].SetStartOpt(HdfsServerConstants.StartupOption.Regular
);
cluster.RestartNameNode(0, false);
// Make sure we can still do FS ops after upgrading.
cluster.TransitionToActive(0);
NUnit.Framework.Assert.IsTrue(fs.Mkdirs(new Path("/foo3")));
NUnit.Framework.Assert.IsTrue(GetCommittedTxnIdValue(qjCluster) > cidBeforeUpgrade
);
// Now bootstrap the standby with the upgraded info.
int rc = BootstrapStandby.Run(new string[] { "-force" }, cluster.GetConfiguration
(1));
NUnit.Framework.Assert.AreEqual(0, rc);
// Now restart NN1 and make sure that we can do ops against that as well.
cluster.RestartNameNode(1);
cluster.TransitionToStandby(0);
cluster.TransitionToActive(1);
NUnit.Framework.Assert.IsTrue(fs.Mkdirs(new Path("/foo4")));
AssertCTimesEqual(cluster);
}
finally
{
if (fs != null)
{
fs.Close();
}
if (qjCluster != null)
{
qjCluster.Shutdown();
}
}
}