public virtual void TestFailoverWithBK()
{
MiniDFSCluster cluster = null;
try
{
Configuration conf = new Configuration();
conf.SetInt(DFSConfigKeys.DfsHaTaileditsPeriodKey, 1);
conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, BKJMUtil.CreateJournalURI("/hotfailover"
).ToString());
BKJMUtil.AddJournalManagerDefinition(conf);
cluster = new MiniDFSCluster.Builder(conf).NnTopology(MiniDFSNNTopology.SimpleHATopology
()).NumDataNodes(0).ManageNameDfsSharedDirs(false).Build();
NameNode nn1 = cluster.GetNameNode(0);
NameNode nn2 = cluster.GetNameNode(1);
cluster.WaitActive();
cluster.TransitionToActive(0);
Path p = new Path("/testBKJMfailover");
FileSystem fs = HATestUtil.ConfigureFailoverFs(cluster, conf);
fs.Mkdirs(p);
cluster.ShutdownNameNode(0);
cluster.TransitionToActive(1);
NUnit.Framework.Assert.IsTrue(fs.Exists(p));
}
finally
{
if (cluster != null)
{
cluster.Shutdown();
}
}
}
public virtual void TestInitializeBKSharedEdits()
{
MiniDFSCluster cluster = null;
try
{
Configuration conf = new Configuration();
HAUtil.SetAllowStandbyReads(conf, true);
conf.SetInt(DFSConfigKeys.DfsHaTaileditsPeriodKey, 1);
MiniDFSNNTopology topology = MiniDFSNNTopology.SimpleHATopology();
cluster = new MiniDFSCluster.Builder(conf).NnTopology(topology).NumDataNodes(0).Build
();
cluster.WaitActive();
// Shutdown and clear the current filebased shared dir.
cluster.ShutdownNameNodes();
FilePath shareddir = new FilePath(cluster.GetSharedEditsDir(0, 1));
NUnit.Framework.Assert.IsTrue("Initial Shared edits dir not fully deleted", FileUtil
.FullyDelete(shareddir));
// Check namenodes should not start without shared dir.
AssertCanNotStartNamenode(cluster, 0);
AssertCanNotStartNamenode(cluster, 1);
// Configure bkjm as new shared edits dir in both namenodes
Configuration nn1Conf = cluster.GetConfiguration(0);
Configuration nn2Conf = cluster.GetConfiguration(1);
nn1Conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, BKJMUtil.CreateJournalURI
("/initializeSharedEdits").ToString());
nn2Conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, BKJMUtil.CreateJournalURI
("/initializeSharedEdits").ToString());
BKJMUtil.AddJournalManagerDefinition(nn1Conf);
BKJMUtil.AddJournalManagerDefinition(nn2Conf);
// Initialize the BKJM shared edits.
NUnit.Framework.Assert.IsFalse(NameNode.InitializeSharedEdits(nn1Conf));
// NameNode should be able to start and should be in sync with BKJM as
// shared dir
AssertCanStartHANameNodes(cluster, conf, "/testBKJMInitialize");
}
finally
{
if (cluster != null)
{
cluster.Shutdown();
}
}
}
public override void SetupCluster()
{
Configuration conf = SetupCommonConfig();
conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, BKJMUtil.CreateJournalURI("/checkpointing"
+ journalCount++).ToString());
BKJMUtil.AddJournalManagerDefinition(conf);
MiniDFSNNTopology topology = new MiniDFSNNTopology().AddNameservice(new MiniDFSNNTopology.NSConf
("ns1").AddNN(new MiniDFSNNTopology.NNConf("nn1").SetHttpPort(10001)).AddNN(new
MiniDFSNNTopology.NNConf("nn2").SetHttpPort(10002)));
cluster = new MiniDFSCluster.Builder(conf).NnTopology(topology).NumDataNodes(1).ManageNameDfsSharedDirs
(false).Build();
cluster.WaitActive();
nn0 = cluster.GetNameNode(0);
nn1 = cluster.GetNameNode(1);
fs = HATestUtil.ConfigureFailoverFs(cluster, conf);
cluster.TransitionToActive(0);
}
public virtual void TestMultiplePrimariesStarted()
{
Path p1 = new Path("/testBKJMMultiplePrimary");
MiniDFSCluster cluster = null;
try
{
Configuration conf = new Configuration();
conf.SetInt(DFSConfigKeys.DfsHaTaileditsPeriodKey, 1);
conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, BKJMUtil.CreateJournalURI("/hotfailoverMultiple"
).ToString());
BKJMUtil.AddJournalManagerDefinition(conf);
cluster = new MiniDFSCluster.Builder(conf).NnTopology(MiniDFSNNTopology.SimpleHATopology
()).NumDataNodes(0).ManageNameDfsSharedDirs(false).CheckExitOnShutdown(false).Build
();
NameNode nn1 = cluster.GetNameNode(0);
NameNode nn2 = cluster.GetNameNode(1);
cluster.WaitActive();
cluster.TransitionToActive(0);
FileSystem fs = HATestUtil.ConfigureFailoverFs(cluster, conf);
fs.Mkdirs(p1);
nn1.GetRpcServer().RollEditLog();
cluster.TransitionToActive(1);
fs = cluster.GetFileSystem(0);
// get the older active server.
try
{
fs.Delete(p1, true);
NUnit.Framework.Assert.Fail("Log update on older active should cause it to exit");
}
catch (RemoteException re)
{
NUnit.Framework.Assert.IsTrue(re.GetClassName().Contains("ExitException"));
}
}
finally
{
if (cluster != null)
{
cluster.Shutdown();
}
}
}
public virtual void SetUp()
{
Configuration conf = new Configuration();
conf.SetInt(DFSConfigKeys.DfsNamenodeCheckpointCheckPeriodKey, 1);
conf.SetInt(DFSConfigKeys.DfsNamenodeCheckpointTxnsKey, 5);
conf.SetInt(DFSConfigKeys.DfsHaTaileditsPeriodKey, 1);
conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, BKJMUtil.CreateJournalURI("/bootstrapStandby"
).ToString());
BKJMUtil.AddJournalManagerDefinition(conf);
conf.SetBoolean(DFSConfigKeys.DfsImageCompressKey, true);
conf.Set(DFSConfigKeys.DfsImageCompressionCodecKey, typeof(TestStandbyCheckpoints.SlowCodec
).GetCanonicalName());
CompressionCodecFactory.SetCodecClasses(conf, ImmutableList.Of <Type>(typeof(TestStandbyCheckpoints.SlowCodec
)));
MiniDFSNNTopology topology = new MiniDFSNNTopology().AddNameservice(new MiniDFSNNTopology.NSConf
("ns1").AddNN(new MiniDFSNNTopology.NNConf("nn1").SetHttpPort(10001)).AddNN(new
MiniDFSNNTopology.NNConf("nn2").SetHttpPort(10002)));
cluster = new MiniDFSCluster.Builder(conf).NnTopology(topology).NumDataNodes(1).ManageNameDfsSharedDirs
(false).Build();
cluster.WaitActive();
}
public virtual void TestNameNodeMultipleSwitchesUsingBKJM()
{
MiniDFSCluster cluster = null;
try
{
Configuration conf = new Configuration();
conf.SetInt(DFSConfigKeys.DfsHaTaileditsPeriodKey, 1);
conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, BKJMUtil.CreateJournalURI("/correctEditLogSelection"
).ToString());
BKJMUtil.AddJournalManagerDefinition(conf);
cluster = new MiniDFSCluster.Builder(conf).NnTopology(MiniDFSNNTopology.SimpleHATopology
()).NumDataNodes(0).ManageNameDfsSharedDirs(false).Build();
NameNode nn1 = cluster.GetNameNode(0);
NameNode nn2 = cluster.GetNameNode(1);
cluster.WaitActive();
cluster.TransitionToActive(0);
nn1.GetRpcServer().RollEditLog();
// Roll Edits from current Active.
// Transition to standby current active gracefully.
cluster.TransitionToStandby(0);
// Make the other Active and Roll edits multiple times
cluster.TransitionToActive(1);
nn2.GetRpcServer().RollEditLog();
nn2.GetRpcServer().RollEditLog();
// Now One more failover. So NN1 should be able to failover successfully.
cluster.TransitionToStandby(1);
cluster.TransitionToActive(0);
}
finally
{
if (cluster != null)
{
cluster.Shutdown();
}
}
}
public virtual void TestFailoverWithFailingBKCluster()
{
int ensembleSize = numBookies + 1;
BookieServer newBookie = bkutil.NewBookie();
NUnit.Framework.Assert.AreEqual("New bookie didn't start", ensembleSize, bkutil.CheckBookiesUp
(ensembleSize, 10));
BookieServer replacementBookie = null;
MiniDFSCluster cluster = null;
try
{
Configuration conf = new Configuration();
conf.SetInt(DFSConfigKeys.DfsHaTaileditsPeriodKey, 1);
conf.Set(DFSConfigKeys.DfsNamenodeSharedEditsDirKey, BKJMUtil.CreateJournalURI("/hotfailoverWithFail"
).ToString());
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperEnsembleSize, ensembleSize);
conf.SetInt(BookKeeperJournalManager.BkjmBookkeeperQuorumSize, ensembleSize);
BKJMUtil.AddJournalManagerDefinition(conf);
cluster = new MiniDFSCluster.Builder(conf).NnTopology(MiniDFSNNTopology.SimpleHATopology
()).NumDataNodes(0).ManageNameDfsSharedDirs(false).CheckExitOnShutdown(false).Build
();
NameNode nn1 = cluster.GetNameNode(0);
NameNode nn2 = cluster.GetNameNode(1);
cluster.WaitActive();
cluster.TransitionToActive(0);
Path p1 = new Path("/testBKJMFailingBKCluster1");
Path p2 = new Path("/testBKJMFailingBKCluster2");
FileSystem fs = HATestUtil.ConfigureFailoverFs(cluster, conf);
fs.Mkdirs(p1);
newBookie.Shutdown();
// will take down shared storage
NUnit.Framework.Assert.AreEqual("New bookie didn't stop", numBookies, bkutil.CheckBookiesUp
(numBookies, 10));
try
{
fs.Mkdirs(p2);
NUnit.Framework.Assert.Fail("mkdirs should result in the NN exiting");
}
catch (RemoteException re)
{
NUnit.Framework.Assert.IsTrue(re.GetClassName().Contains("ExitException"));
}
cluster.ShutdownNameNode(0);
try
{
cluster.TransitionToActive(1);
NUnit.Framework.Assert.Fail("Shouldn't have been able to transition with bookies down"
);
}
catch (ExitUtil.ExitException ee)
{
NUnit.Framework.Assert.IsTrue("Should shutdown due to required journal failure",
ee.Message.Contains("starting log segment 3 failed for required journal"));
}
replacementBookie = bkutil.NewBookie();
NUnit.Framework.Assert.AreEqual("Replacement bookie didn't start", ensembleSize,
bkutil.CheckBookiesUp(ensembleSize, 10));
cluster.TransitionToActive(1);
// should work fine now
NUnit.Framework.Assert.IsTrue(fs.Exists(p1));
NUnit.Framework.Assert.IsFalse(fs.Exists(p2));
}
finally
{
newBookie.Shutdown();
if (replacementBookie != null)
{
replacementBookie.Shutdown();
}
if (cluster != null)
{
cluster.Shutdown();
}
}
}