public virtual void TestDisplayRecentEditLogOpCodes()
{
// start a cluster
Configuration conf = new HdfsConfiguration();
MiniDFSCluster cluster = null;
FileSystem fileSys = null;
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(NumDataNodes).EnableManagedDfsDirsRedundancy
(false).Build();
cluster.WaitActive();
fileSys = cluster.GetFileSystem();
FSNamesystem namesystem = cluster.GetNamesystem();
FSImage fsimage = namesystem.GetFSImage();
for (int i = 0; i < 20; i++)
{
fileSys.Mkdirs(new Path("/tmp/tmp" + i));
}
Storage.StorageDirectory sd = fsimage.GetStorage().DirIterator(NNStorage.NameNodeDirType
.Edits).Next();
cluster.Shutdown();
FilePath editFile = FSImageTestUtil.FindLatestEditsLog(sd).GetFile();
NUnit.Framework.Assert.IsTrue("Should exist: " + editFile, editFile.Exists());
// Corrupt the edits file.
long fileLen = editFile.Length();
RandomAccessFile rwf = new RandomAccessFile(editFile, "rw");
rwf.Seek(fileLen - 40);
for (int i_1 = 0; i_1 < 20; i_1++)
{
rwf.Write(FSEditLogOpCodes.OpDelete.GetOpCode());
}
rwf.Close();
StringBuilder bld = new StringBuilder();
bld.Append("^Error replaying edit log at offset \\d+. ");
bld.Append("Expected transaction ID was \\d+\n");
bld.Append("Recent opcode offsets: (\\d+\\s*){4}$");
try
{
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(NumDataNodes).EnableManagedDfsDirsRedundancy
(false).Format(false).Build();
NUnit.Framework.Assert.Fail("should not be able to start");
}
catch (IOException e)
{
NUnit.Framework.Assert.IsTrue("error message contains opcodes message", e.Message
.Matches(bld.ToString()));
}
}
public virtual void TestBackupNodeTailsEdits()
{
Configuration conf = new HdfsConfiguration();
HAUtil.SetAllowStandbyReads(conf, true);
MiniDFSCluster cluster = null;
FileSystem fileSys = null;
BackupNode backup = null;
try
{
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(0).Build();
fileSys = cluster.GetFileSystem();
backup = StartBackupNode(conf, HdfsServerConstants.StartupOption.Backup, 1);
BackupImage bnImage = (BackupImage)backup.GetFSImage();
TestBNInSync(cluster, backup, 1);
// Force a roll -- BN should roll with NN.
NameNode nn = cluster.GetNameNode();
NamenodeProtocols nnRpc = nn.GetRpcServer();
nnRpc.RollEditLog();
NUnit.Framework.Assert.AreEqual(bnImage.GetEditLog().GetCurSegmentTxId(), nn.GetFSImage
().GetEditLog().GetCurSegmentTxId());
// BN should stay in sync after roll
TestBNInSync(cluster, backup, 2);
long nnImageBefore = nn.GetFSImage().GetStorage().GetMostRecentCheckpointTxId();
// BN checkpoint
backup.DoCheckpoint();
// NN should have received a new image
long nnImageAfter = nn.GetFSImage().GetStorage().GetMostRecentCheckpointTxId();
NUnit.Framework.Assert.IsTrue("nn should have received new checkpoint. before: "
+ nnImageBefore + " after: " + nnImageAfter, nnImageAfter > nnImageBefore);
// BN should stay in sync after checkpoint
TestBNInSync(cluster, backup, 3);
// Stop BN
Storage.StorageDirectory sd = bnImage.GetStorage().GetStorageDir(0);
backup.Stop();
backup = null;
// When shutting down the BN, it shouldn't finalize logs that are
// still open on the NN
FileJournalManager.EditLogFile editsLog = FSImageTestUtil.FindLatestEditsLog(sd);
NUnit.Framework.Assert.AreEqual(editsLog.GetFirstTxId(), nn.GetFSImage().GetEditLog
().GetCurSegmentTxId());
NUnit.Framework.Assert.IsTrue("Should not have finalized " + editsLog, editsLog.IsInProgress
());
// do some edits
NUnit.Framework.Assert.IsTrue(fileSys.Mkdirs(new Path("/edit-while-bn-down")));
// start a new backup node
backup = StartBackupNode(conf, HdfsServerConstants.StartupOption.Backup, 1);
TestBNInSync(cluster, backup, 4);
NUnit.Framework.Assert.IsNotNull(backup.GetNamesystem().GetFileInfo("/edit-while-bn-down"
, false));
}
finally
{
Log.Info("Shutting down...");
if (backup != null)
{
backup.Stop();
}
if (fileSys != null)
{
fileSys.Close();
}
if (cluster != null)
{
cluster.Shutdown();
}
}
AssertStorageDirsMatch(cluster.GetNameNode(), backup);
}
/// <exception cref="System.IO.IOException"/>
internal static void TestNameNodeRecoveryImpl(TestNameNodeRecovery.Corruptor corruptor
, bool finalize)
{
string TestPath = "/test/path/dir";
string TestPath2 = "/second/dir";
bool needRecovery = corruptor.NeedRecovery(finalize);
// start a cluster
Configuration conf = new HdfsConfiguration();
SetupRecoveryTestConf(conf);
MiniDFSCluster cluster = null;
FileSystem fileSys = null;
Storage.StorageDirectory sd = null;
try
{
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(0).ManageNameDfsDirs(false
).Build();
cluster.WaitActive();
if (!finalize)
{
// Normally, the in-progress edit log would be finalized by
// FSEditLog#endCurrentLogSegment. For testing purposes, we
// disable that here.
FSEditLog spyLog = Org.Mockito.Mockito.Spy(cluster.GetNameNode().GetFSImage().GetEditLog
());
Org.Mockito.Mockito.DoNothing().When(spyLog).EndCurrentLogSegment(true);
DFSTestUtil.SetEditLogForTesting(cluster.GetNamesystem(), spyLog);
}
fileSys = cluster.GetFileSystem();
FSNamesystem namesystem = cluster.GetNamesystem();
FSImage fsimage = namesystem.GetFSImage();
fileSys.Mkdirs(new Path(TestPath));
fileSys.Mkdirs(new Path(TestPath2));
sd = fsimage.GetStorage().DirIterator(NNStorage.NameNodeDirType.Edits).Next();
}
finally
{
if (cluster != null)
{
cluster.Shutdown();
}
}
FilePath editFile = FSImageTestUtil.FindLatestEditsLog(sd).GetFile();
NUnit.Framework.Assert.IsTrue("Should exist: " + editFile, editFile.Exists());
// Corrupt the edit log
Log.Info("corrupting edit log file '" + editFile + "'");
corruptor.Corrupt(editFile);
// If needRecovery == true, make sure that we can't start the
// cluster normally before recovery
cluster = null;
try
{
Log.Debug("trying to start normally (this should fail)...");
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(0).EnableManagedDfsDirsRedundancy
(false).Format(false).Build();
cluster.WaitActive();
cluster.Shutdown();
if (needRecovery)
{
NUnit.Framework.Assert.Fail("expected the corrupted edit log to prevent normal startup"
);
}
}
catch (IOException e)
{
if (!needRecovery)
{
Log.Error("Got unexpected failure with " + corruptor.GetName() + corruptor, e);
NUnit.Framework.Assert.Fail("got unexpected exception " + e.Message);
}
}
finally
{
if (cluster != null)
{
cluster.Shutdown();
}
}
// Perform NameNode recovery.
// Even if there was nothing wrong previously (needRecovery == false),
// this should still work fine.
cluster = null;
try
{
Log.Debug("running recovery...");
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(0).EnableManagedDfsDirsRedundancy
(false).Format(false).StartupOption(recoverStartOpt).Build();
}
catch (IOException e)
{
NUnit.Framework.Assert.Fail("caught IOException while trying to recover. " + "message was "
+ e.Message + "\nstack trace\n" + StringUtils.StringifyException(e));
}
finally
{
if (cluster != null)
{
cluster.Shutdown();
}
}
// Make sure that we can start the cluster normally after recovery
cluster = null;
try
{
Log.Debug("starting cluster normally after recovery...");
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(0).EnableManagedDfsDirsRedundancy
(false).Format(false).Build();
Log.Debug("successfully recovered the " + corruptor.GetName() + " corrupted edit log"
);
cluster.WaitActive();
NUnit.Framework.Assert.IsTrue(cluster.GetFileSystem().Exists(new Path(TestPath)));
}
catch (IOException e)
{
NUnit.Framework.Assert.Fail("failed to recover. Error message: " + e.Message);
}
finally
{
if (cluster != null)
{
cluster.Shutdown();
}
}
}
public virtual void TestSaveNamespace()
{
DistributedFileSystem fs = null;
try
{
Configuration conf = new HdfsConfiguration();
conf.SetBoolean(DFSConfigKeys.DfsNamenodeDelegationTokenAlwaysUseKey, true);
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(numDatanodes).Build();
cluster.WaitActive();
fs = cluster.GetFileSystem();
FSNamesystem namesystem = cluster.GetNamesystem();
string renewer = UserGroupInformation.GetLoginUser().GetUserName();
Org.Apache.Hadoop.Security.Token.Token <DelegationTokenIdentifier> token1 = namesystem
.GetDelegationToken(new Text(renewer));
Org.Apache.Hadoop.Security.Token.Token <DelegationTokenIdentifier> token2 = namesystem
.GetDelegationToken(new Text(renewer));
// Saving image without safe mode should fail
DFSAdmin admin = new DFSAdmin(conf);
string[] args = new string[] { "-saveNamespace" };
// verify that the edits file is NOT empty
NameNode nn = cluster.GetNameNode();
foreach (Storage.StorageDirectory sd in nn.GetFSImage().GetStorage().DirIterable(
null))
{
FileJournalManager.EditLogFile log = FSImageTestUtil.FindLatestEditsLog(sd);
NUnit.Framework.Assert.IsTrue(log.IsInProgress());
log.ValidateLog();
long numTransactions = (log.GetLastTxId() - log.GetFirstTxId()) + 1;
NUnit.Framework.Assert.AreEqual("In-progress log " + log + " should have 5 transactions"
, 5, numTransactions);
}
// Saving image in safe mode should succeed
fs.SetSafeMode(HdfsConstants.SafeModeAction.SafemodeEnter);
try
{
admin.Run(args);
}
catch (Exception e)
{
throw new IOException(e.Message);
}
// verify that the edits file is empty except for the START txn
foreach (Storage.StorageDirectory sd_1 in nn.GetFSImage().GetStorage().DirIterable
(null))
{
FileJournalManager.EditLogFile log = FSImageTestUtil.FindLatestEditsLog(sd_1);
NUnit.Framework.Assert.IsTrue(log.IsInProgress());
log.ValidateLog();
long numTransactions = (log.GetLastTxId() - log.GetFirstTxId()) + 1;
NUnit.Framework.Assert.AreEqual("In-progress log " + log + " should only have START txn"
, 1, numTransactions);
}
// restart cluster
cluster.Shutdown();
cluster = null;
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(numDatanodes).Format(false
).Build();
cluster.WaitActive();
//Should be able to renew & cancel the delegation token after cluster restart
try
{
RenewToken(token1);
RenewToken(token2);
}
catch (IOException)
{
NUnit.Framework.Assert.Fail("Could not renew or cancel the token");
}
namesystem = cluster.GetNamesystem();
Org.Apache.Hadoop.Security.Token.Token <DelegationTokenIdentifier> token3 = namesystem
.GetDelegationToken(new Text(renewer));
Org.Apache.Hadoop.Security.Token.Token <DelegationTokenIdentifier> token4 = namesystem
.GetDelegationToken(new Text(renewer));
// restart cluster again
cluster.Shutdown();
cluster = null;
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(numDatanodes).Format(false
).Build();
cluster.WaitActive();
namesystem = cluster.GetNamesystem();
Org.Apache.Hadoop.Security.Token.Token <DelegationTokenIdentifier> token5 = namesystem
.GetDelegationToken(new Text(renewer));
try
{
RenewToken(token1);
RenewToken(token2);
RenewToken(token3);
RenewToken(token4);
RenewToken(token5);
}
catch (IOException)
{
NUnit.Framework.Assert.Fail("Could not renew or cancel the token");
}
// restart cluster again
cluster.Shutdown();
cluster = null;
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(numDatanodes).Format(false
).Build();
cluster.WaitActive();
namesystem = cluster.GetNamesystem();
try
{
RenewToken(token1);
CancelToken(token1);
RenewToken(token2);
CancelToken(token2);
RenewToken(token3);
CancelToken(token3);
RenewToken(token4);
CancelToken(token4);
RenewToken(token5);
CancelToken(token5);
}
catch (IOException)
{
NUnit.Framework.Assert.Fail("Could not renew or cancel the token");
}
}
finally
{
if (fs != null)
{
fs.Close();
}
if (cluster != null)
{
cluster.Shutdown();
}
}
}
