public virtual void TestBlockRecoveryWithLessMetafile()
{
Configuration conf = new Configuration();
conf.Set(DFSConfigKeys.DfsBlockLocalPathAccessUserKey, UserGroupInformation.GetCurrentUser
().GetShortUserName());
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(1).Build();
Path file = new Path("/testRecoveryFile");
DistributedFileSystem dfs = cluster.GetFileSystem();
FSDataOutputStream @out = dfs.Create(file);
int count = 0;
while (count < 2 * 1024 * 1024)
{
@out.WriteBytes("Data");
count += 4;
}
@out.Hsync();
// abort the original stream
((DFSOutputStream)@out.GetWrappedStream()).Abort();
LocatedBlocks locations = cluster.GetNameNodeRpc().GetBlockLocations(file.ToString
(), 0, count);
ExtendedBlock block = locations.Get(0).GetBlock();
DataNode dn = cluster.GetDataNodes()[0];
BlockLocalPathInfo localPathInfo = dn.GetBlockLocalPathInfo(block, null);
FilePath metafile = new FilePath(localPathInfo.GetMetaPath());
NUnit.Framework.Assert.IsTrue(metafile.Exists());
// reduce the block meta file size
RandomAccessFile raf = new RandomAccessFile(metafile, "rw");
raf.SetLength(metafile.Length() - 20);
raf.Close();
// restart DN to make replica to RWR
MiniDFSCluster.DataNodeProperties dnProp = cluster.StopDataNode(0);
cluster.RestartDataNode(dnProp, true);
// try to recover the lease
DistributedFileSystem newdfs = (DistributedFileSystem)FileSystem.NewInstance(cluster
.GetConfiguration(0));
count = 0;
while (++count < 10 && !newdfs.RecoverLease(file))
{
Sharpen.Thread.Sleep(1000);
}
NUnit.Framework.Assert.IsTrue("File should be closed", newdfs.RecoverLease(file));
}
/// <summary>Test NN crash and client crash/stuck immediately after block allocation</summary>
/// <exception cref="System.Exception"/>
public virtual void TestOpenFileWhenNNAndClientCrashAfterAddBlock()
{
cluster.GetConfiguration(0).Set(DFSConfigKeys.DfsNamenodeSafemodeThresholdPctKey,
"1.0f");
string testData = "testData";
// to make sure we write the full block before creating dummy block at NN.
cluster.GetConfiguration(0).SetInt("io.bytes.per.checksum", testData.Length);
cluster.RestartNameNode(0);
try
{
cluster.WaitActive();
cluster.TransitionToActive(0);
cluster.TransitionToStandby(1);
DistributedFileSystem dfs = cluster.GetFileSystem(0);
string pathString = "/tmp1.txt";
Path filePath = new Path(pathString);
FSDataOutputStream create = dfs.Create(filePath, FsPermission.GetDefault(), true,
1024, (short)3, testData.Length, null);
create.Write(Sharpen.Runtime.GetBytesForString(testData));
create.Hflush();
long fileId = ((DFSOutputStream)create.GetWrappedStream()).GetFileId();
FileStatus fileStatus = dfs.GetFileStatus(filePath);
DFSClient client = DFSClientAdapter.GetClient(dfs);
// add one dummy block at NN, but not write to DataNode
ExtendedBlock previousBlock = DFSClientAdapter.GetPreviousBlock(client, fileId);
DFSClientAdapter.GetNamenode(client).AddBlock(pathString, client.GetClientName(),
new ExtendedBlock(previousBlock), new DatanodeInfo[0], DFSClientAdapter.GetFileId
((DFSOutputStream)create.GetWrappedStream()), null);
cluster.RestartNameNode(0, true);
cluster.RestartDataNode(0);
cluster.TransitionToActive(0);
// let the block reports be processed.
Sharpen.Thread.Sleep(2000);
FSDataInputStream @is = dfs.Open(filePath);
@is.Close();
dfs.RecoverLease(filePath);
// initiate recovery
NUnit.Framework.Assert.IsTrue("Recovery also should be success", dfs.RecoverLease
(filePath));
}
finally
{
cluster.Shutdown();
}
}
/// <exception cref="System.Exception"/>
private void RecoverLease(Path filepath, DistributedFileSystem dfs)
{
if (dfs == null)
{
dfs = (DistributedFileSystem)GetFSAsAnotherUser(conf);
}
while (!dfs.RecoverLease(filepath))
{
AppendTestUtil.Log.Info("sleep " + 5000 + "ms");
Sharpen.Thread.Sleep(5000);
}
}
/// <summary>Test append over storage quota does not mark file as UC or create lease</summary>
/// <exception cref="System.Exception"/>
public virtual void TestAppendOverStorageQuota()
{
Path dir = new Path("/TestAppendOverQuota");
Path file = new Path(dir, "file");
// create partial block file
dfs.Mkdirs(dir);
DFSTestUtil.CreateFile(dfs, file, Blocksize / 2, Replication, seed);
// lower quota to cause exception when appending to partial block
dfs.SetQuota(dir, long.MaxValue - 1, 1);
INodeDirectory dirNode = fsdir.GetINode4Write(dir.ToString()).AsDirectory();
long spaceUsed = dirNode.GetDirectoryWithQuotaFeature().GetSpaceConsumed().GetStorageSpace
();
try
{
DFSTestUtil.AppendFile(dfs, file, Blocksize);
NUnit.Framework.Assert.Fail("append didn't fail");
}
catch (DSQuotaExceededException)
{
}
// ignore
// check that the file exists, isn't UC, and has no dangling lease
INodeFile inode = fsdir.GetINode(file.ToString()).AsFile();
NUnit.Framework.Assert.IsNotNull(inode);
NUnit.Framework.Assert.IsFalse("should not be UC", inode.IsUnderConstruction());
NUnit.Framework.Assert.IsNull("should not have a lease", cluster.GetNamesystem().
GetLeaseManager().GetLeaseByPath(file.ToString()));
// make sure the quota usage is unchanged
long newSpaceUsed = dirNode.GetDirectoryWithQuotaFeature().GetSpaceConsumed().GetStorageSpace
();
NUnit.Framework.Assert.AreEqual(spaceUsed, newSpaceUsed);
// make sure edits aren't corrupted
dfs.RecoverLease(file);
cluster.RestartNameNodes();
}
public virtual void TestLeaseRecoveryAndAppend()
{
Configuration conf = new Configuration();
try
{
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(1).Build();
Path file = new Path("/testLeaseRecovery");
DistributedFileSystem dfs = cluster.GetFileSystem();
// create a file with 0 bytes
FSDataOutputStream @out = dfs.Create(file);
@out.Hflush();
@out.Hsync();
// abort the original stream
((DFSOutputStream)@out.GetWrappedStream()).Abort();
DistributedFileSystem newdfs = (DistributedFileSystem)FileSystem.NewInstance(cluster
.GetConfiguration(0));
// Append to a file , whose lease is held by another client should fail
try
{
newdfs.Append(file);
NUnit.Framework.Assert.Fail("Append to a file(lease is held by another client) should fail"
);
}
catch (RemoteException e)
{
NUnit.Framework.Assert.IsTrue(e.Message.Contains("file lease is currently owned")
);
}
// Lease recovery on first try should be successful
bool recoverLease = newdfs.RecoverLease(file);
NUnit.Framework.Assert.IsTrue(recoverLease);
FSDataOutputStream append = newdfs.Append(file);
append.Write(Sharpen.Runtime.GetBytesForString("test"));
append.Close();
}
finally
{
if (cluster != null)
{
cluster.Shutdown();
cluster = null;
}
}
}
/// <summary>
/// Test the scenario where the NN fails over after issuing a block
/// synchronization request, but before it is committed.
/// </summary>
/// <remarks>
/// Test the scenario where the NN fails over after issuing a block
/// synchronization request, but before it is committed. The
/// DN running the recovery should then fail to commit the synchronization
/// and a later retry will succeed.
/// </remarks>
/// <exception cref="System.Exception"/>
public virtual void TestFailoverRightBeforeCommitSynchronization()
{
Configuration conf = new Configuration();
// Disable permissions so that another user can recover the lease.
conf.SetBoolean(DFSConfigKeys.DfsPermissionsEnabledKey, false);
conf.SetInt(DFSConfigKeys.DfsBlockSizeKey, BlockSize);
FSDataOutputStream stm = null;
MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).NnTopology(MiniDFSNNTopology
.SimpleHATopology()).NumDataNodes(3).Build();
try
{
cluster.WaitActive();
cluster.TransitionToActive(0);
Sharpen.Thread.Sleep(500);
Log.Info("Starting with NN 0 active");
FileSystem fs = HATestUtil.ConfigureFailoverFs(cluster, conf);
stm = fs.Create(TestPath);
// write a half block
AppendTestUtil.Write(stm, 0, BlockSize / 2);
stm.Hflush();
// Look into the block manager on the active node for the block
// under construction.
NameNode nn0 = cluster.GetNameNode(0);
ExtendedBlock blk = DFSTestUtil.GetFirstBlock(fs, TestPath);
DatanodeDescriptor expectedPrimary = DFSTestUtil.GetExpectedPrimaryNode(nn0, blk);
Log.Info("Expecting block recovery to be triggered on DN " + expectedPrimary);
// Find the corresponding DN daemon, and spy on its connection to the
// active.
DataNode primaryDN = cluster.GetDataNode(expectedPrimary.GetIpcPort());
DatanodeProtocolClientSideTranslatorPB nnSpy = DataNodeTestUtils.SpyOnBposToNN(primaryDN
, nn0);
// Delay the commitBlockSynchronization call
GenericTestUtils.DelayAnswer delayer = new GenericTestUtils.DelayAnswer(Log);
Org.Mockito.Mockito.DoAnswer(delayer).When(nnSpy).CommitBlockSynchronization(Org.Mockito.Mockito
.Eq(blk), Org.Mockito.Mockito.AnyInt(), Org.Mockito.Mockito.AnyLong(), Org.Mockito.Mockito
.Eq(true), Org.Mockito.Mockito.Eq(false), (DatanodeID[])Org.Mockito.Mockito.AnyObject
(), (string[])Org.Mockito.Mockito.AnyObject());
// new genstamp
// new length
// close file
// delete block
// new targets
// new target storages
DistributedFileSystem fsOtherUser = CreateFsAsOtherUser(cluster, conf);
NUnit.Framework.Assert.IsFalse(fsOtherUser.RecoverLease(TestPath));
Log.Info("Waiting for commitBlockSynchronization call from primary");
delayer.WaitForCall();
Log.Info("Failing over to NN 1");
cluster.TransitionToStandby(0);
cluster.TransitionToActive(1);
// Let the commitBlockSynchronization call go through, and check that
// it failed with the correct exception.
delayer.Proceed();
delayer.WaitForResult();
Exception t = delayer.GetThrown();
if (t == null)
{
NUnit.Framework.Assert.Fail("commitBlockSynchronization call did not fail on standby"
);
}
GenericTestUtils.AssertExceptionContains("Operation category WRITE is not supported"
, t);
// Now, if we try again to recover the block, it should succeed on the new
// active.
LoopRecoverLease(fsOtherUser, TestPath);
AppendTestUtil.Check(fs, TestPath, BlockSize / 2);
}
finally
{
IOUtils.CloseStream(stm);
cluster.Shutdown();
}
}
/// <summary>Test race between delete operation and commitBlockSynchronization method.
/// </summary>
/// <remarks>
/// Test race between delete operation and commitBlockSynchronization method.
/// See HDFS-6825.
/// </remarks>
/// <param name="hasSnapshot"/>
/// <exception cref="System.Exception"/>
private void TestDeleteAndCommitBlockSynchronizationRace(bool hasSnapshot)
{
Log.Info("Start testing, hasSnapshot: " + hasSnapshot);
AList <AbstractMap.SimpleImmutableEntry <string, bool> > testList = new AList <AbstractMap.SimpleImmutableEntry
<string, bool> >();
testList.AddItem(new AbstractMap.SimpleImmutableEntry <string, bool>("/test-file",
false));
testList.AddItem(new AbstractMap.SimpleImmutableEntry <string, bool>("/test-file1"
, true));
testList.AddItem(new AbstractMap.SimpleImmutableEntry <string, bool>("/testdir/testdir1/test-file"
, false));
testList.AddItem(new AbstractMap.SimpleImmutableEntry <string, bool>("/testdir/testdir1/test-file1"
, true));
Path rootPath = new Path("/");
Configuration conf = new Configuration();
// Disable permissions so that another user can recover the lease.
conf.SetBoolean(DFSConfigKeys.DfsPermissionsEnabledKey, false);
conf.SetInt(DFSConfigKeys.DfsBlockSizeKey, BlockSize);
FSDataOutputStream stm = null;
IDictionary <DataNode, DatanodeProtocolClientSideTranslatorPB> dnMap = new Dictionary
<DataNode, DatanodeProtocolClientSideTranslatorPB>();
try
{
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(3).Build();
cluster.WaitActive();
DistributedFileSystem fs = cluster.GetFileSystem();
int stId = 0;
foreach (AbstractMap.SimpleImmutableEntry <string, bool> stest in testList)
{
string testPath = stest.Key;
bool mkSameDir = stest.Value;
Log.Info("test on " + testPath + " mkSameDir: " + mkSameDir + " snapshot: " + hasSnapshot
);
Path fPath = new Path(testPath);
//find grandest non-root parent
Path grandestNonRootParent = fPath;
while (!grandestNonRootParent.GetParent().Equals(rootPath))
{
grandestNonRootParent = grandestNonRootParent.GetParent();
}
stm = fs.Create(fPath);
Log.Info("test on " + testPath + " created " + fPath);
// write a half block
AppendTestUtil.Write(stm, 0, BlockSize / 2);
stm.Hflush();
if (hasSnapshot)
{
SnapshotTestHelper.CreateSnapshot(fs, rootPath, "st" + stId.ToString());
++stId;
}
// Look into the block manager on the active node for the block
// under construction.
NameNode nn = cluster.GetNameNode();
ExtendedBlock blk = DFSTestUtil.GetFirstBlock(fs, fPath);
DatanodeDescriptor expectedPrimary = DFSTestUtil.GetExpectedPrimaryNode(nn, blk);
Log.Info("Expecting block recovery to be triggered on DN " + expectedPrimary);
// Find the corresponding DN daemon, and spy on its connection to the
// active.
DataNode primaryDN = cluster.GetDataNode(expectedPrimary.GetIpcPort());
DatanodeProtocolClientSideTranslatorPB nnSpy = dnMap[primaryDN];
if (nnSpy == null)
{
nnSpy = DataNodeTestUtils.SpyOnBposToNN(primaryDN, nn);
dnMap[primaryDN] = nnSpy;
}
// Delay the commitBlockSynchronization call
GenericTestUtils.DelayAnswer delayer = new GenericTestUtils.DelayAnswer(Log);
Org.Mockito.Mockito.DoAnswer(delayer).When(nnSpy).CommitBlockSynchronization(Org.Mockito.Mockito
.Eq(blk), Org.Mockito.Mockito.AnyInt(), Org.Mockito.Mockito.AnyLong(), Org.Mockito.Mockito
.Eq(true), Org.Mockito.Mockito.Eq(false), (DatanodeID[])Org.Mockito.Mockito.AnyObject
(), (string[])Org.Mockito.Mockito.AnyObject());
// new genstamp
// new length
// close file
// delete block
// new targets
// new target storages
fs.RecoverLease(fPath);
Log.Info("Waiting for commitBlockSynchronization call from primary");
delayer.WaitForCall();
Log.Info("Deleting recursively " + grandestNonRootParent);
fs.Delete(grandestNonRootParent, true);
if (mkSameDir && !grandestNonRootParent.ToString().Equals(testPath))
{
Log.Info("Recreate dir " + grandestNonRootParent + " testpath: " + testPath);
fs.Mkdirs(grandestNonRootParent);
}
delayer.Proceed();
Log.Info("Now wait for result");
delayer.WaitForResult();
Exception t = delayer.GetThrown();
if (t != null)
{
Log.Info("Result exception (snapshot: " + hasSnapshot + "): " + t);
}
}
// end of loop each fPath
Log.Info("Now check we can restart");
cluster.RestartNameNodes();
Log.Info("Restart finished");
}
finally
{
if (stm != null)
{
IOUtils.CloseStream(stm);
}
if (cluster != null)
{
cluster.Shutdown();
}
}
}
/// <exception cref="System.IO.IOException"/>
internal override int Run(IList <string> args)
{
if (args.Count == 0)
{
System.Console.Out.WriteLine(this.usageText);
System.Console.Out.WriteLine(this.helpText + "\n");
return(1);
}
string pathStr = StringUtils.PopOptionWithArgument("-path", args);
string retriesStr = StringUtils.PopOptionWithArgument("-retries", args);
if (pathStr == null)
{
System.Console.Error.WriteLine("You must supply a -path argument to " + "recoverLease."
);
return(1);
}
int maxRetries = 1;
if (retriesStr != null)
{
try
{
maxRetries = System.Convert.ToInt32(retriesStr);
}
catch (FormatException e)
{
System.Console.Error.WriteLine("Failed to parse the argument to -retries: " + StringUtils
.StringifyException(e));
return(1);
}
}
FileSystem fs;
try
{
fs = FileSystem.NewInstance(new URI(pathStr), this._enclosing.GetConf(), null);
}
catch (URISyntaxException e)
{
System.Console.Error.WriteLine("URISyntaxException for " + pathStr + ":" + StringUtils
.StringifyException(e));
return(1);
}
catch (Exception e)
{
System.Console.Error.WriteLine("InterruptedException for " + pathStr + ":" + StringUtils
.StringifyException(e));
return(1);
}
DistributedFileSystem dfs = null;
try
{
dfs = (DistributedFileSystem)fs;
}
catch (InvalidCastException)
{
System.Console.Error.WriteLine("Invalid filesystem for path " + pathStr + ": " +
"needed scheme hdfs, but got: " + fs.GetScheme());
return(1);
}
for (int retry = 0; true;)
{
bool recovered = false;
IOException ioe = null;
try
{
recovered = dfs.RecoverLease(new Path(pathStr));
}
catch (IOException e)
{
ioe = e;
}
if (recovered)
{
System.Console.Out.WriteLine("recoverLease SUCCEEDED on " + pathStr);
return(0);
}
if (ioe != null)
{
System.Console.Error.WriteLine("recoverLease got exception: ");
Sharpen.Runtime.PrintStackTrace(ioe);
}
else
{
System.Console.Error.WriteLine("recoverLease returned false.");
}
retry++;
if (retry >= maxRetries)
{
break;
}
System.Console.Error.WriteLine("Retrying in " + DebugAdmin.RecoverLeaseCommand.TimeoutMs
+ " ms...");
Uninterruptibles.SleepUninterruptibly(DebugAdmin.RecoverLeaseCommand.TimeoutMs, TimeUnit
.Milliseconds);
System.Console.Error.WriteLine("Retry #" + retry);
}
System.Console.Error.WriteLine("Giving up on recoverLease for " + pathStr + " after "
+ maxRetries + (maxRetries == 1 ? " try." : " tries."));
return(1);
}