public virtual void TestDeletionWithZeroSizeBlock2()
{
Path foo = new Path("/foo");
Path subDir = new Path(foo, "sub");
Path bar = new Path(subDir, "bar");
DFSTestUtil.CreateFile(hdfs, bar, Blocksize, Replication, 0L);
hdfs.Append(bar);
INodeFile barNode = fsdir.GetINode4Write(bar.ToString()).AsFile();
BlockInfoContiguous[] blks = barNode.GetBlocks();
NUnit.Framework.Assert.AreEqual(1, blks.Length);
ExtendedBlock previous = new ExtendedBlock(fsn.GetBlockPoolId(), blks[0]);
cluster.GetNameNodeRpc().AddBlock(bar.ToString(), hdfs.GetClient().GetClientName(
), previous, null, barNode.GetId(), null);
SnapshotTestHelper.CreateSnapshot(hdfs, foo, "s1");
barNode = fsdir.GetINode4Write(bar.ToString()).AsFile();
blks = barNode.GetBlocks();
NUnit.Framework.Assert.AreEqual(2, blks.Length);
NUnit.Framework.Assert.AreEqual(Blocksize, blks[0].GetNumBytes());
NUnit.Framework.Assert.AreEqual(0, blks[1].GetNumBytes());
hdfs.Delete(subDir, true);
Path sbar = SnapshotTestHelper.GetSnapshotPath(foo, "s1", "sub/bar");
barNode = fsdir.GetINode(sbar.ToString()).AsFile();
blks = barNode.GetBlocks();
NUnit.Framework.Assert.AreEqual(1, blks.Length);
NUnit.Framework.Assert.AreEqual(Blocksize, blks[0].GetNumBytes());
}
/// <exception cref="System.Exception"/>
internal static INodeFile AssertBlockCollection(string path, int numBlocks, FSDirectory
dir, BlockManager blkManager)
{
INodeFile file = INodeFile.ValueOf(dir.GetINode(path), path);
NUnit.Framework.Assert.AreEqual(numBlocks, file.GetBlocks().Length);
foreach (BlockInfoContiguous b in file.GetBlocks())
{
AssertBlockCollection(blkManager, file, b);
}
return(file);
}
/// <exception cref="System.Exception"/>
internal virtual void AssertAllNull(INodeFile inode, Path path, string[] snapshots
)
{
NUnit.Framework.Assert.IsNull(inode.GetBlocks());
AssertINodeNull(path.ToString());
AssertINodeNullInSnapshots(path, snapshots);
}
public virtual void TestDeletionOfLaterBlocksWithZeroSizeFirstBlock()
{
Path foo = new Path("/foo");
Path bar = new Path(foo, "bar");
byte[] testData = Sharpen.Runtime.GetBytesForString("foo bar baz");
// Create a zero-length file.
DFSTestUtil.CreateFile(hdfs, bar, 0, Replication, 0L);
NUnit.Framework.Assert.AreEqual(0, fsdir.GetINode4Write(bar.ToString()).AsFile().
GetBlocks().Length);
// Create a snapshot that includes that file.
SnapshotTestHelper.CreateSnapshot(hdfs, foo, "s0");
// Extend that file.
FSDataOutputStream @out = hdfs.Append(bar);
@out.Write(testData);
@out.Close();
INodeFile barNode = fsdir.GetINode4Write(bar.ToString()).AsFile();
BlockInfoContiguous[] blks = barNode.GetBlocks();
NUnit.Framework.Assert.AreEqual(1, blks.Length);
NUnit.Framework.Assert.AreEqual(testData.Length, blks[0].GetNumBytes());
// Delete the file.
hdfs.Delete(bar, true);
// Now make sure that the NN can still save an fsimage successfully.
cluster.GetNameNode().GetRpcServer().SetSafeMode(HdfsConstants.SafeModeAction.SafemodeEnter
, false);
cluster.GetNameNode().GetRpcServer().SaveNamespace();
}
public virtual void TestBlockReceived()
{
Configuration conf = new HdfsConfiguration();
conf.SetLong(DFSConfigKeys.DfsBlockSizeKey, 1024);
MiniDFSCluster cluster = null;
try
{
cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(DatanodeCount).Build();
cluster.WaitActive();
DistributedFileSystem hdfs = cluster.GetFileSystem();
FSNamesystem fsn = cluster.GetNamesystem();
BlockManager blkManager = fsn.GetBlockManager();
string file = "/tmp.txt";
Path filePath = new Path(file);
short replFactor = 1;
DFSTestUtil.CreateFile(hdfs, filePath, 1024L, replFactor, 0);
// temporarily stop the heartbeat
AList <DataNode> datanodes = cluster.GetDataNodes();
for (int i = 0; i < DatanodeCount; i++)
{
DataNodeTestUtils.SetHeartbeatsDisabledForTests(datanodes[i], true);
}
hdfs.SetReplication(filePath, (short)DatanodeCount);
BlockManagerTestUtil.ComputeAllPendingWork(blkManager);
NUnit.Framework.Assert.AreEqual(1, blkManager.pendingReplications.Size());
INodeFile fileNode = fsn.GetFSDirectory().GetINode4Write(file).AsFile();
Block[] blocks = fileNode.GetBlocks();
NUnit.Framework.Assert.AreEqual(DatanodeCount - 1, blkManager.pendingReplications
.GetNumReplicas(blocks[0]));
LocatedBlock locatedBlock = hdfs.GetClient().GetLocatedBlocks(file, 0).Get(0);
DatanodeInfo existingDn = (locatedBlock.GetLocations())[0];
int reportDnNum = 0;
string poolId = cluster.GetNamesystem().GetBlockPoolId();
// let two datanodes (other than the one that already has the data) to
// report to NN
for (int i_1 = 0; i_1 < DatanodeCount && reportDnNum < 2; i_1++)
{
if (!datanodes[i_1].GetDatanodeId().Equals(existingDn))
{
DatanodeRegistration dnR = datanodes[i_1].GetDNRegistrationForBP(poolId);
StorageReceivedDeletedBlocks[] report = new StorageReceivedDeletedBlocks[] { new
StorageReceivedDeletedBlocks("Fake-storage-ID-Ignored", new ReceivedDeletedBlockInfo
[] { new ReceivedDeletedBlockInfo(blocks[0], ReceivedDeletedBlockInfo.BlockStatus
.ReceivedBlock, string.Empty) }) };
cluster.GetNameNodeRpc().BlockReceivedAndDeleted(dnR, poolId, report);
reportDnNum++;
}
}
NUnit.Framework.Assert.AreEqual(DatanodeCount - 3, blkManager.pendingReplications
.GetNumReplicas(blocks[0]));
// let the same datanodes report again
for (int i_2 = 0; i_2 < DatanodeCount && reportDnNum < 2; i_2++)
{
if (!datanodes[i_2].GetDatanodeId().Equals(existingDn))
{
DatanodeRegistration dnR = datanodes[i_2].GetDNRegistrationForBP(poolId);
StorageReceivedDeletedBlocks[] report = new StorageReceivedDeletedBlocks[] { new
StorageReceivedDeletedBlocks("Fake-storage-ID-Ignored", new ReceivedDeletedBlockInfo
[] { new ReceivedDeletedBlockInfo(blocks[0], ReceivedDeletedBlockInfo.BlockStatus
.ReceivedBlock, string.Empty) }) };
cluster.GetNameNodeRpc().BlockReceivedAndDeleted(dnR, poolId, report);
reportDnNum++;
}
}
NUnit.Framework.Assert.AreEqual(DatanodeCount - 3, blkManager.pendingReplications
.GetNumReplicas(blocks[0]));
// re-enable heartbeat for the datanode that has data
for (int i_3 = 0; i_3 < DatanodeCount; i_3++)
{
DataNodeTestUtils.SetHeartbeatsDisabledForTests(datanodes[i_3], false);
DataNodeTestUtils.TriggerHeartbeat(datanodes[i_3]);
}
Sharpen.Thread.Sleep(5000);
NUnit.Framework.Assert.AreEqual(0, blkManager.pendingReplications.Size());
}
finally
{
if (cluster != null)
{
cluster.Shutdown();
}
}
}
/// <summary>Test deleting a file with snapshots.</summary>
/// <remarks>
/// Test deleting a file with snapshots. Need to check the blocksMap to make
/// sure the corresponding record is updated correctly.
/// </remarks>
/// <exception cref="System.Exception"/>
public virtual void TestDeletionWithSnapshots()
{
Path file0 = new Path(sub1, "file0");
Path file1 = new Path(sub1, "file1");
Path sub2 = new Path(sub1, "sub2");
Path file2 = new Path(sub2, "file2");
Path file3 = new Path(sub1, "file3");
Path file4 = new Path(sub1, "file4");
Path file5 = new Path(sub1, "file5");
// Create file under sub1
DFSTestUtil.CreateFile(hdfs, file0, 4 * Blocksize, Replication, seed);
DFSTestUtil.CreateFile(hdfs, file1, 2 * Blocksize, Replication, seed);
DFSTestUtil.CreateFile(hdfs, file2, 3 * Blocksize, Replication, seed);
{
// Normal deletion
INodeFile f2 = AssertBlockCollection(file2.ToString(), 3, fsdir, blockmanager);
BlockInfoContiguous[] blocks = f2.GetBlocks();
hdfs.Delete(sub2, true);
// The INode should have been removed from the blocksMap
foreach (BlockInfoContiguous b in blocks)
{
NUnit.Framework.Assert.IsNull(blockmanager.GetBlockCollection(b));
}
}
// Create snapshots for sub1
string[] snapshots = new string[] { "s0", "s1", "s2" };
DFSTestUtil.CreateFile(hdfs, file3, 5 * Blocksize, Replication, seed);
SnapshotTestHelper.CreateSnapshot(hdfs, sub1, snapshots[0]);
DFSTestUtil.CreateFile(hdfs, file4, 1 * Blocksize, Replication, seed);
SnapshotTestHelper.CreateSnapshot(hdfs, sub1, snapshots[1]);
DFSTestUtil.CreateFile(hdfs, file5, 7 * Blocksize, Replication, seed);
SnapshotTestHelper.CreateSnapshot(hdfs, sub1, snapshots[2]);
{
// set replication so that the inode should be replaced for snapshots
INodeFile f1 = AssertBlockCollection(file1.ToString(), 2, fsdir, blockmanager);
NUnit.Framework.Assert.AreSame(typeof(INodeFile), f1.GetType());
hdfs.SetReplication(file1, (short)2);
f1 = AssertBlockCollection(file1.ToString(), 2, fsdir, blockmanager);
NUnit.Framework.Assert.IsTrue(f1.IsWithSnapshot());
NUnit.Framework.Assert.IsFalse(f1.IsUnderConstruction());
}
// Check the block information for file0
INodeFile f0 = AssertBlockCollection(file0.ToString(), 4, fsdir, blockmanager);
BlockInfoContiguous[] blocks0 = f0.GetBlocks();
// Also check the block information for snapshot of file0
Path snapshotFile0 = SnapshotTestHelper.GetSnapshotPath(sub1, "s0", file0.GetName
());
AssertBlockCollection(snapshotFile0.ToString(), 4, fsdir, blockmanager);
// Delete file0
hdfs.Delete(file0, true);
// Make sure the blocks of file0 is still in blocksMap
foreach (BlockInfoContiguous b_1 in blocks0)
{
NUnit.Framework.Assert.IsNotNull(blockmanager.GetBlockCollection(b_1));
}
AssertBlockCollection(snapshotFile0.ToString(), 4, fsdir, blockmanager);
// Compare the INode in the blocksMap with INodes for snapshots
string s1f0 = SnapshotTestHelper.GetSnapshotPath(sub1, "s1", file0.GetName()).ToString
();
AssertBlockCollection(s1f0, 4, fsdir, blockmanager);
// Delete snapshot s1
hdfs.DeleteSnapshot(sub1, "s1");
// Make sure the first block of file0 is still in blocksMap
foreach (BlockInfoContiguous b_2 in blocks0)
{
NUnit.Framework.Assert.IsNotNull(blockmanager.GetBlockCollection(b_2));
}
AssertBlockCollection(snapshotFile0.ToString(), 4, fsdir, blockmanager);
try
{
INodeFile.ValueOf(fsdir.GetINode(s1f0), s1f0);
NUnit.Framework.Assert.Fail("Expect FileNotFoundException when identifying the INode in a deleted Snapshot"
);
}
catch (IOException e)
{
GenericTestUtils.AssertExceptionContains("File does not exist: " + s1f0, e);
}
}
/// <summary>Apply a CacheDirective to a file.</summary>
/// <param name="directive">The CacheDirective to apply.</param>
/// <param name="file">The file.</param>
private void RescanFile(CacheDirective directive, INodeFile file)
{
BlockInfoContiguous[] blockInfos = file.GetBlocks();
// Increment the "needed" statistics
directive.AddFilesNeeded(1);
// We don't cache UC blocks, don't add them to the total here
long neededTotal = file.ComputeFileSizeNotIncludingLastUcBlock() * directive.GetReplication
();
directive.AddBytesNeeded(neededTotal);
// The pool's bytesNeeded is incremented as we scan. If the demand
// thus far plus the demand of this file would exceed the pool's limit,
// do not cache this file.
CachePool pool = directive.GetPool();
if (pool.GetBytesNeeded() > pool.GetLimit())
{
Log.Debug("Directive {}: not scanning file {} because " + "bytesNeeded for pool {} is {}, but the pool's limit is {}"
, directive.GetId(), file.GetFullPathName(), pool.GetPoolName(), pool.GetBytesNeeded
(), pool.GetLimit());
return;
}
long cachedTotal = 0;
foreach (BlockInfoContiguous blockInfo in blockInfos)
{
if (!blockInfo.GetBlockUCState().Equals(HdfsServerConstants.BlockUCState.Complete
))
{
// We don't try to cache blocks that are under construction.
Log.Trace("Directive {}: can't cache block {} because it is in state " + "{}, not COMPLETE."
, directive.GetId(), blockInfo, blockInfo.GetBlockUCState());
continue;
}
Block block = new Block(blockInfo.GetBlockId());
CachedBlock ncblock = new CachedBlock(block.GetBlockId(), directive.GetReplication
(), mark);
CachedBlock ocblock = cachedBlocks.Get(ncblock);
if (ocblock == null)
{
cachedBlocks.Put(ncblock);
ocblock = ncblock;
}
else
{
// Update bytesUsed using the current replication levels.
// Assumptions: we assume that all the blocks are the same length
// on each datanode. We can assume this because we're only caching
// blocks in state COMPLETE.
// Note that if two directives are caching the same block(s), they will
// both get them added to their bytesCached.
IList <DatanodeDescriptor> cachedOn = ocblock.GetDatanodes(DatanodeDescriptor.CachedBlocksList.Type
.Cached);
long cachedByBlock = Math.Min(cachedOn.Count, directive.GetReplication()) * blockInfo
.GetNumBytes();
cachedTotal += cachedByBlock;
if ((mark != ocblock.GetMark()) || (ocblock.GetReplication() < directive.GetReplication
()))
{
//
// Overwrite the block's replication and mark in two cases:
//
// 1. If the mark on the CachedBlock is different from the mark for
// this scan, that means the block hasn't been updated during this
// scan, and we should overwrite whatever is there, since it is no
// longer valid.
//
// 2. If the replication in the CachedBlock is less than what the
// directive asks for, we want to increase the block's replication
// field to what the directive asks for.
//
ocblock.SetReplicationAndMark(directive.GetReplication(), mark);
}
}
Log.Trace("Directive {}: setting replication for block {} to {}", directive.GetId
(), blockInfo, ocblock.GetReplication());
}
// Increment the "cached" statistics
directive.AddBytesCached(cachedTotal);
if (cachedTotal == neededTotal)
{
directive.AddFilesCached(1);
}
Log.Debug("Directive {}: caching {}: {}/{} bytes", directive.GetId(), file.GetFullPathName
(), cachedTotal, neededTotal);
}
