/// <summary>Scan through the cached block map.</summary>
/// <remarks>
/// Scan through the cached block map.
/// Any blocks which are under-replicated should be assigned new Datanodes.
/// Blocks that are over-replicated should be removed from Datanodes.
/// </remarks>
private void RescanCachedBlockMap()
{
for (IEnumerator <CachedBlock> cbIter = cachedBlocks.GetEnumerator(); cbIter.HasNext
();)
{
scannedBlocks++;
CachedBlock cblock = cbIter.Next();
IList <DatanodeDescriptor> pendingCached = cblock.GetDatanodes(DatanodeDescriptor.CachedBlocksList.Type
.PendingCached);
IList <DatanodeDescriptor> cached = cblock.GetDatanodes(DatanodeDescriptor.CachedBlocksList.Type
.Cached);
IList <DatanodeDescriptor> pendingUncached = cblock.GetDatanodes(DatanodeDescriptor.CachedBlocksList.Type
.PendingUncached);
// Remove nodes from PENDING_UNCACHED if they were actually uncached.
for (IEnumerator <DatanodeDescriptor> iter = pendingUncached.GetEnumerator(); iter
.HasNext();)
{
DatanodeDescriptor datanode = iter.Next();
if (!cblock.IsInList(datanode.GetCached()))
{
Log.Trace("Block {}: removing from PENDING_UNCACHED for node {} " + "because the DataNode uncached it."
, cblock.GetBlockId(), datanode.GetDatanodeUuid());
datanode.GetPendingUncached().Remove(cblock);
iter.Remove();
}
}
BlockInfoContiguous blockInfo = blockManager.GetStoredBlock(new Block(cblock.GetBlockId
()));
string reason = FindReasonForNotCaching(cblock, blockInfo);
int neededCached = 0;
if (reason != null)
{
Log.Trace("Block {}: can't cache block because it is {}", cblock.GetBlockId(), reason
);
}
else
{
neededCached = cblock.GetReplication();
}
int numCached = cached.Count;
if (numCached >= neededCached)
{
// If we have enough replicas, drop all pending cached.
for (IEnumerator <DatanodeDescriptor> iter_1 = pendingCached.GetEnumerator(); iter_1
.HasNext();)
{
DatanodeDescriptor datanode = iter_1.Next();
datanode.GetPendingCached().Remove(cblock);
iter_1.Remove();
Log.Trace("Block {}: removing from PENDING_CACHED for node {}" + "because we already have {} cached replicas and we only"
+ " need {}", cblock.GetBlockId(), datanode.GetDatanodeUuid(), numCached, neededCached
);
}
}
if (numCached < neededCached)
{
// If we don't have enough replicas, drop all pending uncached.
for (IEnumerator <DatanodeDescriptor> iter_1 = pendingUncached.GetEnumerator(); iter_1
.HasNext();)
{
DatanodeDescriptor datanode = iter_1.Next();
datanode.GetPendingUncached().Remove(cblock);
iter_1.Remove();
Log.Trace("Block {}: removing from PENDING_UNCACHED for node {} " + "because we only have {} cached replicas and we need "
+ "{}", cblock.GetBlockId(), datanode.GetDatanodeUuid(), numCached, neededCached
);
}
}
int neededUncached = numCached - (pendingUncached.Count + neededCached);
if (neededUncached > 0)
{
AddNewPendingUncached(neededUncached, cblock, cached, pendingUncached);
}
else
{
int additionalCachedNeeded = neededCached - (numCached + pendingCached.Count);
if (additionalCachedNeeded > 0)
{
AddNewPendingCached(additionalCachedNeeded, cblock, cached, pendingCached);
}
}
if ((neededCached == 0) && pendingUncached.IsEmpty() && pendingCached.IsEmpty())
{
// we have nothing more to do with this block.
Log.Trace("Block {}: removing from cachedBlocks, since neededCached " + "== 0, and pendingUncached and pendingCached are empty."
, cblock.GetBlockId());
cbIter.Remove();
}
}
}
/// <summary>Add new entries to the PendingCached list.</summary>
/// <param name="neededCached">The number of replicas that need to be cached.</param>
/// <param name="cachedBlock">The block which needs to be cached.</param>
/// <param name="cached">A list of DataNodes currently caching the block.</param>
/// <param name="pendingCached">
/// A list of DataNodes that will soon cache the
/// block.
/// </param>
private void AddNewPendingCached(int neededCached, CachedBlock cachedBlock, IList
<DatanodeDescriptor> cached, IList <DatanodeDescriptor> pendingCached)
{
// To figure out which replicas can be cached, we consult the
// blocksMap. We don't want to try to cache a corrupt replica, though.
BlockInfoContiguous blockInfo = blockManager.GetStoredBlock(new Block(cachedBlock
.GetBlockId()));
if (blockInfo == null)
{
Log.Debug("Block {}: can't add new cached replicas," + " because there is no record of this block "
+ "on the NameNode.", cachedBlock.GetBlockId());
return;
}
if (!blockInfo.IsComplete())
{
Log.Debug("Block {}: can't cache this block, because it is not yet" + " complete."
, cachedBlock.GetBlockId());
return;
}
// Filter the list of replicas to only the valid targets
IList <DatanodeDescriptor> possibilities = new List <DatanodeDescriptor>();
int numReplicas = blockInfo.GetCapacity();
ICollection <DatanodeDescriptor> corrupt = blockManager.GetCorruptReplicas(blockInfo
);
int outOfCapacity = 0;
for (int i = 0; i < numReplicas; i++)
{
DatanodeDescriptor datanode = blockInfo.GetDatanode(i);
if (datanode == null)
{
continue;
}
if (datanode.IsDecommissioned() || datanode.IsDecommissionInProgress())
{
continue;
}
if (corrupt != null && corrupt.Contains(datanode))
{
continue;
}
if (pendingCached.Contains(datanode) || cached.Contains(datanode))
{
continue;
}
long pendingBytes = 0;
// Subtract pending cached blocks from effective capacity
IEnumerator <CachedBlock> it = datanode.GetPendingCached().GetEnumerator();
while (it.HasNext())
{
CachedBlock cBlock = it.Next();
BlockInfoContiguous info = blockManager.GetStoredBlock(new Block(cBlock.GetBlockId
()));
if (info != null)
{
pendingBytes -= info.GetNumBytes();
}
}
it = datanode.GetPendingUncached().GetEnumerator();
// Add pending uncached blocks from effective capacity
while (it.HasNext())
{
CachedBlock cBlock = it.Next();
BlockInfoContiguous info = blockManager.GetStoredBlock(new Block(cBlock.GetBlockId
()));
if (info != null)
{
pendingBytes += info.GetNumBytes();
}
}
long pendingCapacity = pendingBytes + datanode.GetCacheRemaining();
if (pendingCapacity < blockInfo.GetNumBytes())
{
Log.Trace("Block {}: DataNode {} is not a valid possibility " + "because the block has size {}, but the DataNode only has {}"
+ "bytes of cache remaining ({} pending bytes, {} already cached.", blockInfo.GetBlockId
(), datanode.GetDatanodeUuid(), blockInfo.GetNumBytes(), pendingCapacity, pendingBytes
, datanode.GetCacheRemaining());
outOfCapacity++;
continue;
}
possibilities.AddItem(datanode);
}
IList <DatanodeDescriptor> chosen = ChooseDatanodesForCaching(possibilities, neededCached
, blockManager.GetDatanodeManager().GetStaleInterval());
foreach (DatanodeDescriptor datanode_1 in chosen)
{
Log.Trace("Block {}: added to PENDING_CACHED on DataNode {}", blockInfo.GetBlockId
(), datanode_1.GetDatanodeUuid());
pendingCached.AddItem(datanode_1);
bool added = datanode_1.GetPendingCached().AddItem(cachedBlock);
System.Diagnostics.Debug.Assert(added);
}
// We were unable to satisfy the requested replication factor
if (neededCached > chosen.Count)
{
Log.Debug("Block {}: we only have {} of {} cached replicas." + " {} DataNodes have insufficient cache capacity."
, blockInfo.GetBlockId(), (cachedBlock.GetReplication() - neededCached + chosen.
Count), cachedBlock.GetReplication(), outOfCapacity);
}
}
public virtual void TestNodeCount()
{
// start a mini dfs cluster of 2 nodes
Configuration conf = new HdfsConfiguration();
MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).NumDataNodes(ReplicationFactor
).Build();
try
{
FSNamesystem namesystem = cluster.GetNamesystem();
BlockManager bm = namesystem.GetBlockManager();
HeartbeatManager hm = bm.GetDatanodeManager().GetHeartbeatManager();
FileSystem fs = cluster.GetFileSystem();
// populate the cluster with a one block file
Path FilePath = new Path("/testfile");
DFSTestUtil.CreateFile(fs, FilePath, 1L, ReplicationFactor, 1L);
DFSTestUtil.WaitReplication(fs, FilePath, ReplicationFactor);
ExtendedBlock block = DFSTestUtil.GetFirstBlock(fs, FilePath);
// keep a copy of all datanode descriptor
DatanodeDescriptor[] datanodes = hm.GetDatanodes();
// start two new nodes
cluster.StartDataNodes(conf, 2, true, null, null);
cluster.WaitActive();
// bring down first datanode
DatanodeDescriptor datanode = datanodes[0];
MiniDFSCluster.DataNodeProperties dnprop = cluster.StopDataNode(datanode.GetXferAddr
());
// make sure that NN detects that the datanode is down
BlockManagerTestUtil.NoticeDeadDatanode(cluster.GetNameNode(), datanode.GetXferAddr
());
// the block will be replicated
DFSTestUtil.WaitReplication(fs, FilePath, ReplicationFactor);
// restart the first datanode
cluster.RestartDataNode(dnprop);
cluster.WaitActive();
// check if excessive replica is detected (transient)
InitializeTimeout(Timeout);
while (CountNodes(block.GetLocalBlock(), namesystem).ExcessReplicas() == 0)
{
CheckTimeout("excess replicas not detected");
}
// find out a non-excess node
DatanodeDescriptor nonExcessDN = null;
foreach (DatanodeStorageInfo storage in bm.blocksMap.GetStorages(block.GetLocalBlock
()))
{
DatanodeDescriptor dn = storage.GetDatanodeDescriptor();
ICollection <Block> blocks = bm.excessReplicateMap[dn.GetDatanodeUuid()];
if (blocks == null || !blocks.Contains(block.GetLocalBlock()))
{
nonExcessDN = dn;
break;
}
}
NUnit.Framework.Assert.IsTrue(nonExcessDN != null);
// bring down non excessive datanode
dnprop = cluster.StopDataNode(nonExcessDN.GetXferAddr());
// make sure that NN detects that the datanode is down
BlockManagerTestUtil.NoticeDeadDatanode(cluster.GetNameNode(), nonExcessDN.GetXferAddr
());
// The block should be replicated
InitializeTimeout(Timeout);
while (CountNodes(block.GetLocalBlock(), namesystem).LiveReplicas() != ReplicationFactor
)
{
CheckTimeout("live replica count not correct", 1000);
}
// restart the first datanode
cluster.RestartDataNode(dnprop);
cluster.WaitActive();
// check if excessive replica is detected (transient)
InitializeTimeout(Timeout);
while (CountNodes(block.GetLocalBlock(), namesystem).ExcessReplicas() != 2)
{
CheckTimeout("excess replica count not equal to 2");
}
}
finally
{
cluster.Shutdown();
}
}