/// <summary>
/// This function identifies and returns the hosts that contribute
/// most for a given split.
/// </summary>
/// <remarks>
/// This function identifies and returns the hosts that contribute
/// most for a given split. For calculating the contribution, rack
/// locality is treated on par with host locality, so hosts from racks
/// that contribute the most are preferred over hosts on racks that
/// contribute less
/// </remarks>
/// <param name="blkLocations">The list of block locations</param>
/// <param name="offset"></param>
/// <param name="splitSize"></param>
/// <returns>
/// two arrays - one of hosts that contribute most to this split, and
/// one of hosts that contribute most to this split that have the data
/// cached on them
/// </returns>
/// <exception cref="System.IO.IOException"/>
private string[][] GetSplitHostsAndCachedHosts(BlockLocation[] blkLocations, long
offset, long splitSize, NetworkTopology clusterMap)
{
int startIndex = GetBlockIndex(blkLocations, offset);
long bytesInThisBlock = blkLocations[startIndex].GetOffset() + blkLocations[startIndex
].GetLength() - offset;
//If this is the only block, just return
if (bytesInThisBlock >= splitSize)
{
return(new string[][] { blkLocations[startIndex].GetHosts(), blkLocations[startIndex
].GetCachedHosts() });
}
long bytesInFirstBlock = bytesInThisBlock;
int index = startIndex + 1;
splitSize -= bytesInThisBlock;
while (splitSize > 0)
{
bytesInThisBlock = Math.Min(splitSize, blkLocations[index++].GetLength());
splitSize -= bytesInThisBlock;
}
long bytesInLastBlock = bytesInThisBlock;
int endIndex = index - 1;
IDictionary <Node, FileInputFormat.NodeInfo> hostsMap = new IdentityHashMap <Node,
FileInputFormat.NodeInfo>();
IDictionary <Node, FileInputFormat.NodeInfo> racksMap = new IdentityHashMap <Node,
FileInputFormat.NodeInfo>();
string[] allTopos = new string[0];
// Build the hierarchy and aggregate the contribution of
// bytes at each level. See TestGetSplitHosts.java
for (index = startIndex; index <= endIndex; index++)
{
// Establish the bytes in this block
if (index == startIndex)
{
bytesInThisBlock = bytesInFirstBlock;
}
else
{
if (index == endIndex)
{
bytesInThisBlock = bytesInLastBlock;
}
else
{
bytesInThisBlock = blkLocations[index].GetLength();
}
}
allTopos = blkLocations[index].GetTopologyPaths();
// If no topology information is available, just
// prefix a fakeRack
if (allTopos.Length == 0)
{
allTopos = FakeRacks(blkLocations, index);
}
// NOTE: This code currently works only for one level of
// hierarchy (rack/host). However, it is relatively easy
// to extend this to support aggregation at different
// levels
foreach (string topo in allTopos)
{
Node node;
Node parentNode;
FileInputFormat.NodeInfo nodeInfo;
FileInputFormat.NodeInfo parentNodeInfo;
node = clusterMap.GetNode(topo);
if (node == null)
{
node = new NodeBase(topo);
clusterMap.Add(node);
}
nodeInfo = hostsMap[node];
if (nodeInfo == null)
{
nodeInfo = new FileInputFormat.NodeInfo(node);
hostsMap[node] = nodeInfo;
parentNode = node.GetParent();
parentNodeInfo = racksMap[parentNode];
if (parentNodeInfo == null)
{
parentNodeInfo = new FileInputFormat.NodeInfo(parentNode);
racksMap[parentNode] = parentNodeInfo;
}
parentNodeInfo.AddLeaf(nodeInfo);
}
else
{
nodeInfo = hostsMap[node];
parentNode = node.GetParent();
parentNodeInfo = racksMap[parentNode];
}
nodeInfo.AddValue(index, bytesInThisBlock);
parentNodeInfo.AddValue(index, bytesInThisBlock);
}
}
// for all topos
// for all indices
// We don't yet support cached hosts when bytesInThisBlock > splitSize
return(new string[][] { IdentifyHosts(allTopos.Length, racksMap), new string[0] });
}