//
// Main Dryad LINQ execution stuff
//
public int ExecLinqToDryad(string[] args)
{
//
// must be at least two arguments (program name and query XML file name)
//
if (args.Length < 2)
{
DryadLogger.LogCritical(0, null, "Must provide at least query XML file name.");
return(-1);
}
//
// break if --break is included in arguments (and eliminate it, as it is not known downstream)
//
if (ConsumeSingleArgument("--break", ref args))
{
DebugHelper.WaitForDebugger();
}
//
// parse the XML input, producing a DryadLINQ Query
//
Query query = new Query();
QueryPlanParser parser = new QueryPlanParser();
if (!parser.ParseQueryXml(args[1], query))
{
DryadLogger.LogCritical(0, null, "Invalid query plan");
return(-1);
}
//
// build internal app arguments
//
List <string> internalArgs = new List <string>();
//
// add the XmlExecHost args to the internal app arguments
//
foreach (string xmlExecHostArg in query.xmlExecHostArgs)
{
if (xmlExecHostArg == "--break")
{
DebugHelper.WaitForDebugger();
}
else
{
internalArgs.Add(xmlExecHostArg);
}
}
//
// combine internal arguments with any additional arguments received on the command line
// don't include argv[0] and argv[1] (program name and query XML file name)
//
int internalArgc = (int)internalArgs.Count;
int externalArgc = args.Length - 2; // don't include argv[0] and argv[1]
int combinedArgc = internalArgc + externalArgc;
string[] combinedArgv = new string[combinedArgc];
string msg = "";
// internal arguments first
for (int i = 0; i < internalArgc; i++)
{
combinedArgv[i] = internalArgs[i];
msg += String.Format("{0} ", combinedArgv[i]);
}
// then external arguments
for (int i = 0; i < externalArgc; i++)
{
combinedArgv[i + internalArgc] = args[i + 2]; // don't include argv[0] and argv[1]
msg += String.Format("{0} ", combinedArgv[i + internalArgc]);
}
DryadLogger.LogInformation(null, "Arguments: {0}", msg);
string jobClass = "DryadLINQ";
string dryadBinDir = Environment.GetEnvironmentVariable("DRYAD_HOME");
if (String.IsNullOrEmpty(dryadBinDir))
{
throw new ApplicationException("DryadLINQ requires the DRYAD_HOME environment variable to be set to the Dryad binary folder.");
}
string exeName = Path.Combine(dryadBinDir, "VertexHost.exe");
// create app and run it
//
DrGraphParameters p = DrDefaultParameters.Make(exeName, jobClass, query.enableSpeculativeDuplication);
DrArtemisLegacyReporter reporter = new DrArtemisLegacyReporter();
p.m_defaultProcessTemplate.GetListenerList().Add(reporter);
p.m_defaultVertexTemplate.GetListenerList().Add(reporter);
p.m_topologyReporter = reporter;
p.m_intermediateCompressionMode = query.intermediateDataCompression;
DrGraphExecutor graphExecutor = new DrGraphExecutor();
DrGraph graph = graphExecutor.Initialize(p);
if (graph == null)
{
DryadLogger.LogCritical(0, null, "Failed to initialize Graph Executor");
return(-1);
}
DryadLINQApp app = new DryadLINQApp(graph);
// Initialize with arguments
app.SetXmlFileName(args[1]);
if (!app.ParseCommandLineFlags(combinedArgv))
{
DryadLogger.LogCritical(0, null, "Bad command-line options");
return(-1);
}
// Build graph from query plan
GraphBuilder builder = new GraphBuilder();
builder.BuildGraphFromQuery(app, query);
// Run the app
DryadLogger.LogInformation(null, "Running the app");
graphExecutor.Run();
DrError exitStatus = graphExecutor.Join();
DryadLogger.LogInformation(null, "Finished running the app");
if (exitStatus == null || exitStatus.m_code == 0)
{
FinalizeExecution(query, graph);
DryadLogger.LogInformation(null, "Application completed successfully.");
return(0);
}
else
{
DryadLogger.LogCritical(exitStatus.m_code, null, "Application failed with error code 0x{0:X8}.\n", exitStatus.m_code);
return(exitStatus.m_code);
}
}
//
// Main Dryad LINQ execution stuff
//
public int ExecLinqToDryad(string[] args)
{
//
// must be at least two arguments (program name and query XML file name)
//
if (args.Length < 2)
{
DryadLogger.LogCritical(0, null, "Must provide at least query XML file name.");
return -1;
}
//
// break if --break is included in arguments (and eliminate it, as it is not known downstream)
//
if (ConsumeSingleArgument("--break", ref args))
{
DebugHelper.WaitForDebugger();
}
//
// parse the XML input, producing a DryadLINQ Query
//
Query query = new Query();
QueryPlanParser parser = new QueryPlanParser();
if (!parser.ParseQueryXml(args[1], query))
{
DryadLogger.LogCritical(0, null, "Invalid query plan");
return -1;
}
//
// build internal app arguments
//
List<string> internalArgs = new List<string>();
//
// add the XmlExecHost args to the internal app arguments
//
foreach (string xmlExecHostArg in query.xmlExecHostArgs)
{
if (xmlExecHostArg == "--break")
{
DebugHelper.WaitForDebugger();
}
else
{
internalArgs.Add(xmlExecHostArg);
}
}
//
// combine internal arguments with any additional arguments received on the command line
// don't include argv[0] and argv[1] (program name and query XML file name)
//
int internalArgc = (int)internalArgs.Count;
int externalArgc = args.Length - 2; // don't include argv[0] and argv[1]
int combinedArgc = internalArgc + externalArgc;
string[] combinedArgv = new string[combinedArgc];
string msg = "";
// internal arguments first
for (int i=0; i<internalArgc; i++)
{
combinedArgv[i] = internalArgs[i];
msg += String.Format("{0} ", combinedArgv[i]);
}
// then external arguments
for (int i = 0; i<externalArgc; i++)
{
combinedArgv[i+internalArgc] = args[i+2]; // don't include argv[0] and argv[1]
msg += String.Format("{0} ", combinedArgv[i+internalArgc]);
}
DryadLogger.LogInformation(null, "Arguments: {0}", msg);
string jobClass = "DryadLINQ";
string dryadBinDir = Environment.GetEnvironmentVariable("DRYAD_HOME");
if (String.IsNullOrEmpty(dryadBinDir))
{
throw new ApplicationException("DryadLINQ requires the DRYAD_HOME environment variable to be set to the Dryad binary folder.");
}
string exeName = Path.Combine(dryadBinDir, "VertexHost.exe");
// create app and run it
//
DrGraphParameters p = DrDefaultParameters.Make(exeName, jobClass, query.enableSpeculativeDuplication);
DrArtemisLegacyReporter reporter = new DrArtemisLegacyReporter();
p.m_defaultProcessTemplate.GetListenerList().Add(reporter);
p.m_defaultVertexTemplate.GetListenerList().Add(reporter);
p.m_topologyReporter = reporter;
p.m_intermediateCompressionMode = query.intermediateDataCompression;
DrGraphExecutor graphExecutor = new DrGraphExecutor();
DrGraph graph = graphExecutor.Initialize(p);
if (graph == null)
{
DryadLogger.LogCritical(0, null, "Failed to initialize Graph Executor");
return -1;
}
DryadLINQApp app = new DryadLINQApp(graph);
// Initialize with arguments
app.SetXmlFileName(args[1]);
if (!app.ParseCommandLineFlags(combinedArgv))
{
DryadLogger.LogCritical(0, null, "Bad command-line options");
return -1;
}
// Build graph from query plan
GraphBuilder builder = new GraphBuilder();
builder.BuildGraphFromQuery(app, query);
// Run the app
DryadLogger.LogInformation(null, "Running the app");
graphExecutor.Run();
DrError exitStatus = graphExecutor.Join();
DryadLogger.LogInformation(null, "Finished running the app");
if (exitStatus == null || exitStatus.m_code == 0)
{
FinalizeExecution(query, graph);
DryadLogger.LogInformation(null, "Application completed successfully.");
return 0;
}
else
{
DryadLogger.LogCritical(exitStatus.m_code, null, "Application failed with error code 0x{0:X8}.\n", exitStatus.m_code);
return exitStatus.m_code;
}
}
private DrInputStreamManager CreateInputNode(DryadLINQApp app, VertexInfo info, string inputName)
{
DrInputStreamManager s;
int err = 0;
DryadLogger.LogMethodEntry(inputName);
if (info.ioType == VertexInfo.IOType.PARTITIONEDFILE)
{
DrPartitionInputStream input = new DrPartitionInputStream();
err = input.Open(app.GetUniverse(), info.sources[0]);
if (!SUCCEEDED(err))
{
string msg = String.Format("Could not read DSC input file {0}", info.sources[0]);
throw new LinqToDryadException(msg, err);
}
DrManagerBase inputStage = new DrManagerBase(app.GetGraph(), inputName);
DrInputStreamManager inputManager = new DrInputStreamManager(input, inputStage);
s = inputManager;
}
//else if ( info.ioType == VertexInfo.IOType.STREAM )
//{
// DrDscInputStream input = new DrDscInputStream();
// DryadLogger.LogInformation("Create input node", "Opening DSC input fileset");
// err = input.Open(app.GetUniverse(), info.sources[0]);
// if (!SUCCEEDED(err))
// {
// string msg = String.Format("Could not read DSC input fileset {0}", info.sources[0]);
// throw new LinqToDryadException(msg, err);
// }
// DryadLogger.LogInformation("Create input node", "Opened DSC input fileset");
// DrManagerBase inputStage = new DrManagerBase(app.GetGraph(), inputName);
// DrInputStreamManager inputManager = new DrInputStreamManager(input, inputStage);
// s = inputManager;
//}
else if (info.ioType == VertexInfo.IOType.HDFS_STREAM)
{
DrHdfsInputStream input = new DrHdfsInputStream();
DryadLogger.LogInformation("Create input node", "Opening HDFS input fileset");
err = input.Open(app.GetUniverse(), info.sources[0]);
if (!SUCCEEDED(err))
{
string msg = String.Format("Could not read HDFS input fileset {0}", info.sources[0]);
throw new LinqToDryadException(msg, err);
}
DryadLogger.LogInformation("Create input node", "Opened HDFS input fileset");
DrManagerBase inputStage = new DrManagerBase(app.GetGraph(), inputName);
DrInputStreamManager inputManager = new DrInputStreamManager(input, inputStage);
s = inputManager;
}
else
{
string msg = String.Format("Unknown input type {0}", info.ioType);
throw new LinqToDryadException(msg);
}
DryadLogger.LogMethodExit();
return(s);
}
public void BuildGraphFromQuery(DryadLINQApp app, Query query)
{
// set configurable properties
int highThreshold = app.GetMaxAggregateInputs();
int lowThreshold = 16;
UInt64 highDataThreshold = (UInt64)app.GetAggregateThreshold();
UInt64 lowDataThreshold = (3 * highDataThreshold) / 4;
UInt64 maxSingleDataThreshold = highDataThreshold / 2;
int aggFilterThreshold = app.GetMaxAggregateFilterInputs();
// use a graph stage map to store the vertices as they are created, grouped by stage.
Dictionary <int, GraphStageInfo> graphStageMap = new Dictionary <int, GraphStageInfo>();
DryadLogger.LogInformation("Build Graph From Query", "Building graph");
//
// Create a set of vertices for each vertex (stage) in the query plan
//
DryadLogger.LogInformation("Build Graph From Query", "Adding vertices");
foreach (KeyValuePair <int, Vertex> kvp in query.queryPlan)
{
Vertex v = kvp.Value;
GraphStageInfo value = null;
if (!graphStageMap.TryGetValue(v.uniqueId, out value))
{
DryadLogger.LogInformation("Build Graph From Query", "Adding vertices for stage {0}", v.name);
CreateVertexSet(v, app, query, graphStageMap);
}
}
//
// Add dynamic stage managers
//
DryadLogger.LogInformation("Build Graph From Query", "Adding stage managers");
foreach (KeyValuePair <int, GraphStageInfo> kvp in graphStageMap)
{
Vertex v = kvp.Value.vertex;
//
//There are no dynamic managers
//
if (v.dynamicManager == null)
{
continue;
}
DrStageManager newManager = kvp.Value.stageManager; // newManager
DrGraphParameters parameters = app.GetGraph().GetParameters();
string stdVertexName = "MW";
string cpyVertexName = "CP";
if (v.type != Vertex.Type.INPUTTABLE && v.type != Vertex.Type.CONCAT)
{
if (v.dynamicManager.type == DynamicManager.Type.SPLITTER)
{
if (v.info.predecessors.Length == 1)
{
DrPipelineSplitManager splitter = new DrPipelineSplitManager();
newManager.AddDynamicConnectionManager(graphStageMap[v.info.predecessors[0].uniqueId].stageManager, splitter);
}
else
{
DrSemiPipelineSplitManager splitter = new DrSemiPipelineSplitManager();
newManager.AddDynamicConnectionManager(graphStageMap[v.info.predecessors[0].uniqueId].stageManager, splitter);
}
}
else if (v.dynamicManager.type == DynamicManager.Type.PARTIALAGGR)
{
DrDynamicAggregateManager dynamicMerge = new DrDynamicAggregateManager();
dynamicMerge.SetGroupingSettings(0, 0);
dynamicMerge.SetMachineGroupingSettings(2, aggFilterThreshold);
dynamicMerge.SetDataGroupingSettings(lowDataThreshold, highDataThreshold, maxSingleDataThreshold);
dynamicMerge.SetSplitAfterGrouping(true);
foreach (Predecessor p in v.info.predecessors)
{
newManager.AddDynamicConnectionManager(graphStageMap[p.uniqueId].stageManager, dynamicMerge);
}
}
else if (v.dynamicManager.type == DynamicManager.Type.FULLAGGR ||
v.dynamicManager.type == DynamicManager.Type.HASHDISTRIBUTOR)
{
int idx = 0;
int sz = v.dynamicManager.assemblyNames == null ? 0 : v.dynamicManager.assemblyNames.Length;
DrDynamicAggregateManager dynamicMerge = new DrDynamicAggregateManager();
if (v.dynamicManager.type == DynamicManager.Type.FULLAGGR || sz > 1)
{
dynamicMerge = new DrDynamicAggregateManager();
string name = v.dynamicManager.methodNames[idx];
DrManagerBase newStage = new DrManagerBase(app.GetGraph(), name);
DrActiveVertex mergeVertex = new DrActiveVertex(newStage, parameters.m_defaultProcessTemplate, parameters.m_defaultVertexTemplate);
mergeVertex.AddArgument(stdVertexName);
mergeVertex.AddArgument(v.dynamicManager.assemblyNames[idx]);
mergeVertex.AddArgument(v.dynamicManager.classNames[idx]);
mergeVertex.AddArgument(v.dynamicManager.methodNames[idx]);
idx++;
dynamicMerge.SetInternalVertex(mergeVertex);
dynamicMerge.SetGroupingSettings(0, 0);
dynamicMerge.SetPodGroupingSettings(lowThreshold, highThreshold);
dynamicMerge.SetDataGroupingSettings(lowDataThreshold,
highDataThreshold,
maxSingleDataThreshold);
dynamicMerge.SetMaxAggregationLevel(v.dynamicManager.aggregationLevels);
}
if (v.dynamicManager.type == DynamicManager.Type.FULLAGGR)
{
newManager.AddDynamicConnectionManager(graphStageMap[v.info.predecessors[0].uniqueId].stageManager, dynamicMerge);
}
else
{
string name = v.dynamicManager.methodNames[idx];
DrManagerBase newStage = new DrManagerBase(app.GetGraph(), name);
DrActiveVertex distributeVertex = new DrActiveVertex(newStage, parameters.m_defaultProcessTemplate, parameters.m_defaultVertexTemplate);
distributeVertex.AddArgument(stdVertexName);
distributeVertex.AddArgument(v.dynamicManager.assemblyNames[idx]);
distributeVertex.AddArgument(v.dynamicManager.classNames[idx]);
distributeVertex.AddArgument(v.dynamicManager.methodNames[idx]);
idx++;
DrDynamicDistributionManager dynamicHashDistribute =
new DrDynamicDistributionManager(distributeVertex, dynamicMerge);
dynamicHashDistribute.SetDataPerVertex(highDataThreshold * 2); // 2GB
newManager.AddDynamicConnectionManager(graphStageMap[v.info.predecessors[0].uniqueId].stageManager, dynamicHashDistribute);
}
}
else if (v.dynamicManager.type == DynamicManager.Type.RANGEDISTRIBUTOR)
{
DrStageManager splitManager = graphStageMap[v.dynamicManager.splitVertexId].stageManager;
DrDynamicRangeDistributionManager drdm = new DrDynamicRangeDistributionManager(splitManager, v.dynamicManager.sampleRate);
drdm.SetDataPerVertex(highDataThreshold * 2); // 2GB
newManager.AddDynamicConnectionManager(graphStageMap[v.info.predecessors[0].uniqueId].stageManager, drdm);
}
else if (v.dynamicManager.type == DynamicManager.Type.BROADCAST)
{
// the copy vertex
int bcastNumber = 0;
string nameString = String.Format("CP__{0}", bcastNumber++);
DrManagerBase newStage = new DrManagerBase(app.GetGraph(), nameString);
DrActiveVertex copyVertex =
new DrActiveVertex(newStage,
parameters.m_defaultProcessTemplate,
parameters.m_defaultVertexTemplate);
copyVertex.AddArgument(cpyVertexName);
DrDynamicBroadcastManager bcast = new DrDynamicBroadcastManager(copyVertex);
newManager.AddDynamicConnectionManager(graphStageMap[v.info.predecessors[0].uniqueId].stageManager, bcast);
}
else if (v.dynamicManager.type != DynamicManager.Type.NONE)
{
DryadLogger.LogWarning("Build Graph From Query", "Dynamic manager type {0} not supported yet", v.dynamicManager.type);
}
}
}
//
// Add all the edges
//
DryadLogger.LogInformation("Build Graph From Query", "Adding edges");
foreach (KeyValuePair <int, GraphStageInfo> kvp in graphStageMap)
{
AddEdges(kvp.Value, graphStageMap);
}
//
// Register the actual created vertices with the graph
//
MaterializeToManagers(graphStageMap);
}
private void CreateVertexSet(Vertex v, DryadLINQApp app, Query query, Dictionary <int, GraphStageInfo> graphStageMap)
{
SortedDictionary <int, Vertex> queryPlan = query.queryPlan;
DrVertexSet nodes = null;
DrStageManager newManager = null;
app.GetGraph().GetParameters();
DrGraphParameters parameters = app.GetGraph().GetParameters();
string stdVertexName = "MW";
if (v.type == Vertex.Type.INPUTTABLE)
{
DrInputStreamManager input = CreateInputNode(app, v.info, v.name);
newManager = input.GetStageManager();
nodes = CreateVertexSet(app.GetGraph(), input);
}
else if (v.type == Vertex.Type.OUTPUTTABLE)
{
DrOutputStreamManager output = CreateOutputNode(app, v.info, v.name);
newManager = output.GetStageManager();
nodes = CreateVertexSet(app.GetGraph(), output, v.partitions);
}
else if (v.type == Vertex.Type.CONCAT)
{
newManager = new DrManagerBase(app.GetGraph(), v.name);
// the set of nodes in a concat is just the set of nodes in the predecessor stages concatenated
nodes = new DrVertexSet();
foreach (Predecessor p in v.info.predecessors)
{
GraphStageInfo value = null;
if (graphStageMap.TryGetValue(p.uniqueId, out value))
{
nodes.InsertRange(nodes.Count, value.members);
}
else
{
throw new LinqToDryadException(String.Format("Concat: Failed to find predecessor {0} in graph stage map", p.uniqueId));
}
}
}
else
{
newManager = new DrManagerBase(app.GetGraph(), v.name);
DrVertex vertex;
if (v.type == Vertex.Type.TEE)
{
DrTeeVertex teeVertex = new DrTeeVertex(newManager);
vertex = teeVertex;
}
else
{
DrActiveVertex activeVertex = new DrActiveVertex(newManager, parameters.m_defaultProcessTemplate, parameters.m_defaultVertexTemplate);
activeVertex.AddArgument(stdVertexName);
activeVertex.AddArgument(v.info.assemblyName);
activeVertex.AddArgument(v.info.className);
activeVertex.AddArgument(v.info.methodName);
vertex = activeVertex;
}
nodes = CreateVertexSet(app.GetGraph(), vertex, v.partitions);
if (v.machines != null && v.machines.Length != 0 && v.type != Vertex.Type.TEE)
{
for (int i = 0; i < v.partitions; i++)
{
DrResource r = app.GetUniverse().LookUpResource(v.machines[i]);
if (r != null)
{
DrVertex baseVertex = nodes[i];
DrActiveVertex activeVertex = baseVertex as DrActiveVertex;
activeVertex.GetAffinity().SetHardConstraint(true);
activeVertex.GetAffinity().AddLocality(r);
}
}
}
FileStream mapfile = app.GetIdentityMapFile();
if (mapfile != null)
{
for (int i = 0; i < v.partitions; i++)
{
DrVertex jmv = nodes[i];
string message = String.Format("Mapping {0}[{1}] to {2}\n", v.uniqueId, i, jmv.GetId());
System.Text.ASCIIEncoding encoding = new System.Text.ASCIIEncoding();
mapfile.Seek(0, SeekOrigin.End);
mapfile.Write(encoding.GetBytes(message), 0, message.Length);
}
}
}
graphStageMap.Add(v.uniqueId, new GraphStageInfo(nodes, v, newManager));
}
private DrOutputStreamManager CreateOutputNode(DryadLINQApp app, VertexInfo info, string outputName)
{
DrOutputStreamManager s;
if (info.ioType == VertexInfo.IOType.PARTITIONEDFILE)
{
DrPartitionOutputStream output = new DrPartitionOutputStream();
int err = output.Open(info.sources[0], info.partitionUncPath);
if (!SUCCEEDED(err))
{
string msg = String.Format("Could not open DSC output fileset {0}", info.sources[0]);
throw new LinqToDryadException(msg, err);
}
DrManagerBase outputStage = new DrManagerBase(app.GetGraph(), outputName);
DrOutputStreamManager outputManager = new DrOutputStreamManager(output, outputStage);
app.GetGraph().AddPartitionGenerator(outputManager);
s = outputManager;
}
//else if ( info.ioType == VertexInfo.IOType.STREAM )
//{
// DrDscOutputStream output = new DrDscOutputStream(info.compressionScheme, info.isTemporary);
// int err = 0;
// if (info.recordType == "")
// {
// err = output.Open(info.sources[0], info.partitionUncPath);
// }
// else
// {
// err = output.OpenWithRecordType(info.sources[0], info.partitionUncPath, info.recordType);
// }
// if (!SUCCEEDED(err))
// {
// string msg = String.Format("Could not open DSC output fileset {0}", info.sources[0]);
// throw new LinqToDryadException(msg, err);
// }
// DrManagerBase outputStage = new DrManagerBase(app.GetGraph(), outputName);
// DrOutputStreamManager outputManager = new DrOutputStreamManager(output, outputStage);
// app.GetGraph().AddPartitionGenerator(outputManager);
// s = outputManager;
//}
else if (info.ioType == VertexInfo.IOType.HDFS_STREAM)
{
DrHdfsOutputStream output = new DrHdfsOutputStream();
int err = output.Open(info.sources[0]);
if (!SUCCEEDED(err))
{
string msg = String.Format("Could not open HDFS output fileset {0}", info.sources[0]);
throw new LinqToDryadException(msg, err);
}
DrManagerBase outputStage = new DrManagerBase(app.GetGraph(), outputName);
DrOutputStreamManager outputManager = new DrOutputStreamManager(output, outputStage);
app.GetGraph().AddPartitionGenerator(outputManager);
s = outputManager;
}
else
{
string msg = String.Format("Unknown output type {0}", info.ioType);
throw new LinqToDryadException(msg);
}
return(s);
}
