/// <summary>
/// Create a class to diagnose the problems of a job.
/// </summary>
/// <param name="job">Job to diagnose.</param>
/// <param name="plan">Plan of the diagnosed job.</param>
/// <param name="manager">Communication manager.</param>
protected JobFailureDiagnosis(DryadLinqJobInfo job, DryadJobStaticPlan plan, CommManager manager)
: base(job, plan, manager)
{
this.diagnosisLog = new DiagnosisLog(job, job.Summary);
this.jobManager = this.Job.ManagerVertex;
}
/// <summary>
/// Create a suitable Job Failure diagnosis object for the job being analyzed.
/// </summary>
/// <param name="job">Job to diagnose.</param>
/// <param name="manager">Communication manager.</param>
/// <returns>A subclass of JobFailureDiagnosis with the type appropriate for the job.</returns>
/// <param name="plan">Plan of the job being diagnosed.</param>
public static JobFailureDiagnosis CreateJobFailureDiagnosis(DryadLinqJobInfo job, DryadJobStaticPlan plan, CommManager manager)
{
ClusterConfiguration config = job.ClusterConfiguration;
if (config is CacheClusterConfiguration)
config = (config as CacheClusterConfiguration).ActualConfig(job.Summary);
throw new InvalidOperationException("Configuration of type " + config.TypeOfCluster + " not supported for diagnosis");
}
/// <summary>
/// Create a class to diagnose the problems of a vertex.
/// </summary>
/// <param name="vertex">Vertex to diagnose.</param>
/// <param name="job">Job containing the vertex.</param>
/// <param name="plan">Plan of the executed job.</param>
/// <param name="manager">Communication manager.</param>
protected VertexFailureDiagnosis(DryadLinqJobInfo job, DryadJobStaticPlan plan, ExecutedVertexInstance vertex, CommManager manager)
: base(job, plan, manager)
{
this.Job = job;
this.Vertex = vertex;
// ReSharper disable once DoNotCallOverridableMethodsInConstructor
this.stackTraceFile = "dryadLinqStackTrace.txt";
}
/// <summary>
/// Create a VertexFailureDiagnosis of the appropriate type.
/// </summary>
/// <param name="vertex">Vertex to diagnose.</param>
/// <param name="job">Job containing the vertex.</param>
/// <param name="manager">Communication manager.</param>
/// <returns>A subclass of VertexFailureDiagnosis.</returns>
/// <param name="plan">Plan of the executed job.</param>
public static VertexFailureDiagnosis CreateVertexFailureDiagnosis(DryadLinqJobInfo job,
DryadJobStaticPlan plan,
ExecutedVertexInstance vertex,
CommManager manager)
{
ClusterConfiguration config = job.ClusterConfiguration;
if (config is CacheClusterConfiguration)
config = (config as CacheClusterConfiguration).ActualConfig(job.Summary);
throw new InvalidOperationException("Config of type " + config.TypeOfCluster + " not handled");
}
/// <summary>
/// Try to find the job information from cluster and summary.
/// </summary>
/// <param name="manager">Communication manager.</param>
protected void FindJobInfo(CommManager manager)
{
DryadLinqJobInfo jobinfo = DryadLinqJobInfo.CreateDryadLinqJobInfo(this.cluster, this.Summary, true, manager);
if (jobinfo == null)
{
manager.Status("Cannot collect information for " + Summary.ShortName() + " to diagnose", StatusKind.Error);
return;
}
this.Job = jobinfo;
this.StaticPlan = JobObjectModel.DryadJobStaticPlan.CreatePlan(jobinfo, manager);
}
/// <summary>
/// Create a FailureDiagnosis object.
/// </summary>
/// <param name="job">Job being diagnosed.</param>
/// <param name="plan">Static plan of the job.</param>
/// <param name="manager">Communication manager.</param>
protected FailureDiagnosis(DryadLinqJobInfo job, DryadJobStaticPlan plan, CommManager manager)
{
this.Job = job;
this.StaticPlan = plan;
this.Manager = manager;
this.Summary = job.Summary;
this.cluster = job.ClusterConfiguration;
}
/// <summary>
/// Continuation for refreshing the query plan.
/// </summary>
private void PlanComputed(bool cancelled, DryadJobStaticPlan plan)
{
if (cancelled) return;
this.staticPlan = plan;
if (this.staticPlan != null)
{
this.staticPlan.AddFictitiousStages();
this.staticGraph = this.BuildAGLGraph();
}
this.CreatePlanLayouts();
this.AssignPlanColors();
if (this.planVisible != PlanVisible.Static)
this.FitPlanToWindow();
this.DrawQueryPlan();
}
/// <summary>
/// Build an AGL graph corresponding to a dryadlinq job static plan.
/// </summary>
/// <returns>An AGL graph.</returns>
private static Msagl.Drawing.Graph BuildAGLGraph(DryadJobStaticPlan plan)
{
Msagl.Drawing.Graph retval = new Msagl.Drawing.Graph();
foreach (DryadJobStaticPlan.Stage stage in plan.GetAllStages())
{
Msagl.Drawing.Node node = new Msagl.Drawing.Node(stage.Id.ToString());
node.UserData = stage;
if (stage.IsVirtual)
{
if (stage.IsTee)
node.Attr.Shape = Microsoft.Msagl.Drawing.Shape.Octagon;
else
node.Attr.Shape = Msagl.Drawing.Shape.InvHouse;
}
else
{
if (stage.Name == "JobManager")
{
node.Attr.Shape = Microsoft.Msagl.Drawing.Shape.Diamond;
}
else if (stage.Name == "All vertices")
node.Attr.Shape = Microsoft.Msagl.Drawing.Shape.Box;
else
node.Attr.Shape = Microsoft.Msagl.Drawing.Shape.Ellipse;
}
string nodeName = stage.Name;
if (stage.IsInput || stage.IsOutput)
{
nodeName = string.Join(",", nodeName.Split(',').Select(Path.GetFileName).ToArray());
nodeName = Path.GetFileName(nodeName);
const int maxNodeNameLen = 40;
if (nodeName.Length > maxNodeNameLen)
{
nodeName = nodeName.Substring(0, maxNodeNameLen / 2) + "..." + nodeName.Substring(nodeName.Length - maxNodeNameLen / 2);
}
}
if (stage.Replication != 1)
node.LabelText = stage.Replication + " x " + nodeName;
else
node.LabelText = nodeName;
retval.AddNode(node);
}
foreach (DryadJobStaticPlan.Connection connection in plan.GetAllConnections())
{
Msagl.Drawing.Edge e = retval.AddEdge(connection.From.Id.ToString(), connection.To.Id.ToString());
if (connection.Arity == DryadJobStaticPlan.Connection.ConnectionType.AllToAll)
e.Attr.LineWidth = 3;
if (connection.ConnectionManager != "None")
e.LabelText = connection.ConnectionManager;
e.Attr.Color = FormColorToAglColor(System.Drawing.Color.FromName(connection.Color()));
e.UserData = connection;
}
return retval;
}
/// <summary>
/// Create a class representing the set of edges between two stages in the job plan.
/// These collectively look like a partitioned table.
/// If the source stage has multiple outputs there is not enough information to return meaningful information.
/// </summary>
/// <param name="job">Job whose slice we are displaying.</param>
/// <param name="plan">Static plan of the job.</param>
/// <param name="source">Stage in the job which produces the data.</param>
/// <param name="status">Delegate used to report errors.</param>
/// <param name="showCancelled">If true include cancelled vertices.</param>
public static StaticPartitionedTableInformation StageOutput(
DryadLinqJobInfo job,
DryadJobStaticPlan plan,
DryadJobStaticPlan.Stage source,
StatusReporter status,
bool showCancelled)
{
string header = "Output of " + source.Name;
// First check whether in the static plan this is virtual
while (source.IsTee)
{
var sourceInputs = plan.GetStageConnections(source, true).ToList();
if (sourceInputs.Count() != 1)
throw new DryadException("Unexpected number of inputs for stage " + source.Name);
source = sourceInputs.First().From;
}
// If we reached the input return information about that input
if (source.IsInput)
{
status("Scanning " + source.Name, StatusKind.LongOp);
StaticPartitionedTableInformation result = new StaticPartitionedTableInformation(job.ClusterConfiguration, source.UriType, source.Uri, source.Code, status);
result.Header = "Output of " + header;
result.constructorArguments = new SaveConstructorArguments
{
code = null,
source = source,
plan = plan
};
return result;
}
else
{
StaticPartitionedTableInformation result = new StaticPartitionedTableInformation();
result.Name = "Output of vertices in stage " + source.Name;
result.Header = "Output of " + header;
result.constructorArguments = new SaveConstructorArguments
{
code = null,
source = source,
plan = plan
};
// Check whether this stage has multiple outputs; this can only happen for 'Fork' operators.
var destinations = plan.GetStageConnections(source, false);
if (destinations.Count() > 1)
{
result.Error = "Cannot provide information about one of multiple outputs of a stage.";
return result;
}
DryadLinqJobStage stage = job.GetStage(source.Name);
if (stage == null)
{
result.Error = "There is no information about the output of stage " + source.Name;
return result;
}
result.EstimatedSize = 0;
result.PartitionCount = stage.TotalInitiatedVertices;
int count = 0;
foreach (ExecutedVertexInstance vi in stage.Vertices)
{
if (vi.State == ExecutedVertexInstance.VertexState.Successful ||
vi.State == ExecutedVertexInstance.VertexState.Failed ||
(showCancelled && vi.State == ExecutedVertexInstance.VertexState.Cancelled))
{
StaticPartitionInformation spi = new StaticPartitionInformation(count++, vi.DataWritten, vi.Name + " v." + vi.Version);
result.AddPartition(spi);
if (vi.DataWritten != -1)
result.EstimatedSize += vi.DataWritten;
}
}
return result;
}
}
