Skip to content

Event Pipe : Production Diagnostics in the .NET Core Era

What is Event Pipe Diagnostic?

A simple trace and analysis application to showcase, how Event Pipe in .NET Core works for taking a trace of a .NET Core process and how to extract managed thread stack from the resulting trace, which is very useful while analyzing performance issues.

Using .NET Event Pipe for diagnostics

Event Pipe technology of .NET Core runtime is a cross-platform alternative to ETW on Windows and LTTng on Linux since they work only on a single platform. Event Pipe delivers the same experience on Windows, Linux and macOS.

EventPipe can also be used on any .NET Core applications running on .NET Core 3.0 Preview 5 or later.

Diagnostic IPC Protocol in .NET Core runtime listens and communicates over a platform-specific transport. On Unix/Linux based platforms, a Unix Domain Socket will be used, and on Windows, a Named Pipe will be used.

Why Event Pipe for Production Diagnostics?

EventPipe is the best cross-platform alternative to ETW/LTTng and also the functionality is built-into .NET Core runtime, so you don’t need any external components.

Its very useful for analyzing CPU usage, IO, lock contention, allocation rate, etc where you might want to capture a performance trace. This trace can then be moved to a developer machine where it can be analyzed with profiling tools such as PerfView/VisualStudio or visualized as a flame graph with speedscope.

Lets look at Event Pipe Tracing Code

Main portion of the Event Pipe tracing code is below.
Complete code available on Github. Link available at the end of this article.

	var providers = new List<EventPipeProvider>()
		new EventPipeProvider("Microsoft-DotNETCore-SampleProfiler",
			EventLevel.Informational, (long)ClrTraceEventParser.Keywords.All),    
		new EventPipeProvider("Microsoft-Windows-DotNETRuntime",
			EventLevel.Informational, (long)ClrTraceEventParser.Keywords.Default) 
	// Create client
	var diagClient = new DiagnosticsClient(processId);

	// Create session
	using (var eventPipeSession = diagClient.StartEventPipeSession(providers))
		// Write event stream to trace file
		Task readerTask = Task.Run(async () =>
			using (FileStream fs = new FileStream(traceFilename, FileMode.Create, 
				await eventPipeSession.EventStream.CopyToAsync(fs);

		readerTask.Wait(duration * 1000);


EventPipe-Diagnostics project here contains three functionalities as commands.

  1. PS (lists all process running with EventPipe enabled)
  2. Collect trace (collects trace for a defined period of time)
  3. Analyze trace (prints out the manage stack trace)

Did someone say Help!!!

You can use -? option to get help and check all the command line options

> DotNetEventPipe -?
  DotNetEventPipe [options] [command]

  --version         Show version information
  -?, -h, --help    Show help and usage information

  collect    Capture trace from a process using Event Pipe
  analyze    Analyze trace a process and prints managed call stack
  ps         Process list with Event Pipe available for Diagnostics

What does ps command option do?

The command ps shows all the process with EventPipe listener available to connect

> DotNetEventPipe ps
Process with Event Pipe available for Diagnostics
        29624 iisexpress C:\Program Files\IIS Express\iisexpress.exe

What does Collect command option do?

Collect option runs a trace on a selected .NET process which supports EventPipe for a specific duration.

> DotNetEventPipe collect -?
  Capture trace from a process using Event Pipe

  DotNetEventPipe collect [options]

  -p, --processId <pid> (REQUIRED)    The process id to collect the trace for.
  -d, --duration <duration>           Duration in seconds to run the trace.
  -f, --traceFilename <filename>      Trace file name (without path).
  -?, -h, --help                      Show help and usage information

Typical usage of Collect command below

Here we are tracing a process with pid 29624 for a duration of 20 seconds and writing to trace filename ‘newtrace.nettrace’

DotNetEventPipe collect -p 29624 -d 20 -f "newtrace"

Output looks like below

> DotNetEventPipe collect -p 29624 -d 20 -f "newtrace"
Process ID : 29624
Duration   : 20
Trace file : C:\DotNetEventPipe\traces\newtrace.nettrace
[Trace started...]
[Trace completed.]

What does Analyze command option do?

Analyze option reads the trace and prints out managed stack on console.

> DotNetEventPipe analyze -?
  Analyze trace a process and prints managed call stack

  DotNetEventPipe analyze [options]

  -f, --traceFile <filename> (REQUIRED)    Trace file name (without path).
  -?, -h, --help                           Show help and usage information

Typical usage of Analyze command below

Here we pass the trace file for the analysis

DotNetEventPipe analyze -f C:\DotNetEventPipe\traces\newtrace.nettrace

Let us see the output of the above command

> DotNetEventPipe analyze -f C:\DotNetEventPipe\traces\newtrace.nettrace
[Trace analysis started...]

Stack for Thread (4692):
  UNMANAGED_CODE_TIME!System.Threading.ManualResetEventSlim.Wait(int32,value class System.Threading.CancellationToken)!System.Threading.Tasks.Task.SpinThenBlockingWait(int32,value class System.Threading.CancellationToken)!System.Threading.Tasks.Task.InternalWaitCore(int32,value class System.Threading.CancellationToken)!System.Runtime.CompilerServices.TaskAwaiter.HandleNonSuccessAndDebuggerNotification(class System.Threading.Tasks.Task)!System.Runtime.CompilerServices.TaskAwaiter.GetResult()!Microsoft.Extensions.Hosting.HostingAbstractionsHostExtensions.Run(class Microsoft.Extensions.Hosting.IHost)
  ProblemScenarios!ProblemScenarios.Program.Main(class System.String[])

Stack for Thread (39440):
  UNMANAGED_CODE_TIME!System.Threading.SemaphoreSlim.WaitUntilCountOrTimeout(int32,unsigned int32,value class System.Threading.CancellationToken)!System.Threading.SemaphoreSlim.Wait(int32,value class System.Threading.CancellationToken)!System.Collections.Concurrent.BlockingCollection`1[Microsoft.Extensions.Logging.Console.LogMessageEntry].TryTakeWithNoTimeValidation(!0&,int32,value class System.Threading.CancellationToken,class System.Threading.CancellationTokenSource)!System.Collections.Concurrent.BlockingCollection`1+<GetConsumingEnumerable>d__68[Microsoft.Extensions.Logging.Console.LogMessageEntry].MoveNext()!Microsoft.Extensions.Logging.Console.ConsoleLoggerProcessor.ProcessLogQueue()!System.Threading.ThreadHelper.ThreadStart_Context(class System.Object)!System.Threading.ExecutionContext.RunInternal(class System.Threading.ExecutionContext,class System.Threading.ContextCallback,class System.Object)!System.Threading.ThreadHelper.ThreadStart()

Stack for Thread (18720):
  Anonymously Hosted DynamicMethods Assembly!dynamicClass.lambda_method(pMT: 00007FFEE69552C8,class System.Object,class System.Object[])!Microsoft.Extensions.Internal.ObjectMethodExecutor.Execute(class System.Object,class System.Object[])

Stack for Thread (10064):
  ProblemScenarios!testwebapi.Controllers.DiagScenarioController.<deadlock>b__3_1()!System.Threading.ExecutionContext.RunInternal(class System.Threading.ExecutionContext,class System.Threading.ContextCallback,class System.Object)

Stack for Thread (33240):
  ProblemScenarios!testwebapi.Controllers.DiagScenarioController.<deadlock>b__3_1()!System.Threading.ExecutionContext.RunInternal(class System.Threading.ExecutionContext,class System.Threading.ContextCallback,class System.Object)

Stack for Thread (10900):

[Trace analysis completed...]

Graphical UI for viewing the trace

On Windows, .nettrace files can be viewed on PerfView for analysis. For traces collected on other platforms, the trace file can be moved to a Windows machine to be viewed on PerfView.

This is a more focused version provided in source for experiment with the API. For full options to take traces, please check official CLI dotnet-trace

Event Pipe Diagnostics Github project

1 Comment »

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: