Skip to content

Commit

Permalink
Add EventSource docs
Browse files Browse the repository at this point in the history 
This is an initial set of docs for using EventSource. The goal was to
capture most of the beginner + intermediate material from
https://github.com/Microsoft/dotnet-samples/blob/master/Microsoft.Diagnostics.Tracing/EventSource/docs/EventSource.md
For brevity I dropped some info about lesser used functionality
(event localization and ETW channel support) while adding some
more info on usage outside the PerfView+ETW ecosystem. I also
added information on ActivityIDs which I haven't found documented
anywhere.

After this PR goes in I'd like to trim the EventSource class api-docs
and add a link to this content as the main place to show overall usage. I
also want to add a link to this content from the dotnet-samples github
page and mark that one as being no longer maintained.

This PR is intended to address
dotnet/diagnostics#2908 - EventSource start/stop
dotnet/diagnostics#2904 - "fast" eventsource
dotnet/diagnostics#2902 - custom EventSource with dotnet-trace+vs
dotnet/diagnostics#1347 - activity ids

It partially covers dotnet/diagnostics#2445 - special case LogAlways behavior
but I think we should do more in the API docs before marking that one complete
  • Loading branch information
@noahfalk
noahfalk committed Mar 28, 2022
1 parent 6ad0d90 commit 74a1684
Show file tree
Hide file tree
Showing 17 changed files with 1,089 additions and 3 deletions.
283 changes: 283 additions & 0 deletions docs/core/diagnostics/eventsource-activity-ids.md
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,283 @@
---
title: EventSource Activity IDs
description: Understand how to use EventSource Activity IDs
ms.date: 03/24/2022
---

# EventSource Activity IDs

**This article applies to: ✔️** .NET Core 3.1 and later versions **✔️** .NET Framework 4.5 and later versions

## The challenge of logging concurrent work

Long ago a typical application may have been a simple and single-threaded which makes logging straightforward. We
could write each step to a log file in order and then read the log back exactly in the order it was written to
understand what happened. If the app handled requests then being single threaded only one request was handled
at a time. All log messages for request A would be printed in order, then all the messages for B, and so on so it
is easy to read and understand. When apps become multi-threaded that strategy no longer works because multiple
requests are being handled at the same time. However if each request is assigned to a single thread which processes
it entirely we can solve the problem by recording a thread id for each log message. For example a multi-threaded app
might log:

```
Thread Id Message
--------- -------
12 BeginRequest A
12 Doing some work
190 BeginRequest B
12 uh-oh error happened
190 Doing some work
```

By reading the thread ids we know that thread 12 was processing request A and thread 190 was processing request B,
therefore the 'uh-oh error happened' message related to request A. However application concurrency has continued to
grow ever more sophisticated. It is common to use async/await or queueing schemes so that a single request could
be handled partially on many different threads before the work is complete. Thread ids no longer are sufficient to
correlate together all the messages produced for one request. Activity IDs solve this problem. They provide a finer
grain identifier that can track individual requests, or portions of requests, regardless if the work
is spread across different threads.

> [!NOTE]
> The Activity ID concept refered to here is not the same as the System.Diagnostics.Tracing.Activity, despite the
> unfortunately similar naming.
## Tracking work using an Activity ID

We can run the code below to see Activity tracking in action.

```C#
using System.Diagnostics.Tracing;

public static class Program
{
public static async Task Main(string[] args)
{
ConsoleWriterEventListener listener = new ConsoleWriterEventListener();

Task a = ProcessWorkItem("A");
Task b = ProcessWorkItem("B");
await Task.WhenAll(a, b);
}

private static async Task ProcessWorkItem(string requestName)
{
DemoEventSource.Log.WorkStart(requestName);
await HelperA();
await HelperB();
DemoEventSource.Log.WorkStop();
}

private static async Task HelperA()
{
DemoEventSource.Log.DebugMessage("HelperA");
await Task.Delay(100); // pretend to do some work
}

private static async Task HelperB()
{
DemoEventSource.Log.DebugMessage("HelperB");
await Task.Delay(100); // pretend to do some work
}
}

[EventSource(Name ="Demo")]
class DemoEventSource : EventSource
{
public static DemoEventSource Log = new DemoEventSource();

[Event(1)]
public void WorkStart(string requestName) => WriteEvent(1, requestName);
[Event(2)]
public void WorkStop() => WriteEvent(2);

[Event(3)]
public void DebugMessage(string message) => WriteEvent(3, message);
}

class ConsoleWriterEventListener : EventListener
{
protected override void OnEventSourceCreated(EventSource eventSource)
{
if(eventSource.Name == "Demo")
{
Console.WriteLine("{0,-5} {1,-40} {2,-15} {3}", "TID", "Activity ID", "Event", "Arguments");
EnableEvents(eventSource, EventLevel.Verbose);
}
else if(eventSource.Name== "System.Threading.Tasks.TplEventSource")
{
// Activity IDs aren't enabled by default.
// Enabling Keyword 0x80 on the TplEventSource turns them on
EnableEvents(eventSource, EventLevel.LogAlways, (EventKeywords)0x80);
}
}

protected override void OnEventWritten(EventWrittenEventArgs eventData)
{
lock (this)
{
Console.Write("{0,-5} {1,-40} {2,-15} ", eventData.OSThreadId, eventData.ActivityId, eventData.EventName);
if (eventData.Payload.Count == 1)
{
Console.WriteLine(eventData.Payload[0]);
}
else
{
Console.WriteLine();
}
}
}
}
```

When run, this code prints output:

```
TID Activity ID Event Arguments
21256 00000011-0000-0000-0000-00006ab99d59 WorkStart A
21256 00000011-0000-0000-0000-00006ab99d59 DebugMessage HelperA
21256 00000012-0000-0000-0000-00006bb99d59 WorkStart B
21256 00000012-0000-0000-0000-00006bb99d59 DebugMessage HelperA
14728 00000011-0000-0000-0000-00006ab99d59 DebugMessage HelperB
11348 00000012-0000-0000-0000-00006bb99d59 DebugMessage HelperB
11348 00000012-0000-0000-0000-00006bb99d59 WorkStop
14728 00000011-0000-0000-0000-00006ab99d59 WorkStop
```

> [!NOTE]
> There is a known issue where Visual Studio debugger may cause invalid Activity IDs to be generated. Either don't run this sample under the debugger
> or set a breakpoint at the beginning of Main and evaluate the expression 'System.Threading.Tasks.TplEventSource.Log.TasksSetActivityIds = false' in
> the immediate window before continuing to work around the issue.
Using the Activity IDs we can see that all of the messages for work item A have ID `00000011-...` and all the messages for
work item B have ID `00000012-...`. Both work items first did some work on thread 21256, but then each of them continued their work
on separate threadpool threads 11348 and 14728 so trying to track the request only with thread IDs would not have worked.

EventSource has an automatic heuristic where defining an event named _Something_Start followed immediately by another event named
_Something_Stop is considered the start and stop of a unit of work. Logging the start event for a new unit of work creates a new
Activity ID and begins logging all events on the same thread with that Activity ID until the stop event is logged. The ID also
automatically follows async control flow when using async/await. Although it is recommended to use the Start/Stop naming suffixes
you may name the events anything you like by explicitly annotating them using the
<xref:System.Diagnostics.Tracing.EventAttribute.Opcode?displayProperty=nameWithType> property with the first event set to
EventOpcode.Start and the second set to EventOpcode.Stop.

## Logging requests that do parallel work

Sometimes a single request might do different parts of its work in parallel and you both want to be able to group all the log events
as well as the sub-parts. In the example below we simulate a request that does two database queries in parallel and then does some
processing on the result of each query. We want to be able to isolate the work for each query, but also understand which queries belong
to the same request when many concurrent requests could be happening. We model this as a tree where each top-level request is a root and
then sub-portions of work are branches. Each node in the tree gets its own Activity ID, and the first event logged with the new child
Activity ID logs an extra field called Related Activity ID to describe its parent.

Run the following code:

```C#
using System.Diagnostics.Tracing;

public static class Program
{
public static async Task Main(string[] args)
{
ConsoleWriterEventListener listener = new ConsoleWriterEventListener();

await ProcessWorkItem("A");
}

private static async Task ProcessWorkItem(string requestName)
{
DemoEventSource.Log.WorkStart(requestName);
Task query1 = Query("SELECT bowls");
Task query2 = Query("SELECT spoons");
await Task.WhenAll(query1, query2);
DemoEventSource.Log.WorkStop();
}

private static async Task Query(string query)
{
DemoEventSource.Log.QueryStart(query);
await Task.Delay(100); // pretend to send a query
DemoEventSource.Log.DebugMessage("processing query");
await Task.Delay(100); // pretend to do some work
DemoEventSource.Log.QueryStop();
}
}

[EventSource(Name = "Demo")]
class DemoEventSource : EventSource
{
public static DemoEventSource Log = new DemoEventSource();

[Event(1)]
public void WorkStart(string requestName) => WriteEvent(1, requestName);
[Event(2)]
public void WorkStop() => WriteEvent(2);
[Event(3)]
public void DebugMessage(string message) => WriteEvent(3, message);
[Event(4)]
public void QueryStart(string query) => WriteEvent(4, query);
[Event(5)]
public void QueryStop() => WriteEvent(5);
}

class ConsoleWriterEventListener : EventListener
{
protected override void OnEventSourceCreated(EventSource eventSource)
{
if (eventSource.Name == "Demo")
{
Console.WriteLine("{0,-5} {1,-40} {2,-40} {3,-15} {4}", "TID", "Activity ID", "Related Activity ID", "Event", "Arguments");
EnableEvents(eventSource, EventLevel.Verbose);
}
else if (eventSource.Name == "System.Threading.Tasks.TplEventSource")
{
// Activity IDs aren't enabled by default.
// Enabling Keyword 0x80 on the TplEventSource turns them on
EnableEvents(eventSource, EventLevel.LogAlways, (EventKeywords)0x80);
}
}

protected override void OnEventWritten(EventWrittenEventArgs eventData)
{
lock (this)
{
Console.Write("{0,-5} {1,-40} {2, -40} {3,-15} ", eventData.OSThreadId, eventData.ActivityId, eventData.RelatedActivityId, eventData.EventName);
if (eventData.Payload.Count == 1)
{
Console.WriteLine(eventData.Payload[0]);
}
else
{
Console.WriteLine();
}
}
}
}
```

This example prints output such as:

```
TID Activity ID Related Activity ID Event Arguments
34276 00000011-0000-0000-0000-000086af9d59 00000000-0000-0000-0000-000000000000 WorkStart A
34276 00001011-0000-0000-0000-0000869f9d59 00000011-0000-0000-0000-000086af9d59 QueryStart SELECT bowls
34276 00002011-0000-0000-0000-0000868f9d59 00000011-0000-0000-0000-000086af9d59 QueryStart SELECT spoons
32684 00002011-0000-0000-0000-0000868f9d59 00000000-0000-0000-0000-000000000000 DebugMessage processing query
18624 00001011-0000-0000-0000-0000869f9d59 00000000-0000-0000-0000-000000000000 DebugMessage processing query
18624 00002011-0000-0000-0000-0000868f9d59 00000000-0000-0000-0000-000000000000 QueryStop
32684 00001011-0000-0000-0000-0000869f9d59 00000000-0000-0000-0000-000000000000 QueryStop
32684 00000011-0000-0000-0000-000086af9d59 00000000-0000-0000-0000-000000000000 WorkStop
```

In this example we only ran one top level request which was assigned Activity ID `00000011-...`. Then each QueryStart event began
a new branch of the request with Activity IDs `00001011-...` and `00002011-...` respectively. We know these IDs are children
of the original request because both of the start events logged their parent `00000011-...` in the Related Activity ID field.

> [!NOTE]
> You may have noticed the numerical values of the IDs have some clear patterns between parent and child and are not random. Although it
> can assist in spotting the relationship visually in simple cases it is best for tools not to rely on this and treat the IDs as opaque identifiers.
> As the nesting level grows deeper the byte pattern will change. Using the Related Activity ID field is the best way to ensure tools
> work reliably regardless of nesting level.
Because requests with complex trees of sub-work items will rapidly generate many different Activity IDs these IDs are usually
best parsed by tools rather than trying to reconstruct the tree by hand. [PerfView](https://github.com/Microsoft/perfview) is one tool
that knows how to correlate events annotated with these IDs.
Loading

0 comments on commit 74a1684

Please sign in to comment.