Performance Measuring and Monitoring for .NET 6 Applications

Performance Measuring and Monitoring for .NET 6 Applications

https://www.pluralsight.com/courses/dot-net-6-application-performance-measuring-monitoring

by Ivan Gavryliuk

Module 1: Course Overview

Course Overview

Module 2: Basic Principles of Performance Monitoring

Introduction
Why Monitoring

Monitoring is "Process of analyzing, evaluating, and interpreting data results from a measured parameter withing a specified time to verify conformity"
monitoring framework became part of .net starting in .net 6

When Not to Care About Performance

Performance optimization must be driven by business needs
Any work on performance optimization should never be premature or happen before it's needed
designing code for performance makes it more complicated and less maintainable.

What to Monitor

basic metrics: CPU, memory, disk and internet traffic

you don't need to monitor these in .net, it is done by the OS

What's Available

tracing vs logging: tracing is for understanding program flow and investigative performance
tracing: logs many events, perhaps thousands of events per second
tracing: .NET Runtime and SDK provides native tooling for tracing.
.NET itself emits lots of tracing data.
Distributed Tracing: data generated by multiple applications and submitted to some central server

.NET Runtime has built-in support for distributed tracing.

Dump: a snapshot of an application in time saved to a file.

can be analyised by a tool

Summary

Module 3: Tracing and Logging

Introduction

differences between logging and tracing

Tracing Options in Dotnet

ETW (Event Tracing for Windows)

Windows only
High-performance structured logging

EventPipe aggregates trace events from all possible places.

Eventpipe Runtime Component

Every .net app has a diagnostic port

-- Named Pipes on Windows
-- Domain Socket on Linux and Mac

From Windows, in powershell, to list all named pipes:

-- [System.IO.Directory]::GetFiles("\\.\\pipe\\")
-- the relevant ones start with dotnet-diagnostic-xxxxxx

Overview of Standard Event Providers

default trace events from CoreCLR

Microsoft-Windows-DotNETRuntime: GC, assembly loading, exceptions, threads etc.
Microsoft-DotNETCore-SampleProfiIer: CPU sampling for every thread every IO milliseconds.

from nuget

Microsoft-Extensions-Dependencylnjection

Dependency injection in ASP.NET Core etc.

System.Buffers.ArrayPoolEventSource

Array pooling: rent/return etc.

System.Net.Http

Detailed HTTP stack events

System.Net.NameResolution

DNS resolution

System.Net.Sockets

System sockets: connect/accept etc.

App Code and Dotnet Trace

demo of Spectre.Console: a .NET library that makes it easier to create beautiful console applications.
dotnet_trace

to install: dotnet tool install -g dotnet-trace
it can connect to a diagnostics port

dotnet trace ps

this command will get you all dotnet applications with a diagnostic ports open

Tracing Analyzing

dotnet trace collect

can attach by process id, name, port
output is in .nettrace format
it can also output in speedscope format

Adding Custom Trace Provider

shows how to create an EventSource (custom trace provider)

with dotnet trace, you will need to specify it on the command line

shows how to open nettrace with Visual Studio

Summary

Module 4: Distributed Tracing

Introduction

distributed tracing is always on

Terminology

distributed tracing collections information about spans
opentelemetry is a telemetry standard .net implements

Instrumenting Console Application

in .NET spans are called activities for historical purposes
shows how to add activity instrumentation

Instrumenting Distributed Parts

shows an example of console app and web api distrbuted tracing
HttpClient, once wrapped into activity, will transfer its span id over the network as a part of HTTP headers. Then ASP.NET run time will pick it and propagate down the request

Using Jaeger for Trace Collection

Jaeger is a an open source tool for collection distributed data

shows nuget package needed
shows how to enable it from program.cs

Adding More Data

shows how to add tags and events

Summary

Module 5: Core Dumps

Introduction

when to use crash dumps

There are no logs or traces
Issues can't be reproduced
Cannot generate diagnostic information
Process has died
First Chance Exceptions

Out of memory
Stack overflow
Execution engine
Unhandled exceptions

Analyzing Memory Leaks

use 'dotnet-dump' to create a crash dump

to install: dotnet tool install -g dotnet-dump
to list out process that can be dumped: dotnet dump ps
to create a dump: dotnet dump collect -p {process id}
to analyze dump: dotnet dump analyze {name of dump file}

shows an example of a program that has a memory leak

Analyzing Startup Crashes

you can create dumps after a program has already stopped running (for example it crashes on start up)

set environment variable COMPlus_DbgEnableMiniDump to 1 and then run the application

output will show you where dump file was created and saved

Analyzing Linux Dumps in Windows
Summary

Module 6: Application Metrics

Introduction

.NET metrics reporting:

PerformaceCounter
EventCounters
System.Diagnostics.Metrics

System.Diagnostics.Metrics

compatible with OpenTelemetry
thread safe
.NET 6 +

Tracking Realtime Metrics

track real time metrics using 'dotnet-counters'

to install: dotnet tool install g dotnet-counters
to view metrics you can track: dotnet counters list
to view application that can be tracked: dotnet counters ps
to track an app: dotnet counters monitor -p {process id}

Custom Metrics

shows how to add custom counter using System.Diagnostics.Metrics
you will need to specify the custom counter in command line

Types of Instruments

metering api is not limited to numbers

Collecting with OpenTelemetry and Prometheus

Metrics are exposed via diagnostics port.
demo of Prometheus, a popular tool to collects metrics

Summary

Comments