appmetrics/appmetrics

App Metrics is an open-source and cross-platform .NET library used to record and report metrics within an application.

App Metrics

What is App Metrics?

App Metrics is an open-source and cross-platform .NET library used to record metrics within an application. App Metrics can run on .NET Core or on the full .NET framework. App Metrics abstracts away the underlaying repository of your Metrics for example InfluxDB, Graphite, Prometheus etc, by sampling and aggregating in memory and providing extensibility points to flush metrics to a repository at a specified interval.

App Metrics provides various metric types to measure things such as the rate of requests, counting the number of user logins over time, measure the time taken to execute a database query, measure the amount of free memory and so on. Metrics types supported are Gauges, Counters, Meters, Histograms and Timers and Application Performance Indexes Apdex.

App.Metrics includes an Exponentially Forward Decaying, Sliding Window and Algorithm R reservoir implementations. For more details on reservoir sampling see the docs.

Documentation

• Getting Started
• ASP.NET Core 2.0
• Reporting
• Sample Applications & Grafana Dashboards

Latest Builds, Packages & Repo Stats

Branch	Azure Devops
dev
main

---

Visualization

Dashboards can be imported from Grafana

Grafana Web Monitoring

Grafana OAuth2 Client Web Monitoring

How to build

Azure Devops builds are triggered on commits and PRs to the dev branch

• Install the latest .NET Core 2.x SDK
• Run build.ps1 or build.sh in the root of the repository

How to run benchmarks

App.Metrics includes benchmarking using BenchmarkDotNet.

Two benchmark projects exist targeting App.Metrics.Core and App.Metrics.Concurrency

	cd .\src\Core\benchmarks\App.Metrics.Benchmarks.Runner
	dotnet run -c "Release" --framework netcoreapp3.1

	cd .\src\Concurrency\benchmarks\App.Metrics.Concurrency.Benchmarks.Runner
	dotnet run -c "Release" --framework netcoreapp3.1

You'll then be prompted to choose a benchmark to run which will output a markdown file with the result in directory.

You can find the benchmark results here and here.

Contributing

See the contribution guidlines for details.

Acknowledgements

• ASP.NET Core
• Grafana
• Fluent Assertions
• xUnit.net
• Hugo
• Netlify

Thanks for providing free open source licensing

• NDepend
• Jetbrains

License

This library is release under Apache 2.0 License ( see LICENSE ) Copyright (c) 2016 Allan Hardy

See LICENSE

---

App Metrics is based on the Metrics.NET library, and at the moment uses the same reservoir sampling code from the original library which is a port of the Java Dropwizard Metrics library.

Metrics.NET Licensed under these terms: "Metrics.NET is release under Apache 2.0 License Copyright (c) 2014 Iulian Margarintescu" see LICENSE

Dropwizard Metrics Licensed under these terms*: "Copyright (c) 2010-2013 Coda Hale, Yammer.com Published under Apache Software License 2.0, see LICENSE"

appmetrics's Issues

Add Apdex Reporting to InfluxDB

Report apdex scores in the infuxdb reporter

Add pre-defined HealthChecks

Add health checks to the framework for common use cases

Report health checks to influxdb

Update the influxdb report provide to flush health checks to influxdb for visualizing with grafana

Allow metric filtering via URL on metrics endpoint

Compose a metrics filter from the /metrics url to allowing filtering on the response

Review EWMA

InfluxDB Reporting API test app

Create a sample api with the TestClock configured to confirm metric values are reported correctly when using test scripts .

Adding to sample to have a visual representation of expected beviour as well as unit tests.

Allow reporters to format metric names

Metric name formatting depends on the timeseries database reporting to, it would be good to be able to provide a way for reporters to format metric names on reporting

Apdex Implementation

https://en.wikipedia.org/wiki/Apdex

• Should apdex be it's own Metric type or part of Histograms?
• Just calculate an api's overall apdex or allow an apdex per histogram?
• Separate endpoint for apdex scores?
• Include apdex metrics in health endpoint?
• Include apdex score per histogram in metrics endpoint or separate apdex section?

Provide more reporting options on Meters

Similar to counters allow disabling of flushing percentages and items when reporting

Add "Degraded" health results

Rather than just reporting healthy or un-healthy allow checks to report a degraded status

[Improve Docs] Document available metric types

Measure the percentage of web requests which failed

• Measure the of the overall error rate percentage of web requests as a Gauge
• Update api.sample grafana/influxdb dashboard

Review/Cleanup they way Env info is recorded at set as tags

Code Coverage Report

Add code coverage report on build and github badge

Support .NET 4.5.2

Project's are using .NET Standard can .NET 4.5.2 be supported as well in standard?

[Improve Docs] - Add code sample projects to docs

to https://github.com/alhardy/AppMetrics/tree/dev/docfx_project for reference by docfx

Track reporting errors in InfluxDB report provider

Add metric tracking internal to the influxdb reporter provider to monitoring success and failures in reporting metrics to influxdb

Apdex usage documentation

Metrics text endpoint doesn't show counter and meter items

Fixed as part of #40 Counter and meter items are show, but the endpoint was not report current values

[Improve Docs] Document Sampling Types

Concurrency issue when stress testing api sample with influxdb reporting

Review sampling resevoirs

https://en.wikipedia.org/wiki/Reservoir_sampling

• Algorithm R (Uniform Reservoir Sampling)
• Forward Decaying (Exponentially Decaying Reservoir Sampling)
• Sliding Window - like Algorithm R but producing quantiles which are representative of the past n measurements
• HdrHistogram - designed for recoding histograms of value measurements in latency and performance sensitive applications. Measurements show value recording times as low as 3-6 nanoseconds on modern Intel CPUs

Record Metrics provided by ETW events

[Application.OAuth2Client.HttpRequests] Http Requests Items metrics should be tagged with client id

[Application.OAuth2Client.HttpRequests] Http Requests Items meters should by tagged with client id for grouping

Review Concurrency Utils, Add tests and benchmarks

App crashes on startup when AddReporting is not used

When not reporter is configured through AddReporting and attempt is made resolve the IReporterFactor which fails.

HdrHistogram

Fork https://github.com/HdrHistogram/HdrHistogram.NET and port to .NET Standard

HdrHistogram/HdrHistogram.NET#17

This can replace the current HdrHistogram implmentation

Grafana/InfluxDB Console App Reporting Samples

Dashboard Template for conole sample projects. Host InfluxDB and Graphite for demo

Appveyor build badges for dev and master both show last build status rather than the build for the specific branch

change this to show a single build status badge which points to the latest appveyor build

Migrate xproj to csproj for VS2017

Separate template name resolvers (TemplateRoute vs AttributeRoute) to remove dependency on MVC

duplicate of #11

Grafana dashboard showing the difference between reservoir types

Creating sample for this to validate the expected behavior of each reservoir type on top have unit tests, results are much clearer in a graph

• Add metrics with different types of reservoir sampling to the InfluxDB api sample
• Create a Grafana dashboard showing each reservoir type
• Add to documentation

InfluxDB Report Provider

Exception Monitoring

record metrics on uncaught exception types

NBench Tests

Finialize all api contracts

Review the IMetrics contract to avoid breaking changes before bumping to rc.

• the default IMetrics implementation includes measuring all metrics, seperate this by metric type and expose an api into measurement methods by metric type to allow better discoverabily and a clearer api.
• move sampling implementation from appmetrics core project to its own, may replace some functionality in future with hdrhistogram.net and math.net packages.
• potentially expose hit ratio and other gauge types on the meyrics api for better discover and ease of use.
• IAdvancedMetrics can be improved in the same way as mentioned in the first point above.

Meter and Timer rates should default unit type of mins

When measure throughput rates, this is normally done as requests per min (rpm) so the default rate unit for all timers and meters should be mins

Metrics Text Endpoint Middleware Never Resets the StringReporter buffer

Calling the /metrics-text endpoint appends to the previous output in the 1.0.0-beta2 build using .NET Core.

***** Start - Health Checks *****


	Health Status = {status}

	PASSED CHECKS
	DEGRADED CHECKS
	FAILED CHECKS
***** End - Health Checks *****

-- End String Reporter Report: 2016-12-22T21:41:31.0203Z --

-- Start String Reporter Report: 2016-12-22T21:41:33.2383Z --

***** Start - Environment Information *****

     Assembly Name = src
  Assembly Version = 1.0.26.0
         Host Name = platform-site-2411001161-rsmc3
        Ip Address = 10.233.71.7
        Local Time = 2016-12-22T21:41:33.2406+00:00
      Machine Name = .
                OS = debian
        OS Version = 8

        public void ConfigureServices(IServiceCollection services)
        {
            services.AddLogging();
            services.AddMetrics()
                .AddJsonSerialization()
                .AddHealthChecks()
                .AddMetricsMiddleware();
        }

        public void Configure(IApplicationBuilder app, IHostingEnvironment env, ILoggerFactory loggerFactory)
        {
            app.UseMetrics();
        }

Reporters stop under load and don't start again until app restart

To replicate: run Api.InfluxDB.Sample and send ~100 concurrent request and notice reporting to influxdb stops and the reporter is no longer running.

Add project for benchmarking and include benchmarks for core functionality

• Setup Benchmarking Framework
• Implement Benchmarks for measuring metric types
• Implement Benchmarks for reservoir sampling types

Measure the web request error rate as a percentage of overall web requests

• Measure the of the overall error rate percentage of web requests using a meter, request error rate percentages can also be tracked per endpoint
• Update api.sample grafana/influxdb dashboard

Provide a way to have counters reset when reporting

An example is a count, which is cumulative, in most cases we would want this count reset after flushing to a datastore, this could be implemented as a type of counter.

Update counter type docs.

Benchmark dotnet and build integration

Measure API's/Websites overall Apdex

Updated middleware to record the apdex, apdex T seconds should be configurable

Separate template name resolvers

(TemplateRoute vs AttributeRoute) to remove dependency on MVC

MemoryDiagnoser Benchmarks

Add MemoryDiagnoser to sampling tests using BenchmarkDotNet

Review/Stabilize Reporting framework

[Investigation] Expose simplified metric models to reporters than than valuesource metrics

Grafana/InfluxDB API Reporting Samples

Dashboard Template for Api.Sample project

Metrics 2.0 spec

http://metrics20.org/spec/ + http://dieter.plaetinck.be/post/monitorama-pdx-metrics20/

The benefits from the canonical list of keys include:

easier to define checks, queries, and dashboards because of metric naming consistency
easier to write tools against the more consistent way of reasoning about metrics
easier to define rollup patterns, and more flexible, since one can decide to keep particular dimensions (like cloud server id) for a short time (since they're more ephemeral and only relevant short term), and other dimensions for a longer time
Netflix's Atlas monitoring tool independently seems to arrived at the same sort of conclusions as the spec's author.