kdcllc / bet.aspnetcore Goto Github PK

There is a need to be able to utilize this library on local development with Docker containers Azure/azure-sdk-for-net#4933.

Developing a CLI tool to support this scenario, where Docker Image would retrieve authentication token based on Visual Studio.Net Credential store.

The problem when running/scoring an ML.NET model in multi-threaded applications comes when you want to do single predictions with the PredictionEngine object and you want to cache that object (i.e. as Singleton) so it is being reused by multiple Http requests (therefore it would be accessed by multiple threads). That’s is a problem because the Prediction Engine is not thread-safe ML.NET issue, Nov 2018.

• How-To: Serve Machine Learning Model Through ASP.NET Core Web API
• How to optimize and run ML.NET models on scalable ASP.NET Core WebAPIs or web apps
  REPO corresponding to this article

When multiple configurations providers are used it is hard to troubleshoot what values is being set for a certain Option within the application especially when Azure Vault or other secure vaults are used to provide an application with the configurations.

Desired outcome is to enable this kind of logging on the program start:

            WebHost.CreateDefaultBuilder(args)
                .ConfigureAppConfiguration((hostingContext, configBuilder) =>
                {
                    var configuration = configBuilder.Build();

                    var logFactory = new LoggerFactory()
                                .AddConsole(LogLevel.Debug)
                                .AddDebug();

                    var logger = logFactory.CreateLogger(nameof(Program));
                    configuration.DebugConfigurations(logger);
                })

                .UseStartup<Startup>()
                .ConfigureKestrel(a => a.AddServerHeader = false);

This issue depends on the implementation of https://github.com/kdcllc/dabarblog/issues/8.

Migration of LoggerFactory https://docs.microsoft.com/en-us/aspnet/core/migration/logging-nonaspnetcore?view=aspnetcore-2.2

In preparation to AspNetCore 3.0, it is imperative to split Bet.Extensions.AzureStorage AspNetCore functionality to Bet.AspNetCore.AzureStorage libray.

When Secret Id and App Id is present the exception is raised:

Microsoft.Azure.KeyVault.Models.KeyVaultErrorException
 HResult=0x80131500
 Message=Operation returned an invalid status code 'Forbidden'
 Source=Microsoft.Azure.KeyVault

This was due to the logic of not catching generic exceptions and also the need to remove all azure vault sources from the configurations.

#34

1. Implement ModelCreationBuilder

 public abstract class ModelCreationBuilder<TInput, TOutput, TResult> : IModelCreationBuilder<TInput, TOutput, TResult>

2. Copy trainingPipeline generated code from

  public static IEstimator<ITransformer> BuildTrainingPipeline(MLContext mlContext)

To newly created class

 public override TrainingPipelineResult BuildTrainingPipeline()
        {
            return BuildTrainingPipeline(() =>
            {
                 // STEP 2: Common data process configuration with pipeline data transformations
                var dataProcessPipeline = MLContext.Transforms.Text.FeaturizeText(outputColumnName: "Features", inputColumnName: nameof(SentimentIssue.Text));

                // STEP 3: Set the training algorithm, then create and config the modelBuilder
                var trainer = MLContext.BinaryClassification.Trainers.SdcaLogisticRegression(labelColumnName: "Label", featureColumnName: "Features");
                var trainingPipeline = dataProcessPipeline.Append(trainer);

                return new TrainingPipelineResult(trainingPipeline, trainer.ToString());
            });
        }

ML.NET 1.3.1 release notes

.NET Core 3.0.0 Preview 6 - June 12, 2019

AspNetCore feed

Still not supported in this release

Microsoft.AspNetCore.Mvc.Versioning, Version=3.1.0.0 , ApplyAsync does not have an implementation in .NET Core 3.0 preview6-012264

[BUG] - AspNet Core 3.0 #499

Announcing .NET Core 3.0 Release Candidate 1

ASP.NET Core and Blazor updates in .NET Core 3.0 Release Candidate 1

At the present time, the logic for validation of the Configurations will not return all of the registered validation errors. The desired outcome is to allow to loop thru all of the registered configurations and return a collection of the validation exceptions.

Add Library that supports SPAM validation for the forms submissions. Initial library can include implementation of Google ReCapture Functionality then add ML.NET library with machine learning validation which would be independent of the external services. BinaryClassification_SpamDetection.

https://devblogs.microsoft.com/dotnet/announcing-ml-net-1-2-and-model-builder-updates-machine-learning-for-net/

Add IModelStorageProvider

To add the ability to execute multiple models via schedule and store model and also evaluate results.

Azure introduced Web App Containers but WebJobs are not supported thus LetsEncrypt webjob is not an option for such hosting scenarios for custom certificates.

Nate Mcmaster LetsEncrypt is archived and no longer provides with support. So use some of the ideas to create a middleware that runs the communication protocol of Authentication and stores actual certificate either on the file system or file specified storage.

• On App starts loading Program.cs, checking the validity and if needed Acquiring a certificate
• Hosted job to check for certificate expiration

Desired designed features

• Storing Let's Encrypt SSL Certificates in Azure Vault, file system, and Web Site App.
• Logic to renew SSL certificate
• Support for wild card SSL certificates

In real life senarios, ML models should be built or rebuilt within a certain interval of time for example as new prediction occurs, we would like to retrain the model to reflect the newly collected data. While AutoML tool is great for prototyping it doesn't address the issue of the production training of the models.

• Specs for the ML.NET CLI with AutoML

• ML.NET Model Builder Guide

• Microsoft.ML.AutoML

• Q: ML Model Builder

• Visual Studio.NET ML.NET Extension Preview

Add separate sample project

• To include the start jobs for model generations.
• To include job for app onstart to load the existing model for performance
  Bet.ML.WebApi.Sample.csproj

Define a fluid injectable interface for building ML models on the server side or as Generic Host within Docker Containers hosted in Kubernates.

ML.NET introduced AutoML dotnet/machinelearning#2693

The latest dotnetcore 3.0 preview 7 is released and it comes with production support, so it seems that Microsoft already using it for hosting their websites.

Download and Install

• Announcing .NET Core 3.0 Preview 7

• ASP.NET Core and Blazor updates in .NET Core 3.0 Preview
  Full list of breaking changes in ASP.NET Core 3.0
  API Versioning support

• Announcing Entity Framework Core 3.0 Preview 7 and Entity Framework 6.3 Preview 7

Update HealthChecks with IHostApplicationLifetime dotnet/aspnetcore#7749

Possibly adding a simple interface base on Injecting a Factory Service in ASP.NET Core and the idea of using Func<> as DI like so:

services.AddScoped<ServiceA>();  
    services.AddScoped<ServiceB>();  
    services.AddScoped<ServiceC>();  
  
    services.AddTransient<Func<ServiceEnum, IService>>(serviceProvider => key =>  
    {  
        switch (key)  
        {  
            case ServiceEnum.A:  
                return serviceProvider.GetService<ServiceA>();  
            case ServiceEnum.B:  
                return serviceProvider.GetService<ServiceB>();  
            case ServiceEnum.C:  
                return serviceProvider.GetService<ServiceC>();  
            default:  
                return null;   
        }  
    });  

       public ValuesController(Func<ServiceEnum, IService> serviceResolver)  
       {  
           var service = serviceResolver(ServiceEnum.A);  
       }  

Based on Registering multiple implementations of the same interface in ASP.NET Core

Generic sets of extensions to assist with building and displaying Model Generation information.
Abstract each model into its own self-contained library.

The extension method should provide a way to registers configuration as follows:

 "ParentSetting": {
    "ChildSetting": {
     }
 }

then to configure this via services

services<ChildObject,ChildObjectSettings>("ParentSetting");

Example usage

Two options and ideas

1. http://blogs.msdn.com/b/pfxteam/archive/2012/02/12/10266988.aspx
2. https://github.com/filipw/async-expiring-lazy or article https://www.strathweb.com/2016/11/lazy-async-initialization-for-expiring-objects/

• Upgrade the ML.NET core to support model reload if needed.
• Update PredictionEnginePoolPolicy to support WeakReference as storage.

dotnet/extensions#771

https://docs.microsoft.com/en-us/aspnet/core/security/key-vault-configuration?view=aspnetcore-2.2#reload-secrets

• Using the Microsoft.AspNetCore.HealthChecks Package

Some project Ideas:

• Implementing a Docker HEALTHCHECK using ASP.Net Core 2.2

• How to set up ASP.NET Core 2.2 Health Checks with BeatPulse's AspNetCore.Diagnostics.HealthChecks

• AKS-Angular-AspnetCore-SQLServer

• Kubernetes liveness and readiness probes using HealthChecks

• HealthCheckExample

• Creating singleton named options with IOptionsMonitor

• Running async tasks on app startup in ASP.NET Core (Part 4 - using health checks)

HealthChecks in General

• Healthy - If the microservice/web app is accessible via HTTP and its related services (i.e. Sql Server) is also avaiable. Health checks in ASP.NET Core and Health monitoring

SIGTERM HealthCheck based on IApplicationLifetime

 if (appLifetime.ApplicationStopping.IsCancellationRequested)
                    {
                        return HealthCheckResult.Unhealthy();
                    }

                    return HealthCheckResult.Healthy();

1. AppAuthentication adding support for localhost
   There are scenarios where it is beneficial to enable MSI for localhost testing.
   All possible Connection string options can be found here.
   Examples:

RunAs=Developer; DeveloperTool=AzureCli
RunAs=Developer; DeveloperTool=VisualStudio

Usage:

 appauthentication run -a -l

2. Add LogLevel functionality

Modify all possible switches

• --authority:authid or -a:authid
• --verbose:debug
• --token-provider:AzureCli or -t:VisualStudio (default VisualStudio)
• --environment:Production or -e:Development
• --resource:test or -r:someresource
• --port:1010 or -p:2323
• --config:file or -c:appsettings.config
• --fix or -f
• --local or -l (default Docker)

domaindrivendev/Swashbuckle.AspNetCore#626

Asp.net Core configuration system is very flexible and it is helpful to have validation before the apps starts.

Some ideas:

• Asp.net Core 2.2 Options Validation

• Validating configuration in ASP.NET Core

• Adding validation to strongly typed configuration objects in ASP.NET Core

• ASYNCHRONOUS INITIALIZATION IN ASP.NET CORE, REVISITED

Better yet to implement this https://github.com/aspnet/Extensions/issues/459#issuecomment-455893880.

Previous usage:

 services.ConfigureWithDataAnnotationsValidation<FakeOptionsWithDatatAnnotations>(sectionName: "FakeOptions");

Throws exception even if there are valid values

Value = '((Microsoft.Extensions.Options.OptionsManager<Bet.AspNetCore.UnitTest.FakeOptionsWithDatatAnnotations>)options).Value' threw an exception of type 'Microsoft.Extensions.Options.OptionsValidationException'

Should be:

  services.ConfigureWithDataAnnotationsValidation<FakeOptionsWithDatatAnnotations>(Configuration, sectionName: "FakeOptions");

Must add a positive unit test scenario as well to test a happy path.

• Upgrade to 3.0.100-preview5-011568
• Update sources.props to Daily Builds Sources

ApiVersion is broken in this preview5 services.AddApiVersioning()

• Support for standard kdcllc docker containers to build and run.

• Support for VSCode remote development in the container

Notes for the upgrade ahead...

1. Upgrade Program.cs to support IHost builder impacts the DI for Startup.cs. The is no way to inject ILogger. .

• Adding Serilog to the ASP.NET Core Generic Host
• HostBuilder replaces WebHostBuilder

2. Upgrade Routing

• Update routing startup code

• Endpoint Routing updates starting Preview 4

• Routing in ASP.NET Core

• HealthChecks

                .Configure(app =>
                {
                    app.UseEndpoints(endpoints =>
                    {
                        endpoints.MapHealthChecks("/healthy");
                    });
                })

3. Upgrade To Nullable-references

In many cases, SSL Certificates expire and the health check of the site can be invalided based on that check.

This can be useful in conjunction with a custom publisher to request a new SSL for example for Let's Encrypt Authority.

• Announcing .NET Core 3.0
• ASP.NET Core and Blazor updates in .NET Core 3.0
• Announcing Entity Framework Core 3.0 and Entity Framework 6.3 General Availability
• https://blog.zhaytam.com/2019/09/13/whats-new-asp-net-core-3/

Refactor the code that is being replicated right now for:

• SentimentModelBuilderService
• SpamModelBuilderService

beta package

dotnet add package Bet.Extensions.ML --version 1.3.1-beta1 --source https://www.myget.org/F/kdcllc/api/v3/index.json

Currently OptionsValidationStartupFilter impelments only WebHost Startup validation on configure.

public class OptionsValidationStartupFilter : IStartupFilter

Add unit test to test for Generic host test.

• Announcing ML.NET 1.4 Preview and Model Builder updates (Machine Learning for .NET)

To add Serilog sinks to support the existing Microsoft Logging interfaces.

1. Enable Serilog - A Serilog provider for Microsoft.Extensions.Logging, the logging subsystem used by ASP.NET Core.

• Serilog logging for ASP.NET Core

• A Serilog settings provider that reads from Microsoft.Extensions.Configuration, .NET Core's appsettings.json file.

• A sink for Serilog that writes events to Microsoft Application Insights.

• High performance Serilog sink that writes to Azure Log Analytics

https://docs.microsoft.com/en-us/dotnet/machine-learning/how-to-guides/machine-learning-model-predictions-ml-net#batch-prediction

Simply support for Clean Architecture.

Add library that abstract interaction with Azure Storage Blob and Queue.

• Injectable named StorageAccountOptions that provides with the ability to create CloudStorageAccount based on the configuration provided via configurations or inline.

• IStorageBlob<TOptions> where TOptions : StorageBlobOptions

• IStorageQueue<TOptions> where TOptions : StorageQueueOptions

1. The default name for the Azure Log Analytics was not checking for null and the name for the app had an issue with characters being in the name.

2. Update serilog-sinks-azure-analytics which will require the following being set up.
   For ApplicationInsights Sink to work we need to make sure that we add in Startup.cs file:

            var instrumentId = Configuration.Bind<ApplicationInsightsOptions>("ApplicationInsights",true);

            services.AddApplicationInsightsTelemetry(options =>
            {
                options.InstrumentationKey = instrumentId.InstrumentationKey;
            });

Methos was introduced in 3.0 include this for now in the project GetDebugView

Proposed name Bet.Extensions.AzureTable and support thru Microsoft.Azure.Cosmos.Table

• https://github.com/microsoft/m365-sample-connector-csharp-aspnet/blob/eab656cbb824e730c4f66e889607e622124c3e43/Common/ConnectorCommon/AzureTableProvider.cs#L97

• https://github.com/microsoft/data-accelerator/blob/d7ec69df00d4dcd33ebc70cef307e97f770ca212/Services/DataX.Utilities/DataX.Utility.Blob/TableExtensions.cs

In preparation for the 3.0 release of DotNetCore the libraries must not share IHost and IWebHost in the same libraries. Proposed separation:

1. Bet.Extensions generic functionality for DotNetCore in general.

   • Bet.Extensions.Options - includes Application Options and Bind() validations.
   • Bet.Extensions.Logging - includes shared/common logging functionality.
   • Bet.Extensions.Hosting - includes Generic Hosting functionality.
   • Bet.Extensions.AzureVault - includes Azure Vault functionality.
   • Bet.Extensions - includes extensions methods for DotNetCore.

2. Bet.AspNetCore specific functionality for web applications.

   • Bet.AspNetCore.HealthChecks contains HealthChecks for most common scenarios of the web application.
   • Bet.AspNetCore.Logging contains logging functionality for AspNetCore applications.
   • Bet.AspNetCore - default location for AspNetCore.

Add Azure Vault Key Configuration provider that can be used with MSI and without MSI token provider.

-How to use managed identities for App Service and Azure Functions

-Add method to get an access token

-Using Azure KeyVault to manage secrets of eShopOnContainers

-Connection String Support

Everything can be done thru:

<PackageReference Include="Microsoft.Extensions.Configuration.AzureKeyVault" Version="2.2.0" />

This what the alternative code would look like:

       var kv = new KeyVaultClient(new KeyVaultClient.AuthenticationCallback(GetToken));
        configBuilder.AddAzureKeyVault(options.BaseUrl, kv, new DefaultKeyVaultSecretManager());

        public static async Task<string> GetToken(string authority, string resource, string scope)
        {
            var options = Configuration.Bind<AzureVaultOptions>("AzureVault");

            var authContext = new AuthenticationContext(authority);
            ClientCredential clientCred = new ClientCredential(options.ClientId, options.ClientSecret);
            AuthenticationResult result = await authContext.AcquireTokenAsync(resource, clientCred);

            if (result == null)
                throw new InvalidOperationException("Failed to obtain the JWT token");

            return result.AccessToken;
        }

Issue of using outside of MSI

Get access token using Microsoft.Azure.Services.AppAuthentication in a container

The goal of the project is to provide model building and also model loading on a scheduled basis.

The most recent model that was generated must be loaded first, the other jobs are run as needed and produce ML.NET models that are stored in memory.

To fix two issues:

1. Refactor IHostStartupFilter to be generic and introduce IOptionsValidationFilter
2. Refactor HostStartupService to only run once.

Announcement Core 3.0 Preview 8

• Announcing .NET Core 3.0 Preview 8
• ASP.NET Core and Blazor updates in .NET Core 3.0 Preview 8
• Announcing Entity Framework Core 3.0 Preview 8 and Entity Framework 6.3 Preview 8

Release Notes

-DotNetCore

Breaking Changes

• List of Breaking Changes
• AuthorizationFilterContext is no longer available in HandleRequirementAsync()

Features

• IAsyncEnumerable and Controller

Updates to the Codebase

• Microsoft.AspNetCore.Mvc.Versioning.ApiExplorer.4.0.0-preview8.19405.7 Supports AspNetCore 3.0 Preview 8
• Build issue, removed the following:

    <!--https://github.com/aspnet/AspNetCore/issues/11636-->
    <!--https://github.com/aspnet/AspNetCore/issues/11226-->
    <GeneratePackageOnBuild>false</GeneratePackageOnBuild>

Add a lightweight Scheduling mechanize for the execution of the job.

-Tips and Tricks for Linux development with WSL and Visual Studio Code

• Announcing .NET Core 3.0 Preview 9

• ASP.NET Core and Blazor updates in .NET Core 3.0 Preview 9

• Announcing Entity Framework Core 3.0 Preview 9 and Entity Framework 6.3 Preview 9