| Comments

One of the things that I like about Azure DevOps Pipelines is the ability to make minor changes to your code/branch but not have full CI builds happening.  This is helpful when you are updating docs or README or things like that which don’t materially change the build output.  In Pipelines you have the built-in functionality to put some comments in the commit message that trigger (or don’t trigger rather) the CI build to stop.  The various ones that are supported are identified in ‘Skipping CI for individual commits’ documentation.

Today that functionality isn’t built-in to GitHub Actions, but you can add it as a base part of your workflows with the help of being able to get to the context of the commit before a workflow starts!  Here is an example of my workflow where I look for it:

name: .NET Core Build and Deploy

on:
  push:
    branches:
      - master

jobs:
  build:
    if: github.event_name == 'push' && contains(toJson(github.event.commits), '***NO_CI***') == false && contains(toJson(github.event.commits), '[ci skip]') == false && contains(toJson(github.event.commits), '[skip ci]') == false
    name: Build Package 
    runs-on: ubuntu-latest

You can see at Line 10 that I’m looking at the commit message text for: ***NO_CI***, [ci skip], or [skip ci].  If any of these are present then the job there does not run.  It’s as simple as that!  Here is an example of my last commit where I just was updating the repo to include the build badge:

Screenshot of a commit message on GitHub

And you can see in the workflows that it was not run:

Screenshot of workflow status on GitHub

A helpful little tip to add to your workflows to give you that flexibility!  Hope this helps!

| Comments

I’ve continuing been doing research on GitHub Actions for .NET developers and came across a comment that someone said (paraphrasing): I wish I could use it for .NET Framework apps but it is just .NET Core.

NOT TRUE! And I want to help fix that perception.  There are some bumps in the road, but allow me to explain some simple (yes I realize they are simple) steps to get it working.

NOTE: I’ve been on this research because I’m looking to better get ‘publish’ experiences in Visual Studio for your apps, but I want to help you get into best practices for CI/CD and DevOps practices.  Basically I’m on a mission for right-click, publish to CI to improve for you :-)

So in this post I’ll walk through an ASP.NET Framework (MVC) app and have it build/publish artifacts using GitHub Actions.  Let’s get started…

The simple app

I am starting from File…New Project and selecting the ASP.NET Web Application (.NET Framework):

Screenshot of template selection

So it’s basic vanilla and I’m not changing anything.  The content of the app is not important for this post, just that we have a full .NET Framework (I chose v4.8) app to use.  From here in Visual Studio you can build the app, run, debug, etc.  Everything you need here is in Visual Studio of course.  If you wanted to use a terminal to build this app, you’d be likely (recommended) using MSBuild to build this and not the dotnet CLI.  The command might look something like this:

code

I’m specifying to build the solution and use a release profile.  We’ll come back to this, now let’s move on.

Publish profile

Now for our example, I want to publish this app using some pre-compiled options.  In the end of the publish task I’ll have a folder that I’d be able to deploy to a web server.  To make this simple, I’m using the Publish capabilities in Visual Studio to create a publish profile.  You get there from right-click Publish (don’t worry, we’re not publishing to production but just creating a folder profile).

Publish profile screenshot

The end result is that it will create a pubxml file in the Properties folder in your solution

Publish profile in solution explorer

So we have our app and our publish (to a folder) profile.  Moving on to the next step!

Publish to the repo and create initial GitHub Actions workflow

From Visual Studio we can add this to GitHub directly.  In the lower right of visual Studio you’ll see the ability to ‘Add to Source Control’ and select Git:

Add to source control tray button

which will bring up the UI to create/push a new repository to GitHub directly from Visual Studio:

Publish to GitHub from VS

Now we have our project in GitHub and we can go to our repository and create the initial workflow.

NOTE: This is the area if you have comments about please do so below.  In the workflow (pun intended) right now you leave Visual Studio and go to GitHub to create a new workflow file then have to pull/sync, etc.  You don’t *have* to do this but usually this is the typical workflow to find templates of workflow files for your app.  Got feedback on what Visual Studio might do here, share below!

Now that you have the publish profile created and your solution in GitHub you’ll need to manually add the pubxml file to the source control (as by default it is a part of the .gitignore file).  So right click that file in solution explorer and add to your source control.  Now on your repository in GitHub go to the Actions tab and setup a new workflow:

Setting up new workflow

The reason for this (in choosing new) is that you won’t see a template that is detected for .NET Framework.  And due to whatever reason GitHub thinks this is a JavaScript repository.  Anyhow, we’re effectively starting with blank.  Create the workflow and you’ll get a very blank default:

name: CI

on: [push]

jobs:
  build:

    runs-on: ubuntu-latest

    steps:
    - uses: actions/checkout@v1
    - name: Run a one-line script
      run: echo Hello, world!
    - name: Run a multi-line script
      run: |
        echo Add other actions to build,
        echo test, and deploy your project.

And it will not be helpful, so we’ll be wiping it out.  I’ve named my workflow build.yml as I’m only focusing on build right now. 

Defining the .NET Framework build steps

For this post I’m going to put all the steps here rather than build-up and explain each one so you can see the entirety.  Here’s the final script for me:

name: Build Web App

on: [push]

jobs:
  build:

    runs-on: windows-latest

    steps:
    - uses: actions/checkout@v1
      name: Checkout Code
    
    - name: Setup MSBuild Path
      uses: warrenbuckley/Setup-MSBuild@v1
      
    - name: Setup NuGet
      uses: NuGet/[email protected]
    
    - name: Restore NuGet Packages
      run: nuget restore SimpleFrameworkApp.sln

    - name: Build and Publish Web App
      run: msbuild SimpleFrameworkApp.sln /p:Configuration=Release /p:DeployOnBuild=true /p:PublishProfile=FolderProfile

    - name: Upload Artifact
      uses: actions/[email protected]
      with:
        name: published_webapp
        path: bin\Release\Publish

Let’s start explaining them.

Ensuring the right runner

In a previous post I described what a ‘runner’ is: What is a GitHub Action Runner?  In that post I pointed to the documentation of runners including what is installed on them.  Now for .NET Framework apps we need to use Windows because .NET Framework only works on Windows :-).  Our action needs to specify using Windows and we are using the windows-latest runner image as we are on Line 8.  I won’t spend time talking about self-hosted runners here, but regardless even your self-hosted runner needs to support .NET Framework.  As a part of the windows-latest runner image, you can see what is already provided on the image.  Currently windows-latest is defined as Windows Server 2019 and the documentation shows what is provided on the hardware resource.  This includes already having a version of Visual Studio 2019 installed…which means MSBuild is already there!

Setting up MSBuild

Even though Visual Studio is on the runner, MSBuild is not presently in the default PATH environment (as of the date of this writing)…so you have options.  The documentation provides the path to where Visual Studio is installed and you can determine the right location to MSBuild from there and specify the path fully.  However, I think there should be easier ways to do this and the community agrees!  In the marketplace there is an Action you can use to setup the PATH to have the MSBuild toolset in your path and you can see this being used on Line 14/15.  The action here basically does a ‘vswhere’ and sets up the ability to later just call MSBuild directly.  This only does MSBuild and not other VS tools that are added to PATH as a part of the ‘Visual Studio Command Prompt’ that most people use.  But using this one we have here, we can now build our Framework app with less path ugliness.

Building and publishing the app

With our MSBuild setup in place, we can start building.  The first thing we need to do is restore any NuGet packages.  In Line 20,21 is where we use the NuGet CLI to restore the solution’s packages that are needed.

NOTE: For some reason using msbuild –t:Restore was not working at the time of this writing that I expected to work…

Once we have the packages restored, we can proceed to build.  In Line 24 is our full command to build the solution.  We are specifying some parameters:

  • Configuration – simple, we are building release bits
  • DeployOnBuild – this helps us trigger the publish step
  • PublishProfile – this uses the publish profile we specify to execute that step and all the other options we have set in that configuration.  We just have to specify the name, not the path

After the completion of this step (we didn’t set any different output folders) we will have a bunch of files in the default publish folder (which would be bin\<config>\Publish).

Publish the artifacts

Once we have the final published bits, we can upload them as the artifact for this build pipeline.  As we see starting at Line 26 we are using another action to upload our content (binaries, files) to this completed workflow as an artifact named ‘published_webapp’ and this will be associated with this run and zipped up all these assets you can download or later use these artifacts to publish to your servers, cloud infrastructure, etc.

Summary

So if you thought you couldn’t use GitHub Actions for your .NET Framework now you know you can with some extra steps that may not have been obvious…because they aren’t.  In the end you have a final build:

Picture of a final build log

What I’ve shared here I put in a sample repro: timheuer/SimpleFrameworkApp where you can see the workflow (in .github/workflows/build.yml) and the logs.  I hope this helps, please share your experiences you’d like to see in Visual Studio to help you better for GitHub Actions.

| Comments

So what exactly is a runner and how do I know what’s in it?  When you use GitHub Actions and specify:

jobs:
  build:
    name: Build
    runs-on: ubuntu-latest

What exactly does that mean ‘ubuntu-latest’?  Well a runner is defined as ‘a virtual machine hosted by GitHub with the GitHub Actions runner application installed.’  Clear? LOL, basically it is a machine that has a target operating system (OS) as well as a set of software and/or tools you may desire for completing your job.   GitHub provides a set of these pre-configured runners that you are using when you use the runs-on label and use any one of the combination of: windows-latest, ubuntu-latest (or ubuntu-18.04 or ubuntu-16.04), macosx-latest.  As of this writing the matrix is documented here with also the specs of the virtual environment: Supported runners and hardware resources.

What is on a GitHub-hosted runner?

I personally think it is good practice to never assume the tool you want is on the environment you didn’t create and you should always acquire the SDK, tools, etc. you need.  That’s just me and possibly being overly cautious especially when a definition of a hosted runner provides the tools you need.  But it makes your workflow very explicit, perhaps portable to other runners, etc.  Again, I just think it is good practice. 

Runner log

But you may want to know what exactly you can use on a GitHub-hosted runner when you specify it.  Luckily GitHub publishes this in the documentation Software installed on GitHub-hosted runners.  For example as a .NET developer you might be interested to know that the windows-latest runner has:

  • Chocolatey
  • Powershell Core
  • Visual Studio 2019 Enterprise (as of this writing 16.4)
  • WinAppDriver
  • .NET Core SDK 3.1.100 (and others)

This would be helpful to know that you could use choco install commands to get a new tool for your desired workflow you are trying to accomplish.  What if you don’t see a tool/SDK that you think should be a part of the base image?  You can request to add/update a tool on a virtual environment on their repo!  Better yet, submit a repo if you can.

How much will it cost me to use GitHub-hosted runners?

Well, if you are a public repository, it’s free.  If you are not a public repository your account gets a certain number of minutes per month for free before billing as well.  It’s pretty generous and you can read all the details here: About billing for GitHub Actions.  In your account settings under the Billing section you can see your usage.  They don’t even bother to show your usage for public repositories because it’s free.  I have one private repo that I’ve used 7 minutes on this month.  My bill is $0 so far.  The cool thing is you can setup spending limits there as well.

Can I run my own runner?

Yes! Similar to Azure Pipelines you can create and host your own self-hosted runner.  The GitHub team did an amazing job with the steps here and it seriously couldn’t be simpler.  Details about self-hosted runners (either on your local machine, your own cloud environment, etc.) can be found in About self-hosted runners documentation.  Keep in mind that now the billing is on you and you should understand the security here as well because PRs and such may end up using these agents and the documentation talks all about this.  But if you are needing to do this, the steps are dead simple and the page in your repo pretty much makes it fool proof for most cases:

Screenshot of self-hosted runner config

It’s good to know what is on the environment you are using for your CI/CD and also cool to know you can bring your own and still use the same workflow.  I’ve experimented with both and frankly like the GitHub-hosted model the best for my projects.  They don’t have unique requirements and since they are all public repositories, no cost to me.  Best of all that I don’t have to now manage an environment!

| Comments

Reflecting back on this blog I realized it’s been 16 years of life here.  It hasn’t always been consistent content focus on tech over the early years versus a more random outlet of my thoughts (and apparent lack of concern over punctuation and capitalization).  Some months had more volume and as my career (and perhaps passions) changed some had lower volume.

Month comparison of blog post quantity

I’ve enjoyed getting back in to posting more recently and finding more time (and again, perhaps the passion) to do so.  I think also with newer various outlets of social media, my personal passions are posted elsewhere now like Instagram (if you want to follow my escapades on the bike mostly).  Recently this summer in 2019 I switched job roles at Microsoft back into program management with the .NET team.  I’m focusing on a few different things but having spent so much time in UI frameworks on the client side for so long, I missed some waves of changes in ASP.NET and needed to re-learn.  I spent the first month of my new role doing this and exploring the end-to-end experiences.  Instead of building a To-do app, I wanted to have some real scenario for me to work with so I set off to migrate my blog…it was time anyway as just that week I had received warnings on my server about some errors.  I could avoid this no longer.  It wasn’t an easy path, but here was my journey.

Existing website frameworks

My blog started in 20-Aug-2003 and was built using Community Server (from Telligent) initially.  That quickly forked into a product called .TEXT from Scott Watermasysk and I moved to using that as it was solely content management and not forums or other things I didn’t need.  I stayed on .TEXT for as long as it lived until again another fork happened.  This time the .NET ecosystem around Open Source was improving and this fork was one of those projects.  Phil Haack, along with others, created SubText which was initially a pretty direct fork, but quickly evolved.  I wasn’t much interested in the code at this point, but wanted to follow ‘current’ frameworks so I moved to SubText.  All along this path it was easy because these migrations were similar, using the same frameworks and similar (if not same) data structures.  The SubText site was an ASP.NET 2.0 site using SQL Server as the data store.  Over time this moved to .NET 3.5 but not much more after that (for me at least). 

Over time I made a few adjustments to the SubText environment for me, never really concerning myself about the source, but just patching crap in random binaries that I’d inject into the web.config.  My last one was in 2013 to support Twitter cards and it was a painful reminder at the time that this site was fragile.   By this time as well the SubText project itself was fragile and not really being maintained as ASP.NET had moved to newer things like MVC and such.  The writing was on the wall for me but I ignored it.

Hosting environment

In addition to the platform/framework used, I was using an interesting hosting setup.  Well, not abnormal really considering at the time in 2003 there was no ‘cloud’ as we know it today.  I had a dedicated box (1U server) hosted at my own data center (I was managing, among other things, a data center rack at the time).  This was running Windows Server 2000 and whatever goop that came with.  Additionally this was SQL Server Express [insert some old version here].  I had moved on to another job and after a period of time I needed to move that server.  I was using the server for more than just my site, running about 10 other WordPress blogs for my community, my wife’s business, and various other things.  WordPress sites were constantly being attacked/hacked due to vulnerabilities in WordPress and leading to my server being filled with massive video porn files and me not knowing until my site was down then I had to login remotely and clean crap up.  Loads of fun.  I eventually moved that server to a co-located environment at GoDaddy still maintaining a dedicated 1U server for me.  It was nice having direct access to the server to do whatever I wanted, but I was quickly not needing that type of hands-on configuration anymore…but still dealing with the management.

During each of these moves I was just moving folders around.  I had no builds, no original source code reliably, etc.  “Fixing” things was me writing new code and finding interesting ways to redirect some SubText functionality as I didn’t have the source nor was interested in digging up tooling to get the source to work.  I never upgraded from Windows 2000 server and was well beyond support for things I was doing.  When I wanted to upgrade the OS at GoDaddy, I was faced with “Sure we’ll set up a new server for you and you migrate your apps” approach.  So again I was going to have to re-configure everything.  Another nightmare waiting and I just put it off.

To the Cloud!

My first step was moving data to the cloud.  I wrote about this when I did this task back in 2012.  I quickly learned that if my site wasn’t on Azure as well and with the traffic I was still getting, the egress costs were not going to be attractive for me.  A few years ago I went about moving just my blog app to Azure App Service as well.  Not having anything to build, this was going to be a fun ‘deployment’ where I needed to copy a lot of things manually to my App Service environment.  I felt dirty just FTP-ing in to the environment and continually trying things until it worked.  But it eventually did.  I had my .NET Framework SubText app running on App Service and using Azure SQL.  The cool thing about Azure SQL is the monitoring and diagnostics it provides.  I immediately was met with a few recommendations and configuration changes I should make and/or it automatically made on my behalf.  That was awesome.  I did have one stored procedure from SubText that was causing all kinds of performance havoc and contributing to me hitting capacity with my chosen SQL plan requiring me to bump up to the next plan and more costs.  Neither of which I wanted to do.  And due to what I mentioned prior about not having SubText buildable I couldn’t reliably make a change to the stored proc without really changing the code that called it.  Just another dent in the plan.  I needed a real migration plan.

Migrate to ASP.NET Core

I mentioned that in my new role I needed to spend time re-learning ASP.NET and this was a perfect opportunity.  I decided to dedicate the time and ‘migrate’ to ASP.NET Core.  Why the air quotes?  Because realistically I couldn’t migrate anything but data.  I did not have reliably building source for SubText and despite that it was WebForms and I didn’t know what I might be getting myself in to.  I needed a new plan, which meant a new framework and I went looking.  Immediately I was met with recommendations that I should go static sites, that Jekyll and GitHub pages are the new hotness and why would I want anything else.  I don’t know, for me, I still wanted some flexibility in the way I worked and I wasn’t seeing that I’d be able to get what I want out of a static site approach.  I wanted to move to ASP.NET Core solutions and found a few frameworks that looked attractive.  Most were in varying states and others felt just too verbose for my needs.  I landed on a recommendation to look at miniblogcore.  This was the smallest, simplest, most understandable solution to my needs that I found.  No frills, just render posts with some dynamicism.

I did not even attempt to migrate any of my existing ASP.NET WebForms code or styling as modern platforms were using Bootstrap and other things to do the site, so that was where I needed to start.  I spent a good amount of time working on the simple styling structure for a few things and learning MVC in the process to componentize some of the areas.  I added a few pieces of customization on the miniblog source, adding search routing (using Google site search), a timeline/calendar view thanks to Telerik controls, category browsing (although mine is horrendous due to waaay too many categories used over the years), using Disquss for commenting, adding SyntaxHighlighter for code formatting support, an image provider for my embedded images during authoring, and a few other random things.  I wrote a little MVC controller to do the data migration once from SQL to the XML file-based storage that MiniBlog uses.  That was a lot simpler than I thought it would take to migrate the data to the new structure.  Luckily my old blog had ‘slug’ support and this new one had it as well, so the URI mapping worked fine, but now I had to ensure the old routing would work.  I had to play around with some RegEx skills to accomplish this but in the end I found a pattern that would match and implemented that in my routing, using proper redirect response codes:

// This is for redirecting potential existing URLs from the old Miniblog URL format
// old subtext non-slugged & slugged
// https://timheuer.com/blog/archive/2003/08/19/145.aspx
// https://timheuer.com/blog/archive/2015/04/21/join-windows-engineers-at-free-build-events-around-the-world-xaml.aspx
[Route("/post/{slug}")]
[Route("/blog/archive/{year:regex(\\d{{4}})}/{month:regex(\\d{{2}})}/{day:regex(\\d{{2}})}/{slug}.aspx")]
[HttpGet]
public IActionResult Redirects(string slug)
{
    // if the post was a non-named one we need to append some text to it otherwise it will think it is a page
    // if (slug doesn't contain letters) { redirect to $"post-{slug}" }
    var newSlug = slug;
    var isMatch = Regex.IsMatch(slug, "^[0-9]*$");
    if (isMatch) newSlug = $"post-{slug}";

    return LocalRedirectPermanent($"/blog/{newSlug}");
}

That ended up being remarkably simpler than I thought it would be as well.  This alone was causing me stress to maintain the URIs that had existed over time and using ASP.NET routing with RegEx I got what I needed quickly.

Moving to Azure App Service once this was all done was simple.  When I first moved ASP.NET Core 3.0 wasn’t yet available so I had to deploy as a self-contained app.  This isn’t difficult though and in some cases may be more explicitly what you want to do.  I wrote how to Deploy .NET apps as self-contained so you can follow the steps.  This basically is a ‘bring the framework with you’ approach when the runtime might not be there.  Azure App Service now has .NET Core 3.1 available though so I no longer have to do that, but good to know I can test future versions of .NET by using this mechanism.

Summary

So what did I learn?  Well, not having source for your apps you care about hurts.  I didn’t even get a chance to actually attempt to truly migrate SubText to ASP.NET Core because I had let my implementation rot for so long.  I have become such a huge believer in DevOps now it’s unreal.  I won’t do a simple project even without it.  The confidence you gain when your projects have continuity through automation is amazing.  My new blog app is fully run on DevOps and deploys using that as well…I just commit changes and they are deployed when I approve them.  I learned that even though this was ‘just a blog’ it was a fairly involved app with separation of user controls and things.  It didn’t need to be so complex, but it was and I’m glad for MiniBlog not being so complex.  The performance of my content site and costs are much more manageable now and my stress is reduced knowing that should anything happen I’m in a better place for restoring a good state.  My biggest TODO task I think is re-thinking the XML-based data store though.  This actually is the one thing causing me some DevOps pain because the ‘data’ is content within the web app and when using slot-staging deployment that doesn’t work well.  Azure has a way to use Azure Storage as a mounted point to serve content from in your App Service though and I’ve started to try that with some mixed results so far.  Using this approach separates my app from my data and allows for more meaningful deployment flows and data backup.  I’ve also explored using an Azure Storage provider for my data layer, but the method for how the initial cache is build in MiniBlog right now makes this not a great story due to startup latency when you have 2,000 posts to retrieve from blob container.  I’m still playing around with ideas here, so if you have some I’d love to hear (dasBlog users would hit similar concerns).

I’m happy with where I landed and hope this keeps me on a path for a while.  I’ve got a simple design, responsive design, easy-to-maintain source code, all the features I want (for now), no broken links (I think), works with my editing flow (Open Live Writer), and less stress worrying about a server.  I’ve already updated to ASP.NET Core 3.1 and it was a simple config change to do that now that my setup is so streamlined. 

What are your migration stories?

| Comments

Continuing on my research and playing around with GitHub Actions, I was looking to migrate my Alexa.NET project off of Pipelines and in to one place for my open source project.  Pipelines still has an advantage for me right now as I prefer the approval flow that I have right now.  In this post I’ll cover how I modified my build definition to now also include producing the NuGet package, signing it with my code signing certificate, and pushing it to multiple repositories.

Quick tip: if you haven’t follow Ed Thomson before he’s doing a series on GitHub Actions for the month of December.  Check out his GitHub Actions Advent Calendar!

Pre-requisites

We need to first make sure we have the tools needed in the build step, so let’s be sure to get the .NET SDK so we can use the dotnet CLI commands.  This is the start of my build-and-deploy.yaml file and each other snippet builds on this.

name: "Build and Deploy"

on:
  push:
    branches:
      - master

jobs:
  build:
    if: github.event_name == 'push' && contains(toJson(github.event.commits), '***NO_CI***') == false && contains(toJson(github.event.commits), '[ci skip]') == false && contains(toJson(github.event.commits), '[skip ci]') == false
    name: Build Package
    runs-on: ubuntu-latest

    steps:
    - uses: actions/checkout@v1
    - name: Setup .NET Core SDK
      uses: actions/setup-dotnet@v1
      with:
        dotnet-version: 3.1.100

Starting at line 16 I write the steps to get the .NET SDK that I want to use, in this case the .NET 3.1 (which is the long-term support version now) SDK.  Now we are all set…the tools I need are on the runner.

Building the Package

The first thing we obviously need to do is ensure we have an actual NuGet package.  I perform this step during my ‘build’ job when I know things have been successfully built and tested.  After getting the SDK we can how issue our pack command.  This assumes we’ve already run dotnet build, which I didn’t show here.

    - name: Pack
      run: dotnet pack TestLib --configuration Release -o finalpackage --no-build

    - name: Publish artifact
      uses: actions/upload-artifact@master
      with:
        name: nupkg
        path: finalpackage

You can see in line 2 where we use the dotnet CLI to pack into a NuGet package.  Note I’m using an output argument there to put the final nupkg file in a specific location.  In line 5 I am setting up the action to upload the artifact so that I can use it later in other steps in the job.  The upload-artifact agent will use the path ‘finalpackage’ and upload it into the location ‘nupkg’ for me.  It will available for me later as you’ll see.

Signing the Package

Now I want to be a good trusted provider of a library package so I’ve chosen to sign my package using a code-signing certificate.  I got mine through DigiCert.  One of the main differences between Actions and Pipelines is that Actions only has secure storage for ‘secrets’ as strings.  Pipelines has a library where you can also have secure file storage.  To sign a NuGet package, the command requires a path to a certificate file so we have to somehow get the file available for the CLI command.  Based on all the recommendations from people also doing similar activities (needing files in their actions) it seemed to be the approach was to base64-encode the file and put that as a secret…so that’s the approach I took.  I base64-encoded the contents of my PFX and set it as a secret variable named SIGNING_CERT. 

Now the next thing I need to do is not only retrieve that string, but put that into a temporary file.  Searching as best I could on forums I didn’t see an existing script or anything that people used, so I created a new action for myself to use (and you can to) called timheuer/base64-to-file.  This action takes your encoded string, decodes it to a temporary file and sets the path to that temporary file as an output for the action.  Simple enough.  Now with the pieces in place we can set up the steps:

  deploy:
    needs: build
    name: Deploy Packages
    runs-on: windows-latest # using windows agent due to nuget can't sign on linux yet
    steps:
      - name: Download Package artifact
        uses: actions/download-artifact@master
        with:
          name: nupkg
      
      - name: Setup NuGet
        uses: NuGet/[email protected]
        with:
          nuget-api-key: ${{ secrets.NUGET_API_KEY }}
          nuget-version: latest

      - name: Setup .NET Core SDK
        uses: actions/setup-dotnet@v1
        with:
          dotnet-version: 3.1.100

      - name: Get certificate
        id: cert_file
        uses: timheuer/base64-to-file@master
        with:
          fileName: 'certfile.pfx'
          encodedString: ${{ secrets.SIGNING_CERT }}
      
      # Sign the package
      - name: Sign NuGet Package
        run: nuget sign nupkg\*.nupkg -CertificatePath ${{ steps.cert_file.outputs.filePath }} -CertificatePassword ${{ secrets.CERT_PWD }}  -Timestamper http://timestamp.digicert.com –NonInteractive

The above is my ‘deploy’ job that does the tasks.  On line 6 is where we are retrieving the nupkg file from the artifact drop in the previous job.  After that I’m using the new nuget/setup-nuget action to acquire the NuGet CLI tools for subsequent actions.  At present, you cannot use dotnet CLI to sign a NuGet package so we have to use the NuGet tools directly.  We’ll need this later as well so it’s good we have it now.  On line 22 starts the process mentioned above to use my new action to retrieve the encoded string and put it as a temp file.  One line 31 we execute the NuGet sign CLI command to sign the package.  I have a few arguments here but pay attention to the steps.cert_file.outputs.filePath one.  That is the OUTPUT from the base64-to-file action.  The format of steps.{ID}.outputs.{VARIABLE} is what you see here…and you can see in that step I gave it an id of ‘cert_file’ to easily pull out the variable later.

Now, you may have noticed that this agent job runs on windows-latest as the OS and not ubuntu.  This is because presently package signing for NuGet can only be done on Windows machines.  Now that we have a signed package (in the same location, we just signed it and didn’t move it) we can deploy it to package registries.

Publishing the Package to NuGet

Of course for a public library I want this to be available on NuGet so I’m going to publish it there.  NuGet uses an API key authentication scheme which is supported in the dotnet CLI so we can use dotnet CLI push to publish:

      - name: Push to NuGet
        run: dotnet nuget push nupkg\*.nupkg -k ${{ secrets.NUGET_API_KEY }} -s https://nuget.org

Could I have used the NuGet CLI?  Sure, but I already was using this pattern previously so I’m sticking with this from a previous Pipeline definition.  Choice is yours now that we have both CLI tools on the runner machine.  Done, now on to another registry.

Publishing the Package to GitHub Package Registry

Publishing to the new GitHub Packages Registry takes one extra step.  Since this is not the default location for NuGet, we have to instruct NuGet to let it know where to publish this package.  In your repository you will be provided with a URL from the Packages tab of your repo:

Screenshot of GitHub Packages tab

This is the publishing endpoint for the NuGet CLI.  In our Action we will need two steps: set up the source and publish to it:

      - name: Add GPR Source
        run: nuget sources Add -Name "GPR" -Source ${{ secrets. GPR_URI }} -UserName ${{ GPR_USERNAME }} -Password ${{ secrets.GITHUB_TOKEN }}

      - name: Push to GitHub Packages
        run: nuget push nupkg\*.nupkg -Source "GPR"

In line 2 is where we set up the source we are going to later use.  We can give it any name you want here.  I made the other variables Secrets for my config.  This also requires you to use the UserName/Password scheme as GitHub Packages doesn’t support NuGet API keys right now.  Another reason we need to use the NuGet CLI here.  The password you can use is provided as a default token in any GitHub Action called secrets.GITHUB_TOKEN and your repo’s actions have access to it.  In line 5 then we see us using that source and pushing our package to the GitHub Packages Registry.

Summary

So there you have it!  A GitHub Actions flow packages, signs, and publishes to two package repositories.  It would be nice to standardize on one tooling CLI and I know the teams are looking for feedback here, but it is good to know that you have 2 official supported GitHub Actions in setup-dotnet and setup-nuget to use to get the tools you need.  I hope this helps someone!