So, last week was mostly an introduction to .NET Core and AWS Lambda. Not a lot of meat.

This week I’ll be writing about the build and deployment pipeline that we set up, and how it ties in with our normal process.

Hopefully it will be a bit more nutritious.

The Story So Far

Over the last few years we’ve put together a fairly generic build and deployment process using Powershell that focuses on ensuring the process can be run painlessly on a fresh checkout of the repository. The goal is to have everything either embedded or bootstrapped, and to have any and all logic encapsulated within the scripts.

What this means is that we have a relatively simple setup in TeamCity, which is basically just parameter management and a small bit of logic for kicking off the scripts that live inside the repositories.

The same approach is extended to deployment as well; we have Powershell (or an equivalent scripting language) that does the actual deployment, and then only use Octopus Deploy for orchestration and variable management.

Because we need to build, test and deploy a variety of things (like C# desktop applications, Javscript websites, the old Javascript ELB Logs Processor, CloudFormation environments), the common scripts feature pluggable processes in the form of Powershell script blocks, allowing the user to choose what sort of build is needed (simple Nuget package over some files? A full MSBuild execution that generates an .exe in a Nuget package?), how and when it is tested (Nunit over .dlls containing the phrase “Tests” before deployment? Selenium using Webdriver IO through npm after deployment?) and how it is deployed (pretty much just Octopus Deploy).

The API for the pluggable processes (which I’m going to call engines from this point forward) is relatively lax, in that the Powershell just calls the supplied script block with known parameters, and fails miserably if they don’t exit.

In retrospect, we probably should have just used something already available, like psake, but it just didn’t seem necessary at the time. We’re probably going to get hurt if we try to do anything that only works on Linux systems (no Powershell there), but it hasn’t happened yet, so I’m sure it will never happen.

To summarise, we have a framework that allows for configurable approaches to build, test and deployment, and a .NET Core AWS Lambda function should be able to slot right in.

Vrmm Vrmm Goes The Engine

In order to reliably get the lambda function code from the git repository in which it lives, to the actual AWS Lambda function where it will be executed, we needed to do three things:

  1. Creation of an artefact containing a versioned and validated copy of the code to be executed
  2. The validation logic itself (i.e. testing)
  3. Deploying said artefact to the AWS Lambda function

Coincidentally, these three things line up nicely with the pluggable components in our generic build and deployment process, so all we have to do is provide an engine for each, then run the overall process, and everything should be sweet.

Bob The Builder

The first thing to deal with is the actual creation of something that can be deployed, which for us means a Nuget package.

Referring back to the Javascript version of the ELB Logs Processor, there really wasn’t a lot to the creation of the artefact. Use npm to make sure the appropriate dependencies are present, a quick consolidation step with webpack, then wrap everything up in a Nuget package.

For the C# version, its a similar story, except instead of using node/npm, we can use the dotnet executable available in the SDK, followed by the Nuget command line tool to wrap it all up neatly.

Basically, it looks like this:

dotnet restore {solution-path}
dotnet build –c Release {solution-path}
dotnet publish –o {output-path} –c Release {solution-path}
nuget pack {nuspec-path} {output-directory} {version}

The actual script block for the engine is a little bit more complicated than that (TeamCity integration, bootstrapping the .NET Core SDK, error handling/checking, etc), but its still relatively straightforward.

As I mentioned last week, the only thing to watch out for here is to make sure that you’re targeting the appropriate runtime with your solution configuration, and not to try and build a standalone executable. The code will actually be executed on a Linux machine behind the scenes, and AWS Lambda will make sure the .NET Core runtime is available, so all you have to worry about is your code. I’m pretty sure if you try and supply an OS independent executable it will fail miserably.

Emotional Validation

The second thing to deal with is the validation of the artefact.

Similar to the build process, if we refer back to the Javascript ELB Logs Processor, we had no testing at all. Literally nothing. Our validation was “deploy it and see if the logs stop”.

Don’t get me wrong, there are tonnes of ways to write tests and automate them when using node/npm, we just never went to the effort of actually doing it.

I’m much more familiar with C# (even taking into account the differences of .NET Core), and we don’t write code without tests, so this was an ideal opportunity to put some stuff together for validation purposes.

I’m going to dedicate an entire post to just how we went about testing the rewrite (some interesting stuff there about managing AWS credentials in a secure way), but at a high level we followed our standard pattern of having dedicated test projects that sit in parallel to the code (i.e. Lambda.Tests.Unit, Lambda.Tests.Integration) with automatic execution through the generic build process via an engine.

This particular engine is applied predeployment (i.e. before the Nuget package is pushed to Octopus and a project executed), and is pretty straight forward:

dotnet test {project-path} –c Release –no-build –logger {output-file-path}

The only caveat here was the general immaturity of .NET Core, as I had some problems using NUnit (which is the testing framework we typically use for C#), but XUnit worked perfectly fine. They aren’t that dissimilar, so it wasn’t a huge problem.

Tentacle Joke

The last thing to deal with is the actual deployment of the artefact to AWS.

Assuming step 1 and 2 have gone well (i.e. we have a versioned Nuget package that has been validated), all that’s left is to leverage Octopus Deploy.

For us this means a relatively simple project inside Octopus that just deploys the Nuget package to a worker tentacle, and a Powershell script embedded in the package that does the bulk of the work.

$awsKey = Get-OctopusParameter "AWS.Deployment.Key";
$awsSecret = Get-OctopusParameter "AWS.Deployment.Secret";
$awsRegion = Get-OctopusParameter "AWS.Deployment.Region";

$version = Get-OctopusParameter "Octopus.Release.Number";

$functionName = Get-OctopusParameter "AWS.Lambda.Function.Name";

$aws = Get-AwsCliExecutablePath

$env:AWS_ACCESS_KEY_ID = $awsKey
$env:AWS_SECRET_ACCESS_KEY = $awsSecret
$env:AWS_DEFAULT_REGION = $awsRegion

$functionPath = "$here\function"

Write-Verbose "Compressing lambda code file"
Add-Type -AssemblyName System.IO.Compression.FileSystem
[system.io.compression.zipfile]::CreateFromDirectory($functionPath, "index.zip")

Write-Verbose "Updating Log Processor lambda function [$functionName] to version [$version]"
(& $aws lambda update-function-configuration --function-name $functionName --runtime "dotnetcore1.0" --handler "Lambda::Handler::Handle") | Write-Verbose
(& $aws lambda update-function-code --function-name $functionName --zip-file fileb://index.zip) | Write-Verbose
(& $aws lambda update-function-configuration --function-name $functionName --description $version) | Write-Verbose

In summary, the script takes the contents of the Nuget package that was just deployed, zips up the relevant parts of it (in this case, everything under the “function” directory) and then uses the AWS CLI and some Octopus parameters to push the zip file to a known Lambda function.

Really the only interesting thing here is that we take the version from the Nuget package and apply it directly to the function in AWS for backwards traceability.

To Be Continued

The steps and processes that I’ve outlined above only describe the key parts of the build and deployment process. A lot of the detail is missing, and that is on purpose, because the entire thing is quite complicated and describing it comprehensively would make this blog post into more of a novel.

Its all ready pretty long.

The key things to take away are the .NET Core build and publish process (using the dotnet cmdline application), the .NET Core testing process (again, through the cmdline application) and the simplest way that I’ve found to actually push code to a Lambda function in an automated fashion.

Next week I’ll extrapolate on the testing of the lambda function, specifically around managing and using AWS credentials to provide an end to end test of function logic.