A set of tricky situations with HTTPS and TFS

HTTPS is more and more common-place, not just for public websites but also for internal websites. This is extremely good for a number of reasons, but from an administration standpoint there are a few bits to keep in mind.

In particular, when it comes to Team Foundation Server this is a list of errors and problems that go away with a common denominator: the right certificate.

The number one offender is of course the out-of-domain machine. If you have a domain-joined machines these problems simply do not happen because the internal certificate is deployed by the domain GPO - hence you don't have to fiddle with it. When your machine is not domain-joined, things can easily go south.

Bear in mind - these are not security tips, this is just a collection of situations which you will face if you deploy HTTPS with TFS.

Non domain-joined machines

If you are running a non domain-joined machine then you need to procure the root certificate for your domain and install it in the Trusted Root Certification Authorities store on your machine. This needs to be done on any machine not part of your domain, otherwise you won't be able to do pretty much anything.

Build agents

Build agents need to be reconfigured. You can't run away from this, if you don't do that they will be working until the authentication token expires, and then you will start seeing this error in the Event Log after they go offline:

Agent connect error: The audience of the token is invalid.. Retrying every 30 seconds until reconnected

You need to de-register (config.cmd remove) and re-register your build agents in any pool. Not too bad, but it needs to be planned for.

The Deploy Test Agent task in Build and Release

If you don't have your certificate installed on both the Agent (if outside the domain) and the target machine (again, if outside the domain) then you will get this cryptic error:

The running command stopped because the preference variable "ErrorActionPreference" or common parameter is set to Stop: Exception calling ".ctor" with "2" argument(s): "One or more errors occurred."

It's a communication issue between the target machine and TFS. Once the certificate is installed it goes away and the task works normally. This GitHub issue also recommends enabling TLS v1.2, which is not a bad idea.

 Git

Git holds a special spot in this collection, because of how it handles SSL. While newer versions of Git for Windows made this really straightforward (hint: they support the Windows Credential Manager), but if you aren't running the latest and greatest then this is what could happen with Git on your local machine, even if it is joined to the domain:

C:\>git clone https://myserver/Collection/_git/Project 
Cloning into 'Project'... 
fatal: unable to access 'https://myserver/Collection/_git/Project/': SSL certificate problem: unable to get local issuer certificate

You can sort this out in many ways, but the best one is Philip Kelley's approach. It just works, even if it is a bit of a walkthrough. This applies not only on the client, but also on the build agent if you are not running a recent version of the agent itself. It can be easily corrected by replacing the ca-bundle.crt file over there, it is not going to be replaced until you update the agent to a newer version.

Also, a false friend:

error: RPC failed; curl 56 OpenSSL SSL_read: SSL_ERROR_SYSCALL, errno 10054
fatal: read error: Invalid argument, 255.05 MiB | 1.35 MiB/s
fatal: early EOF
fatal: index-pack failed

It can be all sorts of things, especially as the error points at OpenSSL - but check your connection's stability first before messing up with Git's postBuffer and compression 😃 if the git clone operation starts the problem is not the SSL authentication.

Easily handle internal settings while orchestrating components' deployments and parameters

After ten years of attending, then speaking at conferences it always strikes me what demos often miss are real world details that really make the difference.

Like...deploying an application with a pipeline. Everybody talks about it, right? And everybody (including myself!) have some demo-ready stuff to show around in case it might be required.

I am working on a sample application right now, and I realised how blind I was - even if I am deploying stuff to different slots and environments and whatnot, I am still treating everything as a single monolith. Not really what you want these days, right?

Well' let's sort it out. Say that you have an API component and a Frontend component, the best thing to do is to decouple the two of them so they can be independently deployed *and* mix-matched depending on the requirement.

It is .NET Core in my case, so in my Frontend component's appsettings.json I created this section:

Of course I modified the application so I could add the configuration in my ConfigureServices method and consume it in my Controller. The variable part in this case is the Slot property.

Now comes the fun side of the story - of course I have a pipeline in place. How do I handle these settings?

The best approach here, given the relative complexity of this exercise, is to scope the relevant value by environment. The Dev environment will always point at the Dev environment, Staging to Staging, and the last two environments are effectively production so I do not need to worry about adding a slot. It's not like I have cross-environment settings here.

The reason why the variables are named that way is because I am using the JSON variable substitution option in the Azure App Service Deploy task, and as my property is not on the first level then it needs to be explicitly written that way.

Doing it ensures that each environment has its own setting, and it also makes sure you remain sane while handling internal app settings across your applications and environments 😉 it is really easy to do as well, so there is really no reason to skimp on it.

Quickly deploy a baseline SQL database with VSTS

"Sometimes we go full steam ahead with a complex solution for a very simple problem..."

That was the answer I gave to a friend of mine who asked me how to feed some baseline database for testing purposes with VSTS in Azure.

The obvious one would be to have your versioned SQL scripts in a dedicated repository which you can use to rebuild the whole thing from code (which is by all accounts the most correct solution to this problem). But in this case there are other avenues.

Databases have been treated like second class citizens for years - by tools and practices. For example, why not using BACPAC files for this exercise? At the end of the day, a BACPAC file contains the packaged version of a database at a certain point in time, including its data.

So if you have your BACPAC somewhere, get to an Azure storage account and run this SQLPackage command inside a VSTS PowerShell Script task (of course you need to replace the variables and provide the actual path):

& 'C:\Program Files (x86)\Microsoft Visual Studio\2017\Enterprise\Common7\IDE\Extensions\Microsoft\SQLDB\DAC\130\sqlpackage.exe' /Action:Import  /TargetServerName:$(DBUrl) /TargetDatabaseName:$(DBName) /TargetUser:$(DBAdmin) /TargetPassword:$(DBPassword) /SourceFile:"<your location>/sample.bacpac"

Don't get me wrong, I love seeing a database fully integrated with the pipeline and that's how it should be. But in this specific case, I feel the tradeoff is worth it.

Also - this is a baseline database, nobody prevents us from running delta scripts against it depending on needs. But given it was for testing purposes, I highly doubt there is going to be much development on it in the future!

How to run UI tests in a Deployment Group with TFS and VSTS

Especially if you are testing client applications, you might want to run UI tests on a Deployment Group instead of a Build Agent. While technology is the same, there are a couple of things to keep in mind.

In order to enable a machine to run UI tests you need to make sure your InteractiveSession capability is set to true.

In order to do so, you need to re-configure or manually change the script used to add a machine to the Deployment Group. Given a standard script the first step is removing the --runasservice switch from it.

Once you run the configuration script the process will guide you to configure the agent for interactive interaction. You will set it to auto-start so you will get an unattended experience when rebooting the machine, but you will be able to run interactive sessions on it.

Eventually, I always recommend to use the VSTest Platform Installer task to make sure you have a consistent environment to run your tests from:

and to refer to the tools installed by that in the Visual Studio Test task: