Multitargeting in DotNetCore and Linux and Mac builds

One of the most important features of DotNetStandard is the ability to run on Linux and Mac, but if you need to use a DotNetStandard compiled library in a project that uses full .NET framework, sometimes you can have little problems. Actually you can reference a dll compiled for DotNetCore from a project that uses full Framework, but in a couple of project we experienced some trouble with some assemblies.

Thanks to multitargeting you can simply instruct DotNet compiler to produce libraries compiled against different versions of frameworks, so you can simply tell the compiler that you want both DotNetStandard 2.0 and full framework 4.6.1, you just need to modify the project file to use TargetFrameworks tag to request compilation for different framework.

netstandard2.0;net461

Et voilà, now if you run dotnet build command you will find in the output folder both the versions of the assembly.

image

Figure 1: Multiple version of the same library, compiled for different versions of the assembly.

Multitargeting allows you to produce libraries compiled for different version of the framework in a single build.

The nice aspect of MultiTargeting is that you can use dotnet pack command to request the creation of Nuget Packages: generated packages contain libraries for every version of the framework you choose.

image

Figure 2: Package published in MyGet contains both version of the framework.

The only problem of this approach is when you try to compile multitargeted project in Linux on in Macintosh, because the compiler is unable to compile for the Full Framework. that can be installed only on Windows machines. To solve this problem you should remember that .csproj files of DotNetCore projects are really similar to standard MsBuild project files so you can use conditional options. This is how I defined multitargeting in a project

image

Figure 3: Conditional multitargeting

The Condition attribute is used to instruct the compiler to consider that XML node only if the condition is true, and with Dollar syntax you can reference environment variables. The above example can be read in this way: if  DOTNETCORE_MULTITARGET environment variable is defined and equal to true, the compiler will generate netstandard2.0 and net461 libraries, otherwise (no variable defined or defined with false value) the compiler will generate only netstandard2.0 libraries.

Using the Condition attribute you can specify different target framework with an Environment Variable

All the people with Windows machines will define this variable to true and all projects that uses this syntax, automatically have the project compiled for both the framework. On the contrary, all the people that uses Linux or Macintosh can work perfectly with only netstandard2.0 version simply avoiding defining this variable.

The risk of this solution is: if you always work in Linux, you can potentially introduce code that compiles for netstandard2.0 and not for net461. Even if this situation cannot happen now, working with Linux or Mac actually does not compile and test the code against the full framework. The solution to this problem is simple, just create a build in VSTS that is executed on a Windows agent and remember to set DOTNETCORE_MULTITARGET to true, to be sure that the build will target all desired framework.

image

Figure 4: Use Build variables to setup environment variables during the build

Thanks to VSTS / TFS build system it is super easy to define the DOTNETCORE_MULTITARGET at build level, and you can decide at each build if the value is true or false (and you are able to trigger a build that publish packages only for netstandard2.0). In this build I usually automatically publish NuGet package in MyGet feed, thanks to GitVersion numbering is automatic.

image

Figure 5: Package published in pre-release.

This will publish a pre-release package at each commit, so I can test immediately. Everything is done automatically and is run in parallel with the build running in Linux, so I’m always 100% sure that the code compile both in Windows an Linux and tests are 100% green in each operating system.

Gian Maria.

Configure a VSTS Linux agent with docker in minutes

It is really simple to create a build agent for VSTS that runs in Linux and is capable of building and packaging your DotNetCore project, I’ve explained everything in a previous post, but I want to remind you that, with docker, the whole process is really simple.

Anyone knows that setting up a build machine often takes time. VSTS makes it super simple to install the Agent , just download a zip, call a script to configure the agent and the game is done. But this is only one side of the story. Once the agent is up, if you fire a build, it will fail if you did not install all the tools to compile your project (.NET Framework) and often you need to install the whole Visual Studio environment because you have specific dependencies. I have also code that needs MongoDB and Sql Server to run tests against those two databases, this will usually require more manual work to setup everything.

In this situation Docker is your lifesaver, because it allowed me to setup a build agent in linux in less than one minute.

Here are the steps: first of all unit tests use an Environment Variable to grab the connection string to Mongodb, MsSql and every external service they need. This is a key part, because each build agent can setup those environment variable to point to the right server. You can think that 99% of the time the connection are something like mongodb://localhost:27017/, because the build agent usually have mongodb installed locally to speedup the tests, but you cannot be sure so it is better to leave to each agent the ability to change those variables.

With this prerequisite, I installed a simple Ubuntu machine and then install Docker . Once Docker is up and running I just fire up three Docker environment, first one is the mongo database

sudo docker run -d -p 27017:27017 --restart unless-stopped --name mongommapv1 mongo

Than, thanks to Microsoft, I can run Sql Server in linux in a container, here is the second Docker container to run MSSQL

sudo docker run -e 'ACCEPT_EULA=Y' -e 'SA_PASSWORD=my_password' -p 1433:1433 --name msssql --restart=unless-stopped -d microsoft/mssql-server-linux

This will start a container with Microsoft Sql Server, listening on standard port 1433 and with sa user and password my_password. Finally I start the docker agent for VSTS

sudo docker run \
  -e VSTS_ACCOUNT=prxm \
  -e VSTS_TOKEN=xxx\
  -e TEST_MONGODB=mongodb://172.17.0.1 \
  -e TEST_MSSQL='Server=172.17.0.1;user id=sa;password=my_password' \
  -e VSTS_AGENT='schismatrix' \
  -e VSTS_POOL=linux \
  --restart unless-stopped \
  --name vsts-agent-prxm \
  -it microsoft/vsts-agent

Thanks to the –e option I can specify any environment variable I want, this allows me to specify TEST_MSSQL and TEST_MONGODB variables for the third docker container, the VSTS Agent. The ip of mongodb and MSSql are on a special interface called docker0, that is a virtual network interfaces shared by docker containers.

image

Figure 1: configuration of docker0 interface on the host machine

Since I’ve configured the container to bridge mongo and SQL port on the same port of the host, I can access MongoDB and MSSQL directly using the docker0 interface ip address of the host. You can use docker inspect to know the exact ip of the docker container on this subnet but you can just use the ip of the host.

image

Figure 2: Connecting to mongodb instance

With just three lines of code my agent is up and running and is capable of executing build that require external databases engine to verify the code.

This is the perfect technique to spinup a new build server in minutes (except the time needed for my network to download Docker images 🙂 ) with few lines of code and on a machine that has no UI (clearly you want to do a minimum linux installation to have only the thing you need).

Gian Maria.

Pause build and clear long build queue

In VSTS / TFS Build system, you can change the status of the build, between three states: Enabled, Paused and Disabled. The Paused state is really special, because all the build trigger are still active and builds are queued, but all these queued build does not starts.

image

Figure 1: Paused build

Paused state should be used with great care, because if you forget a build in this state, you can end up with lots of queued build, as you can see in Figure 2:

image

Figure 2: Really high number of build queued, because the build definition is paused.

What happened in Figure 2 is that some user probably set the build to paused, believing that no build will be queued, after some week he want to re-enabled, but we have 172 build in queue.

Now if you are in a situation like this, you probably need to remove all queued builds before re-enable the build definition. If you set the build to active you have the risk to completely saturate your build queue. To solve this problem, just go to the queued tab of the build page.

image

Figure 3: Tab queued for the build page

From this page you can filter and show only queued build for the definition that was paused, you can then select all queued builds, and then delete all scheduling at once. Thanks to the filtering abilities present in the queued tab, you can quickly identify queued build and do massive operation on them.

Now that we deleted all the 172 queued build, we re-enabled the build without the risk of saturating build queue.

Gian Maria.

Check Angular AoT with a TFS Build

Developing with Angular is a real fun, but usually during development you serve the application without any optimization, mainly because you want to speedup the compilation and serving of your Angular application.

When it is time to release the software, usually you build with –prod switch and usually you also use the –aot switch (it seems to me that it is on by default on –prod in latest version of the ng compiler). The aot switch enable the Ahead of Time compiling, that is used to speedup your deployed application and also it find some more error than a normal compilation.

I do not want to made examples when a simple ng build compile just fine but the production build with aot fails, the important concept is, when you fire your production build to generate install packages, it is really annoying to see your angular build fails because of some errors. Since building with aot is really expensive I’ve decided not to trigger this build for each commit, but only for special situations. First of all, here is the whole build.

image

Figure 1: Build to check aot compilation

As you can see, with TFS Build it is really a breeze to create a build that simply trigger the ng compiler to verify that it correctly compiles with aot switch on. Thanks to the trigger panel of my build, I’ve decided to build automatically only master and release branches.

image

Figure 2: Trigger configuration for my AOT build

This is another great advantage of Git over a traditional source control system, I can configure a single build for every branch on my repository. In this specific situation I require that the build will be triggered for release and master branches.

This is usually a good approach, but I want to move a little bit further so I configure that build as optional for pull request for develop branch. With this technique, whenever a pull request is done on the develop branch, that build is triggered, and people can fix errors before they slips to the main develop branch.

Thanks to VSTS build system, it is just a matter of minutes to setup specific build that checks specific part of your code, like angular production compilation and optimization.

Gian Maria.

Choose agent at build queue time

This is a simple feature that is not known very well and deserve a blog post. Sometimes you want to queue a build to a specific agent in a queue and this can be simply done using agent.name as a demand.

Demands are simple key/value pairs that allows the build engine to choose compatible agents and each agent automatically have a couple of capability to store computer name and agent name (they can be different)

image

Figure 1: Capability of agentes, computer name and agent name are automatically present

This allows you to use one of them if you want to run a build on a specific machine, just queue the build and use the demands tab to add agent.name or agent.computername demands.

image

Figure 2: Choosing specific agent at queue time with agent.name capability

If you specify a name that is not valid you will get a warning, if the queue contains that specific agent the build will be queued and will be executed only by that agent.

This technique is useful especially if you add a new agent and want to trigger some specific build on it to verify if it has all the prerequisite ok.

Gian Maria.