Azure DevOps is a Microsoft service developed prior to their acquisition of GitHub. There’s a lot of overlap between the two services, and – even though it seems inevitable that GitHub is the future, there are a lot of organizations still using Azure DevOps.
Azure Pipelines is a part of Azure DevOps where you can build and test code projects. (When programmers use the word “build”, they mean to convert code written by a human programmer into something that can run on a computer.) “CI/CD” has recently become a buzzword in the software development community. It stands for Continuous Integration/Continuous Deployment. Azure Pipelines is the part of Azure DevOps that provides CI/CD services.
The “building software” part of CI/CD requires build agents. A build agent is a computer used to build software, but a it’s a different computer than one used by a human programmer to write code. Azure Pipelines maintains a pool of VMs that can serve as your build agent. When a build process is triggered, one of these is temporarily allocated to your process. It does your build and then reverts back to the pool. (CI/CD is all about automating processes. All of these happens “auto-magically” as directed by a script.)
The VMs provided by Microsoft will meet most needs, but not all. I’m working on project for a client where the amount of RAM provided in the Microsoft-hosted agents wasn’t enough and the build process was failing because it ran out of memory.
When the Microsoft-hosted agents won’t meet your needs, you can plug you own “Self-hosted agents” into the process. This can either be physical or virtual machines that you set up, configure and maintain, or – you can take advantage of Azure virtual machine scale set agents. This lets you specify a “size” of Azure VM (allowing you to specify the processor, amount of RAM, amount of storage, etc.) and tells Azure DevOps to use that for your build agent.
This option is more expensive and requires more administration on your part, but if you have needs beyond what the Microsoft-hosted agents can handle – it’s nice to have this option. The rest of this post will recap some of the technical issues I ran into setting this up recently.
This is a good starting point for background on Azure Pipelines Agents. If you follow the link in the Microsoft-hosted agents section, you can find out about the specific hardware allocated to these VMs. As of this writing, the amount of RAM is 7GB. This client has a particularly large Angular project that required more RAM than that.
I didn’t want to actually “self-host” in a physical sense, so I went with the Azure virtual machine scale set option. Here is the page with the specific instructions. There’s a lot of information here, but basically it consisted of three steps:
- In the Azure portal, create the scale set.
- In the Azure DevOps portal, set up an agent pool pointing to what you created in #1.
- Change your pipelines to use the new agent pool.
Create the Scale Set in Azure
The instructions walk you through using the Azure Cloud Shell to create the scale set. The
az vmss create statement is the one to focus on. The vm-sku setting is probably the most important one. This is what specifies the “size” of the VM. Start here to determine which size will work best for your needs. (Don’t agonize over this too long at the beginning. It’s pretty simple to change this setting on your scale set.) Make sure to pay attention to the usage price of your chosen size. (Here’s the pricing table for Linux VMs.)
The instance-count setting only specifies the number of instances initially created. Once you “hand control of the set over to Azure DevOps” (in step 2). It will override this instance count.
Once you’ve successfully run the az vmss create statement. You’re done with step 1. Azure DevOps will handle everything else (including installing required software on the VMs).
Create the Agent Pool in Azure DevOps
The instructions in the article above are pretty good. Here are some notes of things I ran into:
- You will need to be an Agent Pool Administrator to create the new Agent Pool. If you are missing this, you will get to the end of the form and get an error when you submit saying you need “Manage permission”. If you get this, have the owner of the organization use the Security button on the Agent Pools page (top right – next to blue Add Pool button) to grant this permission to you.
- The Maximum number of virtual machines in the scale set setting overrides the instance-count setting from step 1.
Change your Pipelines to use the new Agent Pool
Here is the relevant part of the Azure Pipelines YAML schema. Basically, if you are using a Microsoft-hosted agent pool, you only specify the vmImage value.
If you want to use a self-hosted agent pool, you need to specify the name. This is the name you provided in step 2 and the name that shows up in the list of Agent Pools in Azure DevOps.
In the case shown above, your YAML would contain this:
pool: Self Hosted Scale Set pool
The 3 main settings on the Azure DevOps Agent Pool will probably involve some trial and error depending on your specific needs.
- Maximum number of virtual machines in the scale set
- Number of agents to keep on standby
- Delay in minutes before deleting the excess idle agents
If you set the Number of agents to keep on standby to 0, you won’t be paying for instances to sit idle, but your Pipeline runs will sit in the Queued state for a little while to give Azure time to “spin up” an image for you. (In my case, this was about 5 minutes.) You can monitor the number (and status) of instances in the Azure portal by finding the Virtual machine scale set and clicking on Instances on the left-side menu. (Hint – if you can’t find it, look at the Resource Group you specified in the
az vmss create statement.)