{"id":175,"date":"2020-02-24T06:33:00","date_gmt":"2020-02-24T06:33:00","guid":{"rendered":"https:\/\/blogs.zeiss.com\/intern\/tech\/?p=175"},"modified":"2020-08-31T11:21:13","modified_gmt":"2020-08-31T11:21:13","slug":"ms-azure-devops-on-demand-agents","status":"publish","type":"post","link":"https:\/\/blogs.zeiss.com\/tech\/ms-azure-devops-on-demand-agents\/","title":{"rendered":"MS Azure DevOps: On-Demand Agents"},"content":{"rendered":"\n

This post is the first in a two-part series.<\/em>

In today\u2019s DevOps world, Continuous Integration and Continuous Delivery allow software engineers to deliver values to customers rapidly. We as Digital Innovation Partners<\/a> at Zeiss use Azure DevOps<\/a> extensively to ensure fast software delivery.<\/p>\n\n\n\n

Pipelines<\/a> is a part of Azure DevOps which helps us to build, configure, and release our code, but for that, all you need is the Agent. Build or release agent is used to run your jobs and sends the result back to Azure DevOps. There are various offerings from Azure DevOps to run your build and release jobs.<\/p>\n\n\n\n

Hosted Agent<\/h2>\n\n\n\n

Azure DevOps provides free hosted agents that have a predefined set of tools installed and configured for building and releasing your apps. Microsoft-hosted agent pool offers 7 different virtual machine images<\/a>.<\/p>\n\n\n\n

Hosted agents are good to start but as soon as your team grows, then you have to face long queue time for the build and release. If the free tier doesn\u2019t meet your needs, you can still buy an additional parallel job with unlimited minutes which costs \u20ac33.74<\/strong> per month.<\/p>\n\n\n\n

Self Hosted Agent <\/h2>\n\n\n\n

There is another option where you set up and manage your own agents. This can be done by simply downloading the agent package, and run it either on your local machine or any other computing platform even on Docker container. It also gives you more freedom to install specific dependencies for your build and release.<\/p>\n\n\n\n

The main disadvantages of the Hosted agent are maintenance and scalability. For instance, if you are running your agent as a docker container or directly on VM you need to make sure that the agent is not filling up disk storage.<\/p>\n\n\n\n

Why not standard Hosted or self-hosted Agent<\/h2>\n\n\n\n

As mentioned before, we own an Azure DevOps account that contains multiple private projects. By default, Azure DevOps offers a hosted agent with one free parallel job.<\/p>\n\n\n\n

We started with provided free tier hosted Agents. As there was only an agent with one parallel job, we suffered a lot from long queue times. We bought a few more parallel jobs but actually, it was not that helpful as our account was hosting around 30 projects.<\/p>\n\n\n\n

As a quick workaround to reduce this long waiting time, most of the projects started setting up their own VMs and running multiple self-hosted agents as Docker containers. This enabled us to have more than one concurrent jobs running. The drawback of this solution was that almost every project was running their own VMs (not very cost-effective as VMs where mostly idling) and increasing maintenance overhead as well we slow scaling when new agents are needed (new VMs had to be set up etc.)<\/p>\n\n\n\n

This leads us to implement something out-of-the-box which could be cost-effective, easily scalable and can be shared among multiple projects. We finally came up with the idea of On-Demand Agent<\/strong> to manage such build\/release agents using some kind of outer orchestration and Kubernetes(to be specific AKS due to Microsoft Stack) was the easy choice.<\/p>\n\n\n\n

The basic idea was to create a pod running as an agent for each build\/release job to ensure clean and reliable agents all the time.<\/p>\n\n\n\n

Note<\/strong>: On-Demand Agents<\/strong> come under the self-hosted agent category but the lifespan of such agent is specific build or release.<\/p>\n\n\n\n

How does it work<\/h2>\n\n\n\n

Before proceeding with more details, it is important to understand the components which we are going to use in our On-Demand Agent Infrastructure on AKS.<\/p>\n\n\n\n