Azure DevOps中Agent Job与Deployment Group Job的差异及创建原因问询

Hey there! Let's dive into the key differences between Agent Jobs and Deployment Group Jobs in Azure DevOps, plus break down when you'd choose one over the other.

These two job types serve distinct purposes in your CI/CD pipelines, so let's break down their key distinctions:

• Primary Purpose

  • Agent Jobs are your workhorses for general CI/CD tasks: building code, running unit tests, packaging artifacts (like Docker images or NuGet packages), or running static code analysis. They're designed for single-machine, workflow-focused tasks.
  • Deployment Group Jobs are purpose-built for deploying to multiple target machines. Think: pushing your web app to a cluster of web servers, updating databases across a farm, or rolling out software to on-premises workstations.

• Execution Environment

  • Agent Jobs run on a single agent (either an Azure-hosted agent from Microsoft's pool, or a self-hosted agent you manage). All tasks in the job execute on that one machine.
  • Deployment Group Jobs execute on multiple machines in a defined deployment group. Each machine in the group has a Deployment Agent installed, which listens for deployment tasks from Azure DevOps. You can run tasks in parallel across all machines, or target specific subsets via tags.

• Target Management

  • Agents are organized into agent pools—groups of agents shared across multiple pipelines or projects. You pick a pool, and Azure DevOps assigns an available agent to run your job.
  • Deployment groups are organized by environment (e.g., Dev, Staging, Production) or machine role (e.g., web servers, database servers). You can tag machines in a group (like prod-web or canary-node) to fine-tune which machines get the deployment.

• Task Execution Model

  • Agent Jobs run tasks in a linear (or parallel) sequence on one agent. If a task fails, the whole job stops unless you configure error handling.
  • Deployment Group Jobs let you control how tasks run across machines: you can run tasks on all machines at once, roll out to machines in batches, or first deploy to a small "canary" subset to validate before full rollout.

Now let's cover when you'd reach for each job type:

When to Create an Agent Job #

• You need to run general CI tasks: Building your code, running tests, packaging artifacts—all these are perfect for Agent Jobs. They're optimized for single-machine workflows that don't require targeting multiple deployment targets.
• You want to use Azure-hosted agents: If you don't want to maintain your own servers, Azure's hosted agents (with pre-installed tools like .NET, Node.js, Docker) integrate seamlessly with Agent Jobs. Just pick the pool that matches your tech stack.
• Your task only needs one machine: Any workflow that doesn't require distributing work across multiple machines—like generating documentation, running code quality scans, or pushing artifacts to a registry—should use an Agent Job.

When to Create a Deployment Group Job #

• You need to deploy to multiple machines: If your app runs on a cluster of web servers, or you need to update software across dozens of on-premises machines, Deployment Group Jobs let you target all those machines in one step.
• You're managing environment-specific deployments: Create separate deployment groups for Dev, Staging, and Production, and you can easily route deployments to the right environment without reconfiguring targets every time.
• You need granular deployment control: Use machine tags to deploy only to web-server machines, or roll out to a canary tag first to catch issues before deploying to your full fleet. This is critical for reducing deployment risk in production.
• You're working with on-premises/hybrid environments: If your target machines are in your local data center (not Azure), Deployment Groups let you connect those machines to your Azure DevOps pipeline without exposing them to the public internet (via the Deployment Agent's outbound connection).

Hope that clarifies the differences and use cases! If you've got a specific pipeline scenario you're trying to build, feel free to share more details.

内容的提问来源于stack exchange，提问作者Bastien Vandamme