An Overview of Workflow in Chef Automate¶
Chef Automate manages changes to both infrastructure and application code, giving your operations and development teams a common platform for developing, testing, and deploying cookbooks, applications, and more.
Chef Automate accelerates the adoption of continuous delivery and encourages DevOps collaboration. It provides a proven, reproducible workflow for managing changes as they flow through the pipeline from a local workstation, through automated tests, and out into production.
Chef Automate handles many types of software systems. Use it to:
- Upload new and updated cookbooks to the Chef server that manages your infrastructure and applications
- Publish new and updated cookbooks to a Chef Supermarket installation
- Release source code or build artifacts to a repository such as GitHub
- Push build artifacts to production servers in real time
If you are new to Chef Automate, you can see it in action in the self-paced tutorial Get started with Chef Automate on AWS. In the tutorial, you’ll set up your own Chef Automate cluster and a sample application to experiment with.
A pipeline is series of automated and manual quality gates that take software changes from development to delivery. The goal of a pipeline is to move changes from your workstation into production quickly and safely.
Pipelines in Chef Automate have six stages: Verify, Build, Acceptance, Union, Rehearsal, and Delivered. Changes progress from one stage to another by passing a suite of automated tests. For the Verify and Acceptance stages, explicit approval by a designated person is required (in addition to the tests).
Here are the stages of a Chef Automate pipeline.
The tests within each stage are organized into phases. The stages and the phases are fixed for all pipelines in Chef Automate. However, what happens within any given phase is completely up to you—if you can describe the activity in a Chef recipe, then you can make it happen in a phase.
The following illustration shows the phases of each pipeline stage.
Chef Automate relies on git and uses its lightweight feature branches as the mechanism for handling changes before they merge, as well as its ability to perform merges automatically. Each pipeline has a designated target branch into which it will merge approved changes. Chef Automate uses a “gated master” model that manages merges to the target branch. (In preparation for using Chef Automate, it is helpful if team members understand how to use feature branches.)
Chef Automate uses projects to organize work across multiple teams. You can create as many projects as you need. A common approach is to have one project for each major component of your system. Each project has its own git repository. (Chef Automate includes a git server for hosting project repositories. It is also possible to integrate with GitHub and GitHub Enterprise for the git-related aspects of the workflow.)
Organizations allow you to group related projects and provide scope for authorization rules.
Each project has one or more pipelines. The typical setup is for each project to have a single pipeline that targets the master branch.
Having multiple pipelines allows the project to target different branches for different changes. A potential use case is maintaining different versions of a project on different branches, enabling you to target a change (for instance, a security fix) against multiple versions quickly and easily.
Changes and Project Pipelines¶
Let’s walk through what happens as a change makes its way through Chef Automate. We’ll assume you have created a project in Chef Automate and want to make a change.
You start with a local checkout of the project’s git repository. You create a feature branch, make a change in that branch and test it locally. When you’re ready, submit the change using the delivery review command (part of the Chef Automate command line tool). This command submits the change to Chef Automate and kicks off the pipeline. The command is the equivalent to git push, although it also creates a change in Chef Automate that is similar to a pull request in GitHub and other git-based version control systems.
When Chef Automate receives the change, it triggers the Verify stage. The purpose of Verify is to run checks so that the system can decide if it’s worth the time of a human to review the change.
When the Verify phases have completed successfully, the change is ready for code review. Chef Automate provides integrated code review through its web UI. There is also an integration with GitHub and Bitbucket Server (by Atlassian) for teams with existing code review workflows.
In code review, team members can comment on the diffs. If more changes are required, they can be made either as additional commits on top of the originally submitted feature branch, or the commit(s) can be reworked using git commit --amend and git rebase.
To submit the updates on a feature branch for review, simply run delivery review again. There’s no need to worry about force pushing if you’ve squashed commits. Chef Automate patchset handling will work with your workflow. When you resubmit a change with updates from code review, Chef Automate triggers a fresh run of the Verify stage using the updated feature branch. This can be repeated as necessary. When Verify has passed and the team is happy with the change, it can be approved. Changes are approved by clicking the Approve button in the web UI.
When someone clicks the Approve button, the feature branch that contains your change is merged into the target branch of the pipeline (usually this is master). At this point, the Build stage begins and the same tests that were run in Verify are run again. This is because the target branch may have moved ahead by other approvals. Assuming these tests pass, the Build stage proceeds with the quality and security phases. The Build stage is also a good place to run additional test suites, as well as security scanning checks, that might be too time consuming to run during Verify.
The Build stage concludes with the publish phase. The purpose of the publish phase is to assemble one or more potentially releasable artifacts and make them available to the remaining stages of the pipeline. You can, for example, publish to a Chef server, to Chef Supermarket, and to JFrog Artifactory.
If the pipeline succeeds in generating and publishing the artifacts, then the Acceptance stage begins. This is the first phase that assesses build artifacts rather than source code.
The Acceptance stage is where your team decides whether the change should ship all the way out to its final destination.
During the Acceptance stage, infrastructure is provisioned (if needed), and the artifacts published at the end of the Build stage are deployed. The deployment is verified with automated smoke tests, and then the health of the resulting system is verified by running a functional test suite. At this point, the pipeline pauses and waits for explicit approval from someone who has the “shipper” role. The Acceptance stage is where you can run ad-hoc tests, and perform manual user acceptance testing. For the internal use of Chef Automate at Chef, we have our product owners review changes in Acceptance and decide whether or not to click the Deliver button.
When you click the Deliver button, the change begins its final journey into production. This journey consists of three stages: Union, Rehearsal, and Delivered. These three stages are special for two reasons.
- The first reason is that they are fully automated. Once you ship a change into Union, it will automatically move through the Rehearsal and Delivered stages if all of the automated checks are successful.
- The second reason is that Union, Rehearsal, and Delivered form the shared pipeline. In these stages you evaluate a change in the context of your system as a whole. Ultimately, it is the health of the entire system—not a particular application—that matters. The Union stage gives you a place to evaluate the impact of a change on the consumers of the application being shipped. Each stage in the shared pipeline has the same set of phases: provision, deploy, smoke, and functional.
How stages of the pipeline are associated with actual infrastructure environments is flexible. For example, you can have dedicated infrastructure for each stage. This allows each stage to operate independently.
In this section, we go into more detail about the pipeline. As we’ve said, the Chef Automate pipeline is made up of six stages: Verify, Build, Acceptance, Union, Rehearsal, and Delivered.
Each stage consists of phases that perform a particular task, such as running some type of test.
One way to think about the stages is whether the set of potentially releasable artifacts has been produced or not. The pipeline creates these artifacts at the end of the Build stage. The remaining stages of the pipeline focus on gaining confidence in those artifacts. Another way to understand the stages is by whether they are isolated at the project level or shared across the system. This diagram shows the relationships among the different stages.
- Each project pipeline has an associated Verify, Build and Acceptance stage. These are called acceptance pipelines
- The Union, Rehearsal, and Delivered stages constitute the shared delivery pipeline
- The Verify and Build stages perform tests on the source code
- The Acceptance, Union, Rehearsal and Delivered stages test potentially releasable artifacts
The Verify stage runs automatically when someone submits a new change or updates an existing change that hasn’t yet been approved. It is made up of the following phases. (Remember that you can skip phases that do not apply to your project and you have complete control over what happens in a given phase job.)
- Lint. Run tools that analyze your source code to identify stylistic problems.
- Syntax. Check that the code can be parsed and, if applicable, that it compiles.
- Unit. Run unit tests.
When a change is approved, Chef Automate merges the change into the pipeline’s target branch and triggers the Build stage. The Build stage repeats the lint, syntax, and unit phases from the Verify stage. This is because the target branch may have moved ahead since the Verify stage ran on this change (this occurs if there are multiple open changes on a project and another change is approved before yours).
In addition to the Verify checks, the Build stage provides three additional phases:
- Quality. A place to run additional test suites and code analysis tools. Some tests are too time consuming to run in Verify and are better reserved for changes that have received approval.
- Security. In many organizations, a suite of security tests must be run before a change can be deployed. The Build phase is the place to run such scans and checks. (You can also add compliance checks into the functional test suites that run against the deployed artifacts.)
- Publish. The goal of the publish phase is to produce the potentially releasable artifacts and to make them available to the rest of the pipeline.
Beginning with the Acceptance stage, the pipeline switches from analyzing the project’s source code to verifying the set of artifacts that were produced in the Build stage. The goal of the Acceptance stage is for the team to make a decision about whether the change should go all the way out to production or not. There are four phases in Acceptance:
- Provision. Provision infrastructure needed to test the artifact(s). Examples include instantiating new infrastructure with Chef provisioning (or another API-accessible mechanism) and manipulating Chef server environments to designate the nodes used by the current stage. Of course, what executes in any phase is up to you and determined by the project’s build cookbook.
- Deploy. Deploy the artifacts published in the Build stage to the portion of your infrastructure that has been set aside for acceptance testing.
- Smoke. Smoke tests should be relatively short-running tests that verify that the code that should have been deployed has indeed been deployed and that the system passes minimal health checks.
- Functional. The functional tests should give you confidence that the system is meeting its business requirements.
Union is the first of the three shared pipeline stages. The purpose of the Union stage is to assess the impact of the change in the context of a complete (or as close as possible) installation of the set of projects that comprise the system as a whole. Union is where you are able to test for interactions between interdependent projects. The phases in Union and the remaining stages in the pipeline are the same: provision, deploy, smoke, and functional.
When an artifact is in Union, Chef Automate ensures that any projects that depend on it can only pass their own Acceptance stages by proving their compatibility with that artifact. Chef Automate does this by pinning the versions of the dependencies to the version of the artifact in Union. In this way, Chef Automate forces projects to consume updates to their dependencies as early as possible and prevents them from shipping before proving that they are compatible with the latest version.
If a problem is discovered in Union (it will happen, that is what Union is for), the cooperating teams need to have a conversation about the right fix. Sometimes the fix may require a change on a different project than the one that initiated the break. To fix the break, you submit a new change through the pipeline. Chef Automate is fundamentally a roll-forward system.
Chef Automate ensures that only one change is active in each of the Union, Rehearsal, and Delivered stages at any one time. This orchestration increases safety by encouraging small batch change. In complex systems, identifying root causes of issues in the context of a single change is much easier than trying to analyze larger batches of changes across many different projects. In the future, Chef Automate’s dependency management features will be enhanced to include all concurrent deploys in Union, Rehearsal, and Delivered, as long as they map to completely unrelated dependency sets.
If all phases of Union succeed, then the Rehearsal stage is triggered. Rehearsal increases confidence in the artifacts and the deployment by repeating the process that occurred in Union in a different environment.
If a failure occurs in Union, Rehearsal serves a different and critical purpose. When you submit a new change and it fixes the break in Union, you will have proved that a sequence of two changes, one that breaks the system, and one that comes after and fixes it, results in a healthy system. You do not yet know what happens when you apply the cumulative change to an environment that never saw the failure. Sometimes a fix’s success depends upon state left behind as a result of a preceding failure. The Rehearsal stage is an opportunity to test the change in an environment that didn’t see the failure.
Delivered is the final stage of the pipeline. What “delivered” means for your system is up to you. It could mean deploying the change so that it is live and receiving production traffic, or it might mean publishing a set of artifacts so they are accessible for your customers.
The following diagram shows the servers that are involved in a Chef Automate installation.
The build cookbook, hosted on the Chef server, determines what happens during each phase job. Runners, under control of the Chef server, run the phase jobs. It’s a good idea to have at least three runners so that the lint, syntax and unit phases can run in parallel.
As changes flow through the Chef Automate pipeline, they are tested in a series of runtime environments that are increasingly similar to the final runtime target environment.
Chef Automate allows you to define the infrastructure that participates in each stage. How you map infrastructure environments to pipeline phases is controlled by the build cookbook. In other words, whether a given phase job distributes work to other infrastructure is up to you. There are many ways to map infrastructure environments to pipeline phases, but here are some possible approaches.
Because they test source code, the Verify and Build stages ordinarily run exclusively on the runners and don’t involve other infrastructure. The necessary runtime environments are created and destroyed during the execution of the stage. For example, they can be established using virtual machines created by testing frameworks such as Kitchen.
The stages that test artifacts—Acceptance, Union, Rehearsal and Delivered—almost always need access to additional infrastructure to perform their tests.
For the Acceptance stage, a common approach is to provision one or more nodes that test the deployment. The Acceptance stage nodes for a project are usually dedicated to that project and can be either persistent, or they can be created and destroyed every time the Acceptance stage runs.
For the shared pipeline (Union, Rehearsal, and Delivered), it makes sense to have persistent infrastructure dedicated to each of the stages. Infrastructure environments mapped to Union and Rehearsal should ideally be identical in topology and should correspond as closely as possible to the live infrastructure of the Delivered stage.
You can set up the infrastructure environments either manually or by using automated, on-the-fly provisioning upon first use. The manual approach is simple, but it has the disadvantage of not having an initial run-list for the nodes in the environment. Automated provisioning requires adding code to the build cookbook, but it is more replicable than the manual approach. Automated provisioning also bootstraps the initial run-list for each node in the environment. The delivery-truck cookbook makes it easy to customize your pipeline’s build cookbook for the environments you want to use for each stage of the pipeline.
Currently, Chef Automate manages cookbook version and application attribute version pins using environment objects of the Chef server. The names of the environments in the Chef server correspond to the stages of a pipeline. (This doesn’t mean, however, that the nodes that participate in a given stage need to remain fixed over time.)
It is also possible to share infrastructure among pipeline stages. For example, you can provision infrastructure needed for performing acceptance tests while relying on enterprise services provided by another pipeline stage or even a production environment. Another possibility is to reserve a portion of infrastructure from production to run acceptance testing.