Ways to run CI/CD with Ravion
You can mix these options. For example, run tests in GitHub Actions on Ravion runners, and build and deploy with a Ravion pipeline.
Separately, stack pipelines apply infrastructure changes for every module. They run no matter which option you choose for your code.
Ravion pipelines
Example pipeline configs for build, test, and deploy.
GitHub Actions runners
Run GitHub Actions jobs on sandboxes in your AWS account.
Stack pipelines
Plan, approve, and apply Terraform for each module.
Ravion pipelines
Ravion pipelines cover everything you use a hosted CI service for:- Triggers — GitHub events such as
push,pull_request, andrelease, GitLab events, schedules, generic webhooks, the API, or a manual run. See Triggers. - CI steps — run tests, linters, and scripts with
customsteps, build images withbuild:image, and build static sites withbuild:static. - Deploy steps — build, deploy, and roll back Ravion modules, and run
terraform:planandterraform:apply. - Flow control —
needsdependencies, matrix steps,parallelandgroupblocks, approvals, andtrigger:pipelineto chain pipelines. - GitHub checks — runs report their status to the commit in GitHub, so you can require them on pull requests. See GitHub checks.
- Fast, cheap compute — steps run in sandboxes that boot in seconds and keep their whole disk as a cache between runs.
main, then deploys only if both succeed:
ravion-pipeline.yaml
GitHub checks
When a pipeline run is connected to a GitHub commit, Ravion reports the run to that commit as a GitHub check. The check shows on the commit and on any pull request that contains it, just like a GitHub Actions or CircleCI check. A run is connected to a commit when either of these is true:- A GitHub trigger started it, such as a
pushorpull_requestevent. - One of its steps builds or checks out a specific commit from a repository connected through the Ravion GitHub App, however the run started.
<project> / <pipeline> (<variant>), for example demo / Build & Deploy (Production). Use that name to make the check required in a GitHub branch protection rule or ruleset. Runs started another way, such as manually, from the API, or on a schedule, add a suffix like · manual. They don’t replace the check that a trigger started.
Every run attempt gets its own check, so a retry posts a fresh check.
Coming from GitHub Actions or CircleCI
Why use Ravion pipelines
1
You get a single pane of glass and CLI/API surface area to see all your tests, builds, deploys, and infrastructure.
2
You get enhanced security and control because you no longer have to open up your database or
image registries to a hosted CI service. Ravion pipelines execute entirely in your cloud account.
3
You can define all your build configuration in the module alongside your runtime config.
GitHub Actions on Ravion runners
GitHub Actions runners are in private preview. Contact support to request access.
runs-on to a Ravion label, and Ravion starts a fresh sandbox for that job in your AWS account:
.github/workflows/test.yml
External CI
Keep building in CircleCI, Buildkite, GitHub-hosted runners, or any other CI system. Your CI builds and pushes the artifact, then starts the deploy with the Ravion CLI:Stack pipelines
Every module stack has a change pipeline and a destroy pipeline. They run Terraform plan → approval → apply whenever a module’s infrastructure config changes or the module is deleted. You don’t author these to ship code. They run alongside whichever CI option you pick. By default, stacks use Ravion-managed system pipelines. Your platform team can replace them with custom pipelines, for example to add policy checks. See System pipelines and customization. Stack pipeline runs appear in your run history next to the pipelines you author. When other docs mention a “stack change pipeline run”, they mean a normal pipeline run that a config change started instead of a git push.Design your pipelines
One pipeline, many environments
A pipeline describes a workflow, not an environment. Don’t create a “Production pipeline” and a “Staging pipeline” — create one pipeline named after what it does, like Build & Deploy, and add a variant per environment. Steps reference module instances through<< pipeline.variant.id >>, so the same steps deploy to whichever environment a run targets:
One workflow per pipeline
A pipeline should contain steps that depend on each other. If two sets of steps never consume each other’s outputs and never need to wait on each other, they are separate workflows — make them separate pipelines, not parallel groups in one pipeline. The common mistake is a pipeline shaped like this:- A deploy that pins the image digest from a build.
- One image build that several module instances deploy.
- A test or
terraform:planstep that gates a deploy or apply. - An approval that several deploys wait on.
- Services that must roll out together, in a fixed order, because of a schema or API contract.
parallel for steps that can overlap inside one of these workflows — build two images before a shared deploy stage, or fan a single build out to several deploys. Reach for group only occasionally: it labels a nested sequence in the UI and can carry a concurrency key for the block as a whole, so it’s useful to hold a multi-step branch inside a parallel block. A pipeline whose top level is nothing but parallel groups is almost always several pipelines. See Blocks for the semantics.
For limits on overlapping runs, see Pipeline concurrency.