Workflows and runs
Where workflow files are read from, what starts a run, where it executes, and the statuses it goes through.
Workflow files
| Folder | Format | Files read |
|---|---|---|
.dcloud/workflows/, .ews/workflows/ |
Native (Workflow syntax) | .yml, .yaml and .json |
.github/workflows/ |
GitHub Actions | .yml and .yaml, directly in the folder |
When .dcloud/workflows/ and .ews/workflows/ hold a file of the same name, the one in
.dcloud/workflows/ is read. A workflow is read at the commit the event names, so a change to a
workflow takes effect on the commit that makes it. Three exceptions: a manual start reads the file
at the tip of the ref you name, a required check reads the protected branch's copy (see
Required checks), and a re-run repeats the file its original run used.
A file that does not parse never blocks a push; it starts no run. Check a file before you commit it (see Add a workflow).
What starts a run
- A push of a branch or tag that the file's
pushtrigger matches. Merging a pull request, committing in the console, creating a branch and updating a pull request's branch start what the same push would. An annotated tag is resolved to its commit first. A push made with a running job's own token starts no run, so a job that pushes a tag or a version bump does not start itself again. - A pull request whose target branch the
pull_requesttrigger matches, built from the pull request's commit. - A merge queue build (
merge_group). - A manual start of a file that allows it (see Start a workflow by hand).
- In GitHub-syntax files also
schedule,releaseandworkflow_run(see Run GitHub Actions workflows). - A re-run of an earlier run (see Cancel or re-run a run).
In the native format, cancelInProgress: true cancels the queued and running runs of the same
workflow name on the same ref when a new one starts, with "Superseded by run #…".
Where a run executes
| The jobs ask for | The run executes on |
|---|---|
Linux only (native linux, GitHub ubuntu-*) |
One platform machine created for the run and deleted after it; its jobs share it |
| Labels a runner pool carries (GitHub syntax) | The pool's machines, one job per machine |
macos (native) |
One of your runners, which takes the whole run; it stays queued until one does |
self-hosted or your runners' labels (GitHub) |
Your runners, which take the run's jobs one by one |
A native run that mixes a linux and a macos job can only be taken by one runner carrying both
labels, and a Mac cannot carry linux: put the macOS jobs in a workflow file of their own. See
Runners and runner groups.
Statuses
| Status | Meaning |
|---|---|
queued |
Waiting: for a free slot under the organization's ceiling on concurrent runs, or for a runner |
provisioning |
The machine is being prepared |
running |
Jobs are executing |
succeeded |
Every job finished successfully |
failed |
A job failed, or the run could not be planned or started |
cancelled |
Cancelled by a person, or superseded by a newer run |
timed_out |
The run passed its time limit |
A queued run says why it waits (see Watch a run). A run on a platform machine or a Deutsche Cloud agent may last two hours. A run whose jobs go to GitHub's runner, yours or a runner pool's, may last 6 hours in a trial organization and 72 hours otherwise.