Skip to main content
Coming soon. Not in this release. This page says what the feature is, so that the shape is public before the code is.
Move the one step that needs a bigger machine, and keep everything else where you can read it. Most steps in a run are cheap. One or two aren’t: the image build, the suite that takes an hour, the step that needs more than the machine on your desk. A remote node will move one of those steps and nothing else. The step executes on another machine; the runtime stays where you started it, and the event record is written where it always was. The change comes back as a commit or a patch, not as a filesystem to keep in sync.

​
What stays local

  • The graph and its decisions, recorded locally, so the record stays one file on one machine.
  • The approvals. A person answers where the person is.
  • The change, as a commit: atomic, with its own description, reviewable without the machine that made it.

​
Limits

  • It won’t distribute a run. One named step is given a place to execute; steps don’t migrate for load.
  • An unreachable machine fails the step where it stands, with the reason on the record.

​
Next steps