Deployments

A deployment is one workflow against a stack. Most deployments use the plan-and-apply flow:

1. GantryCD runs tofu plan.
2. A plan with no changes finishes.
3. A plan with changes waits for approval.
4. Approval starts tofu apply with the saved plan.

The person who approves can choose to apply, plan again with targets, or skip the deployment.

Other modes

Refresh updates the state from the real infrastructure.

Destroy plans the removal of the stack and waits for approval before it applies.

Run command runs a shell command with the stack’s environment and credentials. Treat it as an emergency tool. It does not follow the normal plan-and-apply guard.

Waiting deployments

A stack runs one state-changing deployment at a time. New work waits behind the active deployment. You can move urgent work forward or skip work that is no longer needed.

By default, pushes do not pile up. A newer push skips the older push deployments still waiting on the stack, even for the same commit, and they show as Replaced by newer. Other deployments, and waiting work you moved forward, are left alone. Turn this off in the stack’s Queue settings to keep every push in order.

Cancelling asks an active runner to stop. Work that has not started is stopped at once.

Retry creates a new deployment from the same source commit. Use it when a temporary failure has passed. Use a new commit when the code needs to change.