About Terraform Runs in Atlas
A "run" in Atlas represents the logical grouping of two Terraform steps - a "plan" and an "apply". The distinction between these two phases of a Terraform run are documented below.
When a new run is created, Atlas automatically queues a Terraform plan. Because a plan does not change the state of infrastructure, it is safe to execute a plan multiple times without consequence. An apply executes the output of a plan and actively changes infrastructure. To prevent race conditions, Atlas will only execute one plan/apply at a time (plans for validating GitHub Pull Requests are allowed to happen concurrently, as they do not modify state). You can read more about Terraform plans and applies below.
During the plan phase of a run, Atlas executes the command
Terraform performs a refresh and then determines what actions are necessary to
reach the desired state specified in the Terraform configuration files. A
successful plan outputs an executable file that is securely stored in Atlas
and may be used in the subsequent apply.
Terraform plans in Atlas do not change the state of infrastructure, so it is safe to execute a plan multiple times. In fact, there are a number of components in Atlas that can trigger a Terraform plan. You can read more about this in the starting runs section.
During the apply phase of a run, Atlas executes the command
with the executable result of the prior Terraform plan. This phase can change
infrastructure by applying the changes required to reach the desired state
specified in the Terraform configuration file.
While Terraform plans are safe to run multiple times, Terraform applies often change active infrastructure. Because of this, the default behavior for Atlas is to require user confirmation as part of the Terraform run execution. Upon user confirmation, Atlas will queue and execute the Terraform apply. It is also possible to configure Atlas to automatically apply, but this option is disabled by default.
During run execution, Atlas will lock the environment to prevent other plans and applies from executing simultaneously. When the run completes, the next pending run, if any, will be started.
An administrator of the environment can also manually lock the environment, for example during a maintenance period.
You can see the lock status of an environment, and lock/unlock the environment by visiting that environment's settings page.
To receive alerts when user confirmation is needed or for any other phase of the run process, you can enable run notifications for your organization or environment.