It would be useful to have a rollback command like atlantis rollback -p projectname where it could pull from master or whichever branch the current branch is based from and re-apply to rollback changes. This way if someone applies a project and it fails, they can call for a rollback while they go work on another commit to fix things.
Hi Michael,
I think this would get too complicated. I think in this case it's better for you to manually intervene or close the pull request and open up a new one to revert the changes.
@lkysow -
What about the case where a pull request might be running Terraform in multiple directories/roles, and one succeeds while the other fails? This is a situation where a rollback would be good to have to abort the PR in flight.
Or in the case that there are multiple workspaces being changed at once..
Hi Sharif,
I agree that it would be useful, however I think the implementation would be too complicated to do in Atlantis (and rare, so even if it was doable it might be better to work on other features). In these situations it's better to manually intervene.
Agreed with @lkysow.
In general, we tend to organize tf code into many small independent workspaces to reduce blast radius of the tf code contained within the workspace to make rollback easier for this very reason.
In cases where something goes wrong, it is always better to manually check, verify what state your infrastructure is in currently (using terraform plan + looking at the infrastructure directly) and decide what steps to take.
Most helpful comment
Hi Sharif,
I agree that it would be useful, however I think the implementation would be too complicated to do in Atlantis (and rare, so even if it was doable it might be better to work on other features). In these situations it's better to manually intervene.