It's a design flaw. It should be nearly impossible to delete data accidentally. And it should be possible to roll back state in IIAC.
No serious organization with any scale is going to have the only thing standing between them and a production database deletion being two over tired engineers rubber stamping each other's code changes.
Bad config pushes to prod do happen and they can cause outages like the 2024 Cloudstrike outage. You don't want a tool that takes a minor (but significant) error and turns it into a catastrophic one because of poorly thought out semantics. It's better to just start with a tool that requires at least two engineers to explicitly sign off on deletion.
> It's a design flaw. It should be nearly impossible to delete data accidentally. And it should be possible to roll back state in IIAC.
Still not envisioning what your expected solution would look like.
You check the plan, you asses the actions it will take and you either abandon the plan or apply it. You can't roll back destructive changes in cloud environments that's not possible today for obvious reasons.
What IaC does provide is a mechanism to quickly rebuild if you do accidentally wipe out resources.
I've worked in environments were our entire fleet was wiped out by AWS (thousands of VMs) and we were able to rebuild in hours because of IaC.
> No serious organization with any scale is going to have the only thing standing between them and a production database deletion being two over tired engineers rubber stamping each other's code changes.
Most "serious organizations" either have policy as code (Sentinel) or are running Terraform with credentials/roles that have reduced capabilities.
> Bad config pushes to prod do happen and they can cause outages like the 2024 Cloudstrike outage. You don't want a tool that takes a minor (but significant) error and turns it into a catastrophic one because of poorly thought out semantics. It's better to just start with a tool that requires at least two engineers to explicitly sign off on deletion.
This is a criticism of all software/infrastructure deployments with no guardrails. There's nothing stopping you from having two engineers sign off on a TF plan. You can absolutely build that system on top of whatever pipeline you are running.
I find this to be a very strange criticism and is probably indicative of a poor workflow or CI/CD system if anything.