I Deleted Production With helm upgrade. Here's the Exact Chain of Events.

helm upgrade --install is in every platform team's playbook. It's also the fastest way to delete a PersistentVolumeClaim you've been growing for 8 months if you don't know about the one Helm behavior that nobody puts in the getting-started guide.

I'll walk through the exact sequence: a values file refactor, an immutable field change, Helm deciding a resource needs to be deleted and recreated, and a database volume that went with it. Statefulset. Production. Gone.

The twist: none of this was a bug. It was Helm behaving exactly as documented. We just hadn't read the right paragraph.

I'll cover what Helm actually does during upgrade vs. what engineers think it does, the 3 resource types that silently get nuked during common refactors, and the --dry-run + diff workflow that would have saved us.

• How Helm handles immutable field changes vs. what most engineers assume

• The 3 resource types most likely to get silently deleted during a helm upgrade

• A pre-upgrade diff workflow that catches destructive changes before they happen