CloudFormation stuck in UPDATE_ROLLBACK_FAILED (and DELETE_FAILED)
A rollback fails when CloudFormation cannot restore a resource. Fix the cause, then continue the rollback, or skip the resources that cannot be restored.
What you see
- Stack status
UPDATE_ROLLBACK_FAILED - Stack status
DELETE_FAILED No updates are to be performed
What it means
When an update fails, CloudFormation rolls back. If it cannot return a dependent resource to its old state, the rollback itself fails and the stack is stuck. The docs' example: rolling back to a database instance that someone deleted outside CloudFormation. CloudFormation does not know, so it tries to restore a resource that no longer exists.
Changes made outside CloudFormation are the single most common root cause. After a rollback that skipped resources, the real state no longer matches the template.
How to confirm the cause
- 1Find the failing resource and its reason
aws cloudformation describe-stack-events --stack-name STACK_NAME --max-items 20
Read
ResourceStatusReasonon the firstFAILEDevent. In the console, use the operation ID filter. - 2Check whether someone changed resources by hand
aws cloudformation detect-stack-drift --stack-name STACK_NAME
Drift detection compares the real resources with the template.
Causes and fixes
A resource was changed or deleted outside CloudFormation
- HOW TO TELL
- The failing event says a resource does not exist or does not match.
- ROOT CAUSE
- CloudFormation assumes the resource is as it left it. When it is not, restoring the old version is impossible.
- FIX
- Bring the resource back to match the original template (for a deleted one, recreate it with the same name and properties), then continue the rollback.
Insufficient permissions
- HOW TO TELL
- The reason is an access denied on a modify action.
- ROOT CAUSE
- Your policy lets you create a resource but not modify it, so the rollback cannot change it back.
- FIX
- Add the missing permissions, then continue the rollback.
Quota, expired credentials or a resource that did not stabilize
- HOW TO TELL
- The reason says limit exceeded, invalid security token, or timeout.
- ROOT CAUSE
- These are temporary or fixable without changing the stack.
- FIX
- Delete unneeded resources or request a quota increase. For a token or timeout, no change is needed; continue the rollback.
Nothing else works
- HOW TO TELL
- You cannot fix the resource.
- ROOT CAUSE
- The resource cannot be restored.
- FIX
- Continue the rollback while skipping it. CloudFormation marks it
UPDATE_COMPLETE, but the real state will not match the template, so reconcile it before the next update.aws cloudformation continue-update-rollback --stack-name STACK_NAME --resources-to-skip LOGICAL_RESOURCE_ID
DELETE_FAILED: resource is not empty or protected
- HOW TO TELL
- Stack is
DELETE_FAILED. - ROOT CAUSE
- Some resources must be empty first, such as an S3 bucket with objects. Termination protection also blocks deletion.
- FIX
- Empty the resource, disable termination protection, or retry the delete while retaining the resource that cannot be deleted.
aws cloudformation delete-stack --stack-name STACK_NAME --retain-resources LOGICAL_RESOURCE_ID
How to stop it coming back
- Never edit CloudFormation-managed resources by hand. Change the template.
- Run drift detection on a schedule.
- Use change sets and review them before applying.
Keep going
- Runbook: AWS "is not authorized to perform" (AccessDenied)
- Cheat sheet: AWS CLI
- Lesson: the iac topic in Learn
Sources
AWS documentation (docs.aws.amazon.com), read 18 Nov 2026.