CFN stack left in DELETE_FAILED state for MNG cluster

[couralex6]

CFN stacks are occasionally left in DELETE_FAILED state, which is causing our CNI CI tests to fail:

Nodegroup cni-test-8287-mng-for-cni failed to stabilize: [{Code: Ec2SecurityGroupDeletionFailure,Message: DependencyViolation - resource has a dependent object,ResourceIds: [sg-01aca506d267ae11a]}]

Looking at the aws-k8s-tester log, it seems like the first try to delete MNG fails:

2021-01-22T23:22:56.4117354Z {"level":"info","ts":"2021-01-22T23:22:56.411Z","caller":"eks/eks.go:1423","msg":"failed Down","error":"MNGs["cni-test-8287-mng-for-cni"] failed to delete failed to delete stack (previous status reason "The following resource(s) failed to delete: [MNG]. ", current stack status "DELETE_FAILED", current status reason "The following resource(s) failed to delete: [MNG]. ")","started":"18 minutes ago"}

However, when querying directly with EKS API, the MNG was deleted. Deleting the stack manually works.

This does not happen on every run.