Node Controller retries based off the longest requeue, rather than shortest

[njtran]

Version

Karpenter: Latest

Kubernetes: v1.20.0

The Node controller reconciles nodes for emptiness, expiration, finalizers, liveness, and readiness purposes to help reconcile states of nodes provisioned by Karpenter.

Expected Behavior

We expect that the next time we reconcile a node, we would go through the reconcile loop the next time one of the above 5 processes need to happen.

Actual Behavior

We end up requeuing a reconcile for the largest duration desired, instead of smallest.

Steps to Reproduce the Problem

With provisioner fields ttlSecondsAfterEmpty and ttlSecondsUntilExpired set, where empty is set to something like 30s, and expired is set to something like 80000, we will end up re-reconciling not in 30s.