Kubernetes Pod Disruption Budgets during cluster autoscaler scale-down

Running a 40-node EKS cluster with cluster-autoscaler. We have PDBs set to minAvailable: 75% on several stateless services, but during scale-down events the autoscaler respects the PDB and then gets stuck — can't drain the node, can't terminate it, and the scale-down never completes. How are you handling PDBs in autoscaling scenarios? Options we've considered: 1. Relax PDB to maxUnavailable: 25% instead of minAvailable 2. Add a 'scale-down-friendly' PDB profile with lower availability during maintenance windows 3. Use cluster-autoscaler's --skip-nodes-with-system-pods flag more aggressively 4. Accept longer scale-down times and let CA timeout gracefully What's actually working in production for you?