Karpenter
Your Karpenter NodePools, what runs on each, spot against on-demand, pods that can't be placed, and safe edits that warn before they evict.
Karpenter in the sidebar’s Tools reads the cluster’s Karpenter NodePool objects. If Karpenter isn’t installed, the page says so.
NodePools
A card for each NodePool:
- how many nodes it has provisioned, and how many pods and namespaces run on them;
- its spot and on-demand mix;
- the CPU and memory it uses against its own
spec.limits, or no limit.
Unschedulable pods
Pods the scheduler can’t place, with the scheduler’s own message quoted as it wrote it, such as 0/5 nodes are available: 2 Insufficient cpu. That’s where the blocking constraint is named, and often why Karpenter is (or isn’t) provisioning.
Editing a NodePool
Edit on a card opens the NodePool’s YAML in place. Above it, a banner says what the NodePool backs right now, such as Backs 4 nodes / 61 pods / 9 namespaces · 12 pods covered by a PodDisruptionBudget · 3 spot nodes.
Some edits make Karpenter evict pods. Kubebay flags these four:
spec.limits.cpushrinking;spec.limits.memoryshrinking;- a disruption budget’s
nodeslimit being lowered; consolidationPolicyswitching toWhenEmptyOrUnderutilized, so Karpenter may evict pods from nodes that are underutilized, not just empty.
When your edit makes one of these changes, Kubebay lists it under This change looks risky, and Apply stays disabled until you type the NodePool’s name. Dry-run always works, because it changes nothing.
Related
- On Nodes, the NodePool and Capacity type columns show which pool and capacity type each node came from.
- Cost / Waste shows which nodes could be consolidated, and the risk spot reclaims pose.