peaks-1hrr0s0 has the whole Monster Market defined and almost none of it running. The manifests are in ~/peaks-1hrr0s0/.

kubectl get nodes
kubectl get deploy,ds,pods -n peaks-1hrr0s0 -o wide

market-web     Pending   wants a label no node has
market-db      Pending   wants the ssd node, which is cordoned
market-worker  Pending   wants the batch node, which is tainted
market-cache   2/4       anti-affinity, one node cordoned, three nodes in total
log-agent      2/3       skips the tainted node

For each Pending Pod, kubectl describe pod names the rule it fails. Decide whether to change the node or the workload.

Hints

• web selects tier=frontend. The intended node is peaks-node-2; label it, do not edit the Deployment.
• db selects disk=ssd. The node exists and is cordoned; the maintenance is over.
• worker must run on peaks-node-3, which is tainted dedicated=batch:NoSchedule on purpose. The worker is the batch workload: give it the toleration in worker.yaml and apply the file.
• cache has required anti-affinity to itself and already tolerates the batch taint. Three nodes, four replicas: lower the replica count.
• log-agent has to run everywhere, including the batch node. Add a toleration to its template in agent.yaml.

kubectl get deploy,ds -n peaks-1hrr0s0

Everything at its full count and the DaemonSet on all three nodes. The last objective asks which flag kubectl drain needs when a node runs the log agent.