Chaos drill 8.C5 — fill the disk

Claim — you can state how the eviction path for disk pressure differs from the one for memory pressure, and name the thresholds that drive each.

Topology — pair, continued from 8.C3.

Footprint note — the worker's disk is 20G and it holds container images, the kubelet's volumes and your fill file. Fill toward a threshold, not to 100%, and read the ceiling first: a node whose root filesystem is genuinely full may not recover without a reprovision, and a reprovision costs the rest of this chain.

Do

1. Record kubectl describe node <worker> conditions and the kubelet's eviction thresholds before touching anything.
2. Fill the volume — chaosd disk-fill on the host, or a StressChaos inside the pod — to just past nodefs.available soft.
3. Watch the DiskPressure condition, the eviction, and which pod was chosen.
4. Compare with what you saw for memory pressure in P6: which is graceful, which is immediate, and which one the pod can be Ready through.
5. Free the space and watch the condition clear after its grace period.

Observe

kubectl describe node <worker> | sed -n '/Conditions/,/Addresses/p'
kubectl get events --field-selector reason=Evicted
ssh zain@10.10.10.131 'df -h /var/lib/kubelet; journalctl -u kubelet | grep -i evict'

Expect — disk-pressure eviction is ranked and graceful; memory pressure ends in an OOMKill by the kernel with no ranking and no notice. The kubelet also garbage-collects images before it evicts anything, so the first response to a filling disk is invisible in the pod list.

Write down — the two eviction paths side by side, with the signal, the actor and the pod's experience for each.

Teardown — delete the fill. Leave the cluster up for 8.C6.