Preconditions

• Robot turned ON

Steps

1. Clean up and baseline

rm -f /data/fill_*.img /data/scratch.img /data/directtest.img
umount /mnt/scratch 2>/dev/null
sync
df -h /data

2. Fill /data to 98%

TARGET_PCT=98
total=$(df --output=size -B1 /data | tail -1)
used=$(df --output=used -B1 /data | tail -1)
target_used=$(( total * TARGET_PCT / 100 ))
count=$(( (target_used - used) / (1024*1024) ))
echo "Will write ${count} MB to bring /data to ${TARGET_PCT}%"
dd if=/dev/zero of=/data/fill_1.img bs=1M count=$count status=progress
df -h /data

3. Observe behavior at 98%

df -h /data
kubectl get pods -A
kubectl describe node ea5-control | grep -iA6 conditions

4. Recover immediately

rm -f /data/fill_*.img
sync
df -h /data
kubectl get pods -A

Expected

• At 98% the node reports a DiskPressure condition and may evict pods.
• After recovery, DiskPressure clears and all pods reschedule to Running/Ready.

Notes

Evicted pods may need a moment to restart — watch until they're all healthy.