I created an iSCSI volume on my #MyCloud #NAS to hold the #postgres data that the pod uses. ISCSI is fairly fast (85MB/s writes), and more importantly it’s not stored locally on any of the #k3s nodes, so the server can move to any node and still access the data on the volume. That means that I can drain that node, upgrade it, and the downtime is limited to a few seconds.

Basically I’m creating a redundant array of inexpensive compute (I’m running on three $150 mini PCs) to go with my redundant array of inexpensive disks. If a single node or disk fails, the other nodes/disks will pick up the load until the bad unit is replaced.