This raspberry Pi cluster with 3 control plane nodes and 7 workers has been running for 111 days now. Here are some observations:
- /var/log just grows and grows. My master nodes especially seem to be writing to /k3s-service.log a lot, and there’s no log-rollover happening. I can sh into the nodes and remove the files occasionally, while I’m looking for a more automated solution. Ideally, /var/log would be a tempfs mount.
- system-update seems to be working (all my nodes currently show version v1.20.11+k3s1). However, occasionally a node will remain cordoned, and the result will affect pods that use Longhorn storage. When this happens, the “apply” job will get a “Job was active longer than the expected deadline” event posted. Typically, I can resolve these by removing the cordon on the node.
In my next post, I’ll describe how I created the k3os images for the cluster, how I brought the cluster up, and how things have been installed on it.