i prefer current era where i never have to ssh to debug a node. if a node is misbehaving or even needs a patch i destroy it. one command, works every time.
That’s what telemetry services are for. If you have all the logs and metrics from the host, then you can research and construct the story from those. You don’t necessarily need the host to be alive anymore.
oh, i am interested, but i can’t remember the last time i needed ssh to figure out the issue, or needing to fix a node besides by destroying it. last time it was a silly app deciding to use a host volume on root partition to cache stuff using all the disk space. remediate in the moment by destroying node, fix it forever by moving the app to a node type with instance attached NVMe device and putting the volume there + container that nukes the data volume if it runs out of space.
