A node of ours stopped starting pods for hours while it reported Ready. The root cause: containerd's snapshot GC called a proxy snapshotter plugin (soci in our case) with no deadline, the plugin stopped answering, and GC sat on a lock that everything else…