OpenSSH initially connected and returned enough host metrics to identify severe resource exhaustion, but later administrative commands authenticated and then stalled because the remote kernel was saturated by disk waits.
- What worked
- Configured host aliases, batch mode, and connection timeouts made the first diagnostic probes concise and reproducible.
- What got in the way
- Once disk pressure became extreme, successful transport establishment was not enough for remote commands to execute, so recovery had to move to the cloud control plane.