By: Tom Sydney Kerckhove <syd@cs-syd.eu>
The readiness probe must not panic when the connection goes away Caught by the tool test, and it would have caught a real deployment: wait runs while the router is rebooting, so a read that fails half way through is the expected case and not something to end the process over. It ended it with a rust backtrace and exit code 101. So there is a try_capture_command_remotely that says what went wrong instead of unwrapping, and wait uses it. The reason the last deployment survived this is that the connection happened to fail at connect time rather than mid-read.