Arvados

Use case: you have a compute node image that's misbehaving because it's misconfigured. It would help to be able to run a job that registers/passes through all the system logs for review.

This could be as simple as running journalctl --follow --boot=0 --lines=all and streaming the output somewhere. Maybe capture a few key log files under /var/log too for common system facilities that commonly don't log to syslog, like /var/log/dpkg.log.

Since this needs to run outside the container, I'm guessing it needs to be implemented in crunch-run directly?

Some problems are post-boot (like #20748), so it would help to be able to have the container run for an arbitrary time while in this mode. e.g., run the busybox image with sleep 15min or something.

Need to consider the security implications of this. Maybe only let admins run containers in this mode?