On Mon, Jul 10, 2017 at 5:53 PM, Michael Catanzaro <mike.catanz...@gmail.com > wrote:
> OK, I see the problem. > > On the one hand, systemd needs fixed, as disabled root account is already > a popular pattern on the largest Linux distro out there. This is an issue > for rescue prompts too. That would be great. > But it's true that doesn't help the case where there is no user account at > all. > > One can reboot into Live environment and mount the disks, but that's >> tedious, and most importantly, it requires a reboot, so we lose the >> information about the running system (e.g. which process is stuck in a >> loop). Also, I believe OpenQA uses pre-created root account in anaconda to >> switch to tty2 and upload any error logs in case of any troubles, in a >> fully automated way. So it's not just manual testing affected. >> >> Thoughts on how to resolve that? It seems we need at least the root >> password option kept. >> > > My suggestion is some sort of debug option for Anaconda that would allow > QA to set a root password when needed to investigate something going wrong, > or for OpenQA. It could even allow visiting arbitrary Anaconda spokes. > That's already possible if you edit /etc/sysconfig/anaconda before running > anaconda, which is admittedly one more annoying step that you'd have to > follow before starting Anaconda in order to debug the problem, but it > works. OpenQA could also do that, though that's not superb as then OpenQA > is testing its own special configuration. > That's the problem. We want to test *the very same code* the user runs. If we edit a config file or provide a cmdline argument, we deviate from that and test a different codepath. It might be a minor change, but the purpose here is still test exactly the same as experienced by the user. Only that way we can be sure the composes really work as expected. That's also one of the reasons we use OpenQA (simulating real user input via keyboard and mouse) instead of e.g. modifying the anaconda environment and using e.g. dogtail to control the GUI. If you're completely convinced that those options need to go away, we'll have to rethink our goals. But we'd definitely prefer to continue testing the unaltered environment. Thinking more about it, specifically for OpenQA, I guess we could do something like switch to tty2 once we detect "quit" button is active, and run a command to set root password in the mounted new filesystem. Then switch back and reboot the machine. That doesn't really modify the installer codepath and ensures root is reachable. It's a bit annoying for manually installs, but those are not that important (and for those, we could use the debug option to show the spokes, when we know we're trying to reproduce a specific problem). Adam, wdyt?
_______________________________________________ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
