extend grub boot prompt timeout on platforms with full console access

[dustymabe]

This issue was broken out from an old coreos-assembler PR that attempted to change the grub timeout to 5 seconds.

A lot of work has happened since that PR so it's worth a reset of the landscape and fresh discussion of the issue. With coreos/coreos-assembler#2400 and coreos/fedora-coreos-config#1781 I think we should now have covered the feature request in #110 such that we can now set the timeout per platform (might require some tweaks, but theoretically possible).

The proposal here is to increase the grub console timeout on platforms where it is possible to have read/write access to a console from 1 second to something larger to improve the user experience of catching the console prompt if needed.

More context exists in coreos/coreos-assembler#1263

[bgilbert]

If we have a tradeoff between a relatively uncommon debugging case and the boot time of every Fedora CoreOS system, I'd lean toward making debugging possible but not necessarily easy. After reading through coreos/coreos-assembler#1263, I'd like to hear more about use cases where this would concretely help. I frequently have to catch the FCOS boot menu on multiple platforms (add rd.break, remove console=ttyS0,115200) and the timeout hasn't generally been an issue. I can think of two cases where it's impractically short: bare metal machines that take 20 minutes to POST, and clouds where you can't start the console until the machine has powered on. In both cases, 5 seconds will probably not be enough.

[dustymabe]

Let's say I'm a user and my system won't boot. It hits the grub prompt and then an error shows on screen. At a 1 second timeout it's impossible for me to read the prompt to know what key to hit. Because of my experience I know I can just hit the arrow key to stop the timeout, but it's not discoverable.

I understand not wanting to increase the boot time for the common case. Maybe bumping to 2 seconds could be a reasonable compromise?

[bgilbert]

Actually it looks like almost any key stops the count, including Escape, Backspace, Delete, any F key, Tab, Space, Ctrl-C, and down arrow.

Also, if a user needs to fix a boot failure, hopefully they'd refer to our docs for troubleshooting steps. Those can include detailed instructions on how to get into the boot menu.

[dustymabe]

Added the meeting label to get more input. I'm 100% fine with leaving it at 1s if no other advocates emerge.

Though, we should consider the flip side, which is: on platforms where it's impossible to get to the console, should we set the timeout to 0?

[cgwalters]

Does holding down a key work when timeout=0 too? I think there's an argument that if so, we should go to zero and not have a tiny timeout at all.

[bgilbert]

@cgwalters I did a quick test with timeout=0 in qemu and wasn't able to interrupt the boot at all.

[jlebon]

This was discussed in today's community meeting:

13:15:42 < jlebon> #info there seems to be insufficient desire for a longer timeout. we will
                   not pursue this for now but may reconsider in the future pending more
                   convincing feedback.