ros install stalls (agetty issue)

[rvlander]

RancherOS Version: (ros os version)
0.6.0

Where are you running RancherOS? (docker-machine, AWS, GCE, baremetal, etc.)
pcengine apu via serial port (there is no graphic output on this board).

After typing :
sudo ros install -d /dev/sda

I am prompted :
Continue [y/N]

I type y + Enter and then nothing happens and ros seem stalled. When I quit the process and mount the partition on sda, there is nothing on it.

Note: I can boot via serial port changing boot options. However, I can't see the boot sequence.

[joshwget]

Can this be consistently reproduced, or does this only happen some of the time? Does the same thing happen when installing via SSH?

Regarding seeing the boot sequence, you might need to add the appropriate TTY as a kernel parameter, e.g. console=ttyS0.

[rvlander]

@joshwget I made a usb boot key using unetbootin which generate two grub entries, I already modified its 2 entries using console=ttyS0, 1152000.

The fact that there is no boot sequence is due to the fact that the first entry is generated by unetbootin doing some manipulation on the kernel he finds in the iso. I can boot using this entry (but then ros install stalls). I will try to connect via ssh once booted and come back to you.

Using the second entry (which boots the rancheros kernel as is), the boot sequence is displayed but there is a kernel panic as reported in #984

Trying to use IPXE, I have an error during boot sequence saying that rootfs can't be uncompressed because it appears to be a initrd...

[joshwget]

I've actually been running into a similar issue when trying to switch consoles in a serial console with QEMU. Pretty sure these two are related.

[SvenDowideit]

very interesting. I started kicking this, and it looks like the qemu problem only happens for some consoles, and only on ttyS0 IF you havn't ros console switched in another shell (ssh for eg) previously.

doing a fresh boot with a debian console previously enabled doesn't seem to have a problem (ok, I'm going to try a fresh install with debian just in case)

its not just ros, read -p 'hit enter' doesn't work.

yup, ./scripts/run --append "rancher.debug rancher.console=debian" --fresh meant that read and sudo ros console switch default allowed me to hit y.

so far, I've tested:

• broken: default, centos and the new alpine console I'm working on for 0.8.0
• working: debian, ubuntu

stuffit, starting a git bisect

and boom, its a side effect of the rancher.autologin=ttyS0

[SvenDowideit]

@rvlander, if you remove that from your setup, and add rancher.password=rancher after you login, does the install prompt work again?

or alternatively, if you ros console enable debian and reboot, that works for me.

[SvenDowideit]

and to add insult to injury, using /sbin/getty -n -l /bin/sh 115200 ttyS0 works (you're logged in as root tho), while /sbin/agetty --autologin rancher 11520 ttyS0 doesn't.

but this only fixes the default and alpine consoles, not centos and fedora - which of course have softlinked a modified agetty to getty.!?

[SvenDowideit]

it'd be nice if respawn supported SIGHUP for re-reading the cfg.

[SvenDowideit]

Next TODO is to compare the code for the busybox / alpine getty and agetty as switching between them makes a difference. I fear the centos/fedora pair might have more todo with an assumption of systemd magics, but hopefully that's just my ignorance talking.

[SvenDowideit]

I've just experienced this on a normal hardware console while testing 0.8.0-rc3 - so it looks like this will get more attention very soon.

sadly not repeatably