systemd 229-4ubuntu7 ADT test failure with linux 4.4.0-33.52

[Andy Whitcroft]

Testing failed on:
    ppc64el: https://objectstorage.prodstack4-5.canonical.com/v1/AUTH_77e2ada1e7a84929a74ba3b87153c0ac/autopkgtest-xenial/xenial/ppc64el/s/systemd/20160727_170251@/log.gz

[Andy Whitcroft (apw)]

FAIL: test_no_failed (__main__.ServicesTest)
No failed units
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/tmp/autopkgtest.qQr62c/build.1BH/systemd-229/debian/tests/boot-and-services", line 54, in test_no_failed
    self.assertEqual(failed, [])
AssertionError: Lists differ: ['keyboard-setup.service loaded failed failed Set console keymap'] != []

First list contains 1 additional elements.
First extra element 0:
'keyboard-setup.service loaded failed failed Set console keymap'

- ['keyboard-setup.service loaded failed failed Set console keymap']
+ []

[Martin Pitt (pitti)]

The most recent apparmor test (https://objectstorage.prodstack4-5.canonical.com/v1/AUTH_77e2ada1e7a84929a74ba3b87153c0ac/autopkgtest-xenial/xenial/ppc64el/s/systemd/20160729_104947@/log.gz) ran against the same -31.50 kernel that the same test on the same base cloud image still succeeded. It's also the same systemd version, the same cloud, and console-setup also did not change in xenial.

- ['autopkgtest.service loaded failed failed autopkgtest root shell on ttyS1',
- 'keyboard-setup.service loaded failed failed Set console keymap']

It is a bit odd that all of a sudden *two* console/tty related jobs fail to start. (autopkgtest.service is uninteresting for the cloud, but it previously succeeded as well). So I suspect that something in the cloud config changed.

[Martin Pitt (pitti)]

I did a manual run and reproduced the error. It also happens on a current (July 21) xenial cloud image, not just on the rather outdated adt image from April 20.

autopkgtest.service failing is a consequence of https://anonscm.debian.org/cgit/autopkgtest/autopkgtest.git/commit/?id=eff773b -- i. e. previously the error was hidden because it was a backgrounded SysV init job. So there is a /dev/ttyS1 in the VM (otherwise ConditionPathExists=/dev/ttyS1 would fail), but trying to talk to it fails:

$ sudo cat </dev/ttyS1
cat: -: Input/output error

which is reflected in

 Main PID: 1386 (code=exited, status=208/STDIN)

I added a workaround in https://anonscm.debian.org/cgit/autopkgtest/autopkgtest.git/commit/?id=99717f142 .

[Martin Pitt (pitti)]

keyboard-setup.service fails with

  loadkeys[723]: Couldn't get a file descriptor referring to the console

which also happens with running "sudo loadkeys /etc/console-setup/cached.kmap.gz" manually. So /dev/tty0 exists which signifies the presence of ttys, but using them is broken. I added a workaround: https://git.launchpad.net/~ubuntu-release/+git/autopkgtest-cloud/commit/?id=09ee670bc

I retried the failed test on http://people.canonical.com/~ubuntu-archive/proposed-migration/xenial/update_excuses.html for apparmor, please feel free to retry it for all kernels that are affected.

[Martin Pitt (pitti)]

This got rolled out long ago: http://autopkgtest.ubuntu.com/packages/s/systemd/xenial/ppc64el

[Steve Langasek (vorlon)]

It appears that keyboard-setup.service works reliably in both xenial and eoan. So closing this bug task now.