I still can't reproduce the issue, but I have some more data points. Taking [1] as an example of a "bad" run, we can see that:
- /etc/hosts is mangled via sed (not visible from the logs) - hostname (the package) and cloud-init get upgraded - the system reboots *something* in caused the `sed` to not have effect (/etc/hosts got rewritten? by what? the cloud-init upgrade caused some sort of race?). I tried downgrading the packages, but still couldn't reproduce the problem. Maybe the `sed` should be moved to autopkgtest --setup-commands-boot. For reference, [2] is the postfix bug about it not being to handle the trailing dot. [1] https://autopkgtest.ubuntu.com/results/autopkgtest-noble/noble/amd64/m/mime-construct/20240412_050518_cab97@/log.gz [2] https://bugs.launchpad.net/ubuntu/+source/postfix/+bug/2019195 -- You received this bug notification because you are a member of Canonical's Ubuntu QA, which is subscribed to Auto Package Testing. https://bugs.launchpad.net/bugs/2019472 Title: Trailing dot in fqdn of ps5 VMs regresses some tests (e.g. postfix) Status in Auto Package Testing: Incomplete Bug description: On ps5 amd64 VMs (lcy02) we have that: $ hostname --fqdn adt-paride-mantic-postfix.openstack.prodstack5.lan. Note the trailing dot. This causes regressions in some packages, see for example LP: #2019195. Ideally we should fix the regressed packages (the trailing dot is technically correct), but it makes sense to workaround the problem while the packages are getting fixed. Some possibilities are: (1) Remove `manage_etc_hosts: true` from the cloud-init user-data used by the create-nova-image-new-release script. This should prevent the hostname to be set to the name advertised by openstack, and left to the static string 'autopkgtest'. I did some git archeology and manage_etc_hosts initially came from this snippet: # unbreak my server option :-( userdata=`mktemp` trap "rm $userdata" EXIT TERM INT QUIT PIPE /bin/echo -e "#cloud-config\nmanage_etc_hosts: True" > $userdata and that was the *only* use of user-data back then. Looks like it was a workaround for some issue? In any case I can't be sure that disabling manage_etc_hosts won't cause other issues. (2) Remove the trailing dot in /etc/hosts using the setup-canonical.sh script, e.g. sed -Ei '/^127\.0\.1\.1 /s/([a-z])\. /\1 /' /etc/hosts I don't think this can be racey with cloud-init reconfiguring the hostname because testbed-setup removes cloud-init. Cons of this approach: very hacky. We ask cloud-init to manage /etc/hosts (via manage_etc_hosts), but then mangle it manually (not nice). To manage notifications about this bug go to: https://bugs.launchpad.net/auto-package-testing/+bug/2019472/+subscriptions -- Mailing list: https://launchpad.net/~canonical-ubuntu-qa Post to : canonical-ubuntu-qa@lists.launchpad.net Unsubscribe : https://launchpad.net/~canonical-ubuntu-qa More help : https://help.launchpad.net/ListHelp
