In Focal, there is a bug (#1859858) in udev which returns a
truncated value for ID_SERIAL on some scsi devices; this does
not break dname rules when the ephemeral environment matches
the target system, however if Focal deploys Bionic or in this
case Centos70, the udev in the target system produces a
different ID_SERIAL value which prevents the dname rules from
matching. Skip this test until we get a udev fix.
vmtests: fix network mtu tests, separating ifupdown vs networkd
On networkd based releases, some of the existing ipv6 MTU tests
will not work, by design. The curtin ifupdown helper scripts
installed ensure that the base network interface mtu is always
set correctly; networkd requires an explicit device MTU setting
in addition to an IPV6 MTU value if they are to differ.
- Add a networkd specific mtu test config
- Add a _skip_if_not_ifupdown to avoid tests when on networkd
- Update scripts to capture additional debugging info when on
networkd
Additional changes needed:
- Add '---' before adding the service masking to ensure
we disable snapd during vmtest
- Refactor journalctl commands to separately sync, flush to
ensure we capture the compelte journal when we collect
- allow test class to specify add_repos, system_upgrade and
upgrade_packages, env still wins if set
- use systemd from ppa:ddstreet/systemd until ipv6 mtu
bug is fixed
4fa2d66...
by
Mike Pontillo <email address hidden>
This branch adds support for deploying CentOS and RHEL 8 images. As a
side effect this also enables secure UEFI boot for CentOS 7 when the
grub2-efi-x64 package is installed.
The initramfs-only-one branch modified how we use "dirty-disk" mode
which previously had unset environment variables used by curtin,
OUTPUT_FSTAB; this had some undesirable side-effects and instead
we set OUTPUT_FSTAB to a tmpdir and block-meta will now verify
expected environment variables. We missed a variant of the "dirty-disk"
mode used to recreate a specific bcache bug and now we see stack traces
warning about unset OUTPUT_FSTAB in vmtest output:
On s390x, and aarch64, the kernel package directly calls
arch specific commands (zipl, flash-kernel) which fail to run
correctly if an initramfs isn't already created in the image. This
broke installation when we attempt to only create the initramfs
once. This branch adds arch support into the code path to divert
not only 'update-initramfs' but the arch specific tools as well and
restores them prior to running the real 'update-initramfs' command.
clear-holders: handle arbitrary order of devices to clear
At some point, the vmtest for LVM over bcache device list to clear
was sorted in a way that exposed a bug in the level calculation
resulting in not determining that the LVM device must be shutdown
before the bcache layer.
Resolve this by three changes: First we now sort the holders trees so
we have a consistent plan. Second when adding devices to the shutdown
registry, if the device has not been seen we take the max of either the
current level, or the size of the devices holders and lastly, the level
calculation requires recursion to find the full depth rather than just
taking the length of the holders list.