Description Philipp Hahn 2013-11-20 17:21:19 CET

Created attachment 5665 [details]
installer.log

+++ This bug was initially created as a clone of Bug #31925 +++
My UCS-3.2 PXE installation on the PCs of our Schulungsraum suddenly started failing with the strange error message, that
  "mkdir /etc/ldap/slapd.conf"
failed, because the *file* already exists.

That is from /var/lib/dpkg/info/slapd.postinst:493
>        echo -n "  Creating initial configuration... " >&2
>
>        # Create the slapd.d directory.
>        rm -rf ${SLAPD_CONF}/cn=config ${SLAPD_CONF}/cn=config.ldif
>        mkdir -p ${SLAPD_CONF}
(IMHO an "mkdir" after the "rm" looks very strange)

The real cause for this is that the PCs were installed multiple times: 1st with MBR, which was then changed to GPT. On 3rd try the above mentioned error occurred.

Digging in installer.log shows that on all subsequent installs, where the HD still contains a valid GPT installation, the previous installation is not removed and a new installation if placed on the still existing file system still containing the previous installation.

$ grep ^'PARTITION-GPT:.*run_cmd' /tmp/installer.log
...
> PARTITION-GPT:1460: run_cmd(['/sbin/pvdisplay', '-c'])
> PARTITION-GPT:1460: run_cmd(['/sbin/vgdisplay'])
> PARTITION-GPT:1460: run_cmd(['/sbin/vgdisplay', '-c'])
> PARTITION-GPT:1460: run_cmd(['/sbin/lvdisplay', '-c'])
> PARTITION-GPT:1460: run_cmd(['/sbin/vgchange', '-ay'])

Here the old VG is visible

> PARTITION-GPT:1460: run_cmd(['/sbin/vgchange', '-an'])

and now no longer

> PARTITION-GPT:1460: run_cmd(['/sbin/parted', '-s', '/dev/sda', 'mklabel', 'gpt'])

so here an empty GPT is created

> PARTITION-GPT:1460: run_cmd(['/sbin/vgchange', '-ay'])

so here no VG can be found any longer

> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/lvremove', '-f', 'vg_ucs'])
> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/vgreduce', '-a', '--removemissing', 'vg_ucs'])
> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/vgreduce', '-a', '--removemissing', 'vg_ucs'])

so these have no effect

> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/parted', '--script', '/dev/sda', 'unit', 'B', 'mkpart', '"BIOS Boot Partition"', '16777216', '25165823'])
> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/parted', '-s', '/dev/sda', 'set', '1', 'bios_grub', 'on'])
> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/parted', '--script', '/dev/sda', 'unit', 'B', 'mkpart', '"/boot"', '25165824', '562036735'])
> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/mkfs.ext4', '/dev/sda2'])
> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/parted', '--script', '/dev/sda', 'unit', 'B', 'mkpart', '"linux-swap"', '562036736', '11299454975'])
> PARTITION-GPT:1301: (profile) run_cmd(['/bin/mkswap', '/dev/sda3'])
> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/parted', '--script', '/dev/sda', 'unit', 'B', 'mkpart', '"LVMPV"', '11299454976', '499973619711'])
> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/parted', '-s', '/dev/sda', 'set', '4', 'lvm', 'on'])

now the old PV is visible again

> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/pvcreate', '/dev/sda4'])

and this fails since no "-ff" is given

> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/vgcreate', '--physicalextentsize', '4096.0k', 'vg_ucs', '/dev/sda4'])
> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/vgextend', 'vg_ucs', '/dev/sda4'])
> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/lvcreate', '-l', '116508', '--name', 'rootfs', 'vg_ucs'])

these all error out because of previous errors: missing PV, VG, LVM, ...

> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/mkfs.ext4', '/dev/sda2'])
> PARTITION-GPT:1301: (profile) run_cmd(['/bin/mkswap', '/dev/sda3'])

these run fine

> PARTITION-GPT:1301: (profile) run_cmd(['/sbin/mkfs.ext4', '/dev/vg_ucs/rootfs'])

but this fails and the installation continues on the old FS leading to an fatal error much later...


In interactive-mode install_fresh_gpt() (which does the "vgc -an ; empty_gpt ; vgc -ay" stuff) is run in the right sequence, but in profile-mode it is run before the old LVM is fully removed.

So currently "create_partitiontable" must be given for an untouched HD, but must not be used if the HD already contains a GPT+LVM.

Adding "-ff" to "pvcreate" might be also a good idea.


Exactly the same error has happened at a customer site (see original bug), were one blade was installed multiple times to get the profile and scripts right.