Red Hat Bugzilla – Bug 990828
Cannot use "-" (simple dash/minus) character for volume group name in kickstart partitioning
Last modified: 2014-05-21 01:54:04 EDT
Description of problem: When partitioning in a Kickstart file with volume group names containing a dash or minus sign, the sign is ignored by anaconda during the build. The two issues with this are: 1. It's not documented therefore not expected 2. Can it be fixed or is this an anconda limitation? Version-Release number of selected component (if applicable): RHEL6.4 How reproducible: Steps to Reproduce: 1. Create a kickstart file, In the partitioning, use a "-" in a volume group name, for instance: part pv.2 --ondrive=sda --size=10240 volgroup cinder-volume --pesize=4096 pv.2 2. Install Actual results: The installation will go through without error and will simply remove the character. Expected results: Obvioulsy, the "-" character to stay in place and not being removed.
This is a restriction imposed by anaconda and we've been doing this for a very long time. I think the best way to approach this is to update the documentation.
Hello David, does this restriction apply only to volume group names, or to volume names as well?
Unfortunately, the behaviour was different in RHEL 6.0 to 6.3 so I vote for the bug to be fixed instead of the * fresh systems with volgroup foo-bar created foo-bar as volume group (OK) * reinstalled systems kept dashes foo-bar in existing volume groups (OK) * Linux LVM Tools accept dashes inside of volume group names (OK) RHEL 6.4 shows the following anomalous behaviour and there breaks cobbler "reinstallation" on upgrade from 6.3 to 6.4 wich expected the volume groups to be existing with a dash. * fresh systems with configured volgroup foo-bar get volume group named foobar (FAIL) * reinstalled systems with --noformat keep the dash in volume group foo-bar (OK) * Linux LVM Tools accept dashes inside of volume group names (OK) This is very inconsistent behaviour (especially the difference between noformat and format), which has been introduced with Release 6.4. In cannot tell anything about the logical volume names, since we have only been using dashes in our volume group names (for years now) and we would like to continue doing so, since the volumes exist on many systems, automatically installed with cobbler and our cobbler preinstallation scripts expect the data volumes to contain a dash on existing machines. For now we needed to implement a very dirty workaround which renames the volume group back from foobar to foo-bar in the cobbler postinstall section.
Yes, logical volume names are also affected. Dashes get removed from logvol names on fresh RHEL 6.4 installs.
This limitation actually causes trouble down the road with potentially any system provisionning solution. For instance, on top of Foreman, I have the case where Openstack Cinder expects a "cinder-volume" to be present for automated installation. Of course Cinder default could be changed but the why not fix the trouble at the bottom of the stack, for good? Can someone, please, explain/check why the "-" is an issue in Anaconda and why it cannot be escaped? Cheers
Here's what I've run into, which can hopefully help others seeking answers. It appears (not to point fingers) that when lvm creates the new volume group, if there's a "-" in the name, it somehow doubles that up. (vg_foo-bar becomes vg_foo--bar) To recreate, create a physical volume (e.g. pv.01), a volume group, and a a logical volume. vgcreate foo-bar pv.01 lvcreate -L10m foo-bar The VG Name in vgdisplay and lvdisplay will show as foo-bar but a quick check under /dev/mapper/ will show a dynamic link for foo--bar-lvol0. udev is understandably confused as a query reports the following: udev info --query=all --name=/dev/mapper/foo--bar-lvol0 | grep foo S: mapper/foo--bar-lvol0 S: foo-bar/lvol0 S: disk/by-id/dm-name-foo--bar-lvol0 E: DM_NAME=foo--bar-lvol0 E: DM_VG_NAME=foo-bar E: DEVLINKS=... So, in summation, I'm not certain what's causing the issue, however I believe when lvm writes the block information, it adds an additional "-" to the VG Name. It appears Red Hat was trying to protect us by removing "-" from VG names. As a workaround, you can rename the volume group (vgrename), however you will also need to change your grub.conf and /etc/fstab to reflect. I'm not a big fan of naming the VG "foo-bar" and then putting "foo--bar" in /etc/fstab as that could be confusing, so I change fstab and grub to point to /dev/foo-bar/root. The issue there is, the mount command still reports /dev/mapper/foo--bar-root. It's something you have to live with, unless you can remove "-" from your volumegroup name altogether without issue.
Thanks, thomas, that's interesting and good to know to help the workaround. Maybe someone can explain why this is an Anaconda restriction and if there is any way we could ultimately (RFE) see if the "part" parameter can be fully escaped?
I'm using RHEL 6.5 provisioned via Foreman with this disk layout template: zerombr clearpart --all --initlabel part /boot --fstype ext4 --size=150 part swap --recommended part pv.01 --size=5120 part pv.02 --size=1 --grow volgroup vg_root pv.01 logvol / --fstype=ext4 --vgname=vg_root --name=lv_root --size=1 --grow --percent=100 volgroup cinder-volumes pv.02 The VG is still created without the "-" and I have to manually "vgrename cindervolumes cinder-volumes" after the systems are provisioned. Is is possible that some packages need to be updated in the provisioning media created from the current rhel-server-6.5-x86_64-dvd.iso?
(In reply to Ramon Acedo from comment #18) > The VG is still created without the "-" and I have to manually "vgrename > cindervolumes cinder-volumes" after the systems are provisioned. > This was a fix to the documentation not to anaconda.