Bug 1141398

Not an Alpha blocker as if I boot with just one disk or the other selected I can see the PV and delete it (and hence Alpha criterion "The installer must be able to complete an installation to a single disk using automatic partitioning." is satisfied). But proposing as a Beta blocker - https://fedoraproject.org/wiki/Fedora_21_Beta_Release_Criteria#Custom_partitioning :

"Correctly interpret, and modify as described below, any disk with a valid ms-dos or gpt disk label and partition table containing ext4 partitions, LVM and/or btrfs volumes, and/or software RAID arrays at RAID levels 0, 1 and 5 containing ext4 partitions"

Created attachment 937140 [details]
anaconda.log

Created attachment 937141 [details]
ifcfg.log

Created attachment 937142 [details]
journal output

Created attachment 937143 [details]
program.log

Created attachment 937144 [details]
storage.log

# mdadm --detail /dev/md127

/dev/md127:
        Version : 1.2
  Creation Time : Wed Sep 10 23:59:25 2014
     Raid Level : raid1
     Array Size : 15719424 (14.99 GiB 16.10 GB)
  Used Dev Size : 15719424 (14.99 GiB 16.10 GB)
   Raid Devices : 2
  Total Devices : 2
    Persistence : Superblock is persistent

  Intent Bitmap : Internal

    Update Time : Fri Sep 12 21:14:13 2014
          State : clean 
 Active Devices : 2
Working Devices : 2
 Failed Devices : 0
  Spare Devices : 0

           Name : localhost:pv00  (local to host localhost)
           UUID : dbc9631b:d5d77065:42c4a0f3:ae5a2870
         Events : 152

    Number   Major   Minor   RaidDevice State
       0     252        2        0      active sync   /dev/vda2
       1     252       17        1      active sync   /dev/vdb1

[root@localhost tmp]# lvdisplay -a
  --- Logical volume ---
  LV Path                /dev/fedora-server/root
  LV Name                root
  VG Name                fedora-server
  LV UUID                C70iud-55Tp-D8Le-NyIY-C0dZ-6kmx-FCW41P
  LV Write Access        read/write
  LV Creation host, time localhost, 2014-09-10 23:59:27 -0400
  LV Status              available
  # open                 0
  LV Size                12.91 GiB
  Current LE             3304
  Segments               1
  Allocation             inherit
  Read ahead sectors     auto
  - currently set to     256
  Block device           253:3
   
  --- Logical volume ---
  LV Path                /dev/fedora-server/swap
  LV Name                swap
  VG Name                fedora-server
  LV UUID                cqgBko-Rpa5-kJ4T-z0EN-V0uu-CpB4-ocO7iI
  LV Write Access        read/write
  LV Creation host, time localhost, 2014-09-10 23:59:33 -0400
  LV Status              available
  # open                 2
  LV Size                2.00 GiB
  Current LE             512
  Segments               1
  Allocation             inherit
  Read ahead sectors     auto
  - currently set to     256
  Block device           253:4

[root@localhost tmp]# parted /dev/vda print
Model: Virtio Block Device (virtblk)
Disk /dev/vda: 21.5GB
Sector size (logical/physical): 512B/512B
Partition Table: msdos
Disk Flags: 

Number  Start   End     Size    Type     File system  Flags
 1      1049kB  525MB   524MB   primary  ext4         boot
 2      525MB   16.6GB  16.1GB  primary               raid

[root@localhost tmp]# parted /dev/vdb print
Model: Virtio Block Device (virtblk)
Disk /dev/vdb: 16.1GB
Sector size (logical/physical): 512B/512B
Partition Table: msdos
Disk Flags: 

Number  Start   End     Size    Type     File system  Flags
 1      1049kB  16.1GB  16.1GB  primary               raid

This is related to md UUIDs containing ':' instead of '-'. I know you've made some changes in this area -- do you think they could have caused this issue? I'm asking because this is a new problem and I don't think anything has changed in udev, lvm, or mdadm.

Update: this affects live, but not netinst, something in live is screwing with things, presumably.

Bug reproduced with RC1, let me know if there's anything I can provide.

I note that on the RC1 boot, the systemd unit 'lvm2-pvscan@9:127.service' appears to fire during boot and do something:

Sep 19 13:41:46 localhost lvm[728]: 2 logical volume(s) in volume group "fedora-server" now active

then something more when anaconda starts up:

Sep 19 13:56:20 localhost lvm[2220]: No PV label found on /dev/md127.
Sep 19 13:56:23 localhost lvm[2355]: 2 logical volume(s) in volume group "fedora-server" now active

The F20 Final live works, and the logs from that service are rather different. On boot:

Sep 19 14:03:15 localhost systemd[1]: Starting LVM2 PV scan on device 9:127...
Sep 19 14:03:15 localhost pvscan[717]: 2 logical volume(s) in volume group "fedora-server" now active
Sep 19 14:03:15 localhost systemd[1]: Started LVM2 PV scan on device 9:127.

which perhaps indicates the scan never finishes on F21? I don't believe there's a matching "Started" message for the "Starting", I'll go check again shortly. On F20, when anaconda starts up:

Sep 19 14:11:13 localhost systemd[1]: Stopping LVM2 PV scan on device 9:127...
Sep 19 14:11:13 localhost pvscan[2181]: Device 9:127 not found. Cleared from lvmetad cache.
Sep 19 14:11:13 localhost systemd[1]: Stopped LVM2 PV scan on device 9:127.
Sep 19 14:11:13 localhost systemd[1]: Starting LVM2 PV scan on device 9:127...
Sep 19 14:11:13 localhost pvscan[2241]: 2 logical volume(s) in volume group "fedora-server" now active
Sep 19 14:11:13 localhost systemd[1]: Started LVM2 PV scan on device 9:127.

Not sure if this is relevant, but it looked interesting.

We hope that these will all turn out to be dupes.

Although in this case we can see that the array was actually found with name pv00.

The problem is, instead, that the status is checked and the uuids are verified against each other and one is the wrong format so they turn out not equal and status is False. This should be fixed by the patch for the other bugs, since we don't just sanitize the results of mdadm methods, but also the results of udev methods.

Discussed at 2014-10-03 blocker review meeting: http://meetbot.fedoraproject.org/fedora-blocker-review/2014-10-03/f21-blocker-review.2014-10-03-15.58.log.txt . Accepted as a blocker per criteria cited in nomination.

Same as bz#1147077 fix.

Fix for this should be in 21 Beta TC3 (not TC2).

Fix confirmed in TC3, and finally I can stop keeping that VM lying around!

0.61.4 is now pending stable push, so let's just close this out.

I have the same issues with Fedora 21 Live release.

I have an intel imsm raid-1 and a standalone SSD. Both contain partitions that belong to LVM PV of two different VG. There is another Fedora installations on them. I want to install F21 into some existing LVM partitions.

When I run "install to the hard disk", anaconda proposes me selection with two devices "Raid volume" and "OCZ SSD". I select both with "custom layout". On the next screen I can see the existing fedora installation with the logical volumes only on SSD. 

I can not see any volumes on top of Raid Volume neither in existing Fedora installation nor under "Unknown". Instead there is /dev/md126p2 marked as unknown partition with the filesystem of "LVM PV".

Created attachment 971299 [details]
the storage.log

Created attachment 971300 [details]
anaconda.log

(In reply to Oleg Samarin from comment #17)
> I can not see any volumes on top of Raid Volume neither in existing Fedora
> installation nor under "Unknown". Instead there is /dev/md126p2 marked as
> unknown partition with the filesystem of "LVM PV".

I looks like the raid contains an unformatted LVM PV partition

14:07:18,962 WARN blivet: PV md126p2 has no vg_name
14:07:18,962 WARN blivet: PV md126p2 has no vg_uuid
14:07:18,962 WARN blivet: PV md126p2 has no pe_start

If there is a volume group on it, even a partial volume group, we should have gotten back some information about it from udev.

But lvm recognises this partition successfully.

How can I debug this with working system without running anaconda?

Hi!

When you are running anaconda and you get to the screen you describe in comment 17, you can try switching to a terminal and running the pvs command, like:

pvs --unit=k --nosuffix --nameprefixes --unquoted --noheadings -opv_name,pv_uuid,pe_start,vg_name,vg_uuid,vg_size,vg_free,vg_extent_size,vg_extent_count,vg_free_count,pv_count

This is the precise command blivet uses to gather information about the pvs on any visible device.

Please report the result.

Please also run the same command with the additional flag --all and report that result as well.

This will allow us to discover what information pvs is reporting for the pv at that time.

Another helpful thing to do is to include all logs. Especially useful in this context would be to have included program.log, which would tells us more about pvs output than is available from storage.log.

Today I've made more research. I need to start anaconda two times for obtaining an installer screen. Anaconda does nothing visible at the first start.

But something happens at this time: 

before: lvm has two PVs: md126p2 and sdc2
after: lvm has another PVs: sdb2 and sdc2

(the md raid consists of two disks sda and sdb).

So blivet indicates correctly at the second dime that md126p2 is no more PV.

The remaining question is what and why does anaconda for lvm to use sdb instead of md126p2?

Created attachment 972148 [details]
pvdisplay result before starting anaconda

Created attachment 972150 [details]
program.log of the first run

Created attachment 972151 [details]
storage.log of the first run

Created attachment 972152 [details]
pvdisplay result after the first (unsuccessful) anaconda run

I can'not attach anaconda.log because it has zero length

Starting from Comment #17 this is a whole new bug, so I went ahead and cloned it.

All further comments about that problem should be on #1178181.