ephemeral device filesystems should be labeled with 'ephemeralX'

[Scott Moser]

After bug 827598 is fixed , and there are filesystems already made on the ephemeral devices, it would be a very nice addition to label the filesystems with 'ephemeral0' or 'ephemeral1' corresponding to their block-device-mapping.

This would allow you to very easily add entries to fstab like:
LABEL=ephemeral0 /mnt auto defaults 0 0

Rather than dealing with block-device-mapping which is likely to be wrong.

Thanks to ~ahasenack for this idea.

[Isaku Yamahata (yamahata)]

So what you want is mkfs.ext3 -L ephemeral<N> <target device/file>
It's easy.

[Isaku Yamahata (yamahata)]

I updated lp:~yamahata/nova/lp827598.
Can you please check if that is what you wanted?
If so, I'll request for merge.

[Scott Moser (smoser)]

Maybe I'm reading the diff wrong, but it looks to me like you have:
  utils.execute('mkfs.ext3', '-l', fs_label, '-F', target)

where you'd want:
  utils.execute('mkfs.ext3', '-L', fs_label, '-F', target)

Specifically, the flag for label is upper case L , not lower case.

[Scott Moser (smoser)]

A couple more things. I suspect volumes in windows instances are not formated with ext3. I actually have never looked. If you'd like I *can* dig that up, but the primary point would be that I suspect filesystems type is instance-OS specific (at least to the point of windows/linux).

The other thing is that mkfs on a large volume will take quite a while, so you'd most definitely want to consider that. Maybe cache the volume after mkfs was done?

[Isaku Yamahata (yamahata)]

On Mon, Aug 22, 2011 at 01:11:09PM -0000, Scott Moser wrote:
> Maybe I'm reading the diff wrong, but it looks to me like you have:
> utils.execute('mkfs.ext3', '-l', fs_label, '-F', target)
>
> where you'd want:
> utils.execute('mkfs.ext3', '-L', fs_label, '-F', target)
>
> Specifically, the flag for label is upper case L , not lower case.

Ah, good catch. I'll fix it.

[Isaku Yamahata (yamahata)]

On Mon, Aug 22, 2011 at 01:13:54PM -0000, Scott Moser wrote:
> A couple more things. I suspect volumes in windows instances are not
> formated with ext3. I actually have never looked. If you'd like I
> *can* dig that up, but the primary point would be that I suspect
> filesystems type is instance-OS specific (at least to the point of
> windows/linux).

Okay, os_type is passed to libvirt driver. So we can choose which
file system to use. So which file system should be used?

os_type
   'linux' => mkfs.ext3
              mkfs.ext3 -L <volume-label>

   'windows' => mkfs.vfat
                do we want to use
                -n <volume-name>
                -i <volume-id>

                or

                mkfs.ntfs
                do we want to use
                --label STRING
                --fast: perform a quick format

   other os => We don't know. (Solaris, BSD* ...)
                leave it uninitialized or make it user configurable?

> The other thing is that mkfs on a large volume will take quite a while,
> so you'd most definitely want to consider that. Maybe cache the volume
> after mkfs was done?

Fortunately the files is cached in /var/lib/nova/instance/_base/ as you
described.
If ext4 could be used, uninit_bg option could be used. It reduces mkfs
time much.

[Isaku Yamahata (yamahata)]

Okay, I updated the branch with nova-trunk.
It implements options on how to mkfs a given ephemeral device depending on os_type.
Can you please check it?

[Scott Moser (smoser)]

OK, after *way* to many failures, I finally succeeded in launching a windows m1.large instance with 2 ephemeral devices attached on amazon.

Each of the block devices are formatted as NTFS and have no label.

So, I suggest making the default case NTFS and the comment able to change to vfat.

[Isaku Yamahata (yamahata)]

Thank you for confirming it. I changed the default value to use ntfs and did request-for-merge of the branch.

[Scott Moser (smoser)]

fix-committed at revno 1562.