openstack team mailing list archive

[Scott Moser <smoser@xxxxxxxxxx>]

On Tue, 7 Jun 2011, Muhammad Atif wrote:

> Hi Openstack gurus,
>
> We have recently shifted from eucalyptus to openstack and so far, finding it
> quite stable. However, we are having terrible time re-bundling an existing
> image.
>
> The re-bundled image is able to complete the boot process upto mountall: but the
> cloud-init never fires up.
>
> Here is what we have done so far, please let us know what exactly is wrong on
> our side.
>
> 1- Downloaded the ubuntu1010 image from openstack site as follows:
>     wget
> http://c0179148.cdn1.cloudfiles.rackspacecloud.com/ubuntu1010-UEC-localuser-image.tar.gz

Does anyone know what this image is ?
I'm guessing that Muhammad found reference to it from
http://wiki.openstack.org/GettingImages . The page there mentions
http://uec-images.ubuntu.com, but the code snippit points to something on
http://c0179148.cdn1.cloudfiles.rackspacecloud.com/ .

I'm happy that people find the uec-images.ubuntu.com images useful, and
one reason that we put the files there is so that people can easily use
them as a starting point for recreating images.

However, if there is something that was insufficient about the official
ubuntu images, I'd like to know what it was so that we can make them
better.  What was it that caused someone to create
"ubuntu1010-UEC-localuser-image.tar.gz" ?


> 2- publish this image:
> uec-publish-tarball ubuntu1010-UEC-localuser-image.tar.gz ubuntu x86_64
> 3- Now we run the image using euca-run-instance and it boots up fine, we are
> also able to login using the private key and able to install software.
>
> 4- When we want to save the instace into an image (rebundle), following sequence
> is performed
>
>     -   rm -r /etc/udev/rules.d/70-*

Just curious, where did you see that you should do this? If possible, I'd
like to address the source of the issue, so that people don't have to do
such arcane things.  http://pad.lv/719418 is one such bug that has been
fixed.

>     -   euca-bundle-vol -c ${EC2_CERT} -k ${EC2_PRIVATE_KEY} -u ${EC2_USER_ID}
> --ec2cert ${NOVA_CERT} --no-inherit --kernel aki-00000XXX -d /mnt -r x86_64 -p
> ubuntu-1010-rebundle -s 2048 -e /var/lib/dhcp3

This seems like a good opportunity for simplification.  Anyone interested
in contributing a nice wrapper to http://launchpad.net/cloud-utils ?

Personally, I think that "bundle-vol" should have stopped after creating
the .img, and then have the user use euca-bundle-img and
euca-upload-bundle, then it would not need credentials itself.  And then,
you could use the (more sane, imho):
   uec-publish-image amd64 ubuntu-1010-rebundle.img ubuntu

as a wrapper around your next two commands.

(I realize euca2ools copied ec2-api-tools here, but that doesn't mean i
can't complain).

>     -  euca-upload-bundle -b ubuntu -m  /mnt/ubuntu-1010-rebundle.manifest.xml
>
>    -  euca-register ubuntu/ubuntu-1010-rebundle.manifest.xml
>
> 5- Now we fire up this image once it has been successfully uploaded; we get
> stuck after plymouth-splash. Comapred to the original image which started
> cloud-init after this point. The image is replying to pings but refusing ssh. It
> can be some odd ssh key issue?

If i had to guess, based on nothing else, I suspect that there is an entry
in /etc/fstab for a device that is not present.  mountall will sit and
wait until the proverbial cows come home for it to appear.  The solution
is to put 'nobootwait' in the options field for entries in /etc/fstab that
may not be there on boot.

> [    0.384064] md: autorun ...
> [    0.384435] md: ... autorun DONE.
> [    0.386063] EXT3-fs: barriers not enabled
> [    0.386946] kjournald starting.  Commit interval 5 seconds
> [    0.387731] EXT3-fs (vda): mounted filesystem with ordered data mode
> [    0.388644] VFS: Mounted root (ext3 filesystem) readonly on device 252:0.
> [    0.389738] devtmpfs: mounted
> [    0.390218] Freeing unused kernel memory: 828k freed
> [    0.391214] Write protecting the kernel read-only data: 10240k
> [    0.392153] Freeing unused kernel memory: 308k freed
> [    0.393078] Freeing unused kernel memory: 1612k freed
> mountall: Disconnected from Plymouth
> init: plymouth main process (48) killed by SEGV signal
> init: plymouth-splash main process (323) terminated with status 2
>
>
> Any help is greatly appreciated. I am willing to provide you with other logs if
> requested.

An /etc/fstab from your re-bundled image would confirm or refute my
assertion.