Sensible libvirt xml
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Eucalyptus |
New
|
Undecided
|
Dmitrii Zagorodnov |
Bug Description
Ok This is the start of another bug similar to my -vnc option one but editing that config in /etc/libvirt/
So i think the problem is where the libvirt xml is being generated. So i look at the default style of a eucalyptus vm
<domain... bla bl>
...
<type hvm>
<kernel ...>
<initrd ...>
<cmdline> root=/dev/sda1 console=ttyS0
<\type>
...
</domain>
So i have problems with the whole type in the libvirt xml stuff. So really is it nesecary to have a whole kernel and initrd setup. It is a nice way of having much much smaller virtual image bundles and just adding on extra disks for space as needed. But why not just have a simple rootfs with kernel etc..
Or at least have support for an EMI to work without having to specify and kernel and ramdisk because this is causeing serious problems. I need to use things like AIX or Suse or Solaris. I have no idea how to get them onto eucalyptus i usualy just use kvm in libvirt or kvm manualy.
Then i dont see anywhere where i specify the console=ttyS0 and looking at
ps ax | grep kvm
pastebin here might be easier
everything else is fine i guess. But i just think the virtual machine mangement should be cleaned up a bit. And ec2-run-upload image should throw an error that i didnt specify kernel or initrd with this image. But tbh it should support a single image virtual machine because it will save alot of headaches and support all OS's that KVM and Qemu and XEN support.
Then the console=ttyS0 and no-graphic needs to be rid of i really think you need some kind of backup interface to your virtual machine because all my stuff is having major problems with this because i am behind a corporate proxy and if i follow the ubuntu wiki sure i can do export http_proxy but its not really working properly. Ie i think networking isnt starting properly but it should be it did earlier but recently it didn't.
Another thing is why do we have to specify buckets? Shouldn't it just be i want to upload this manifest of files? Then let eucalyptus assign buckets(folders) properly then ec2-delete instance should only need the instance-id the whole EMI or ERI or EKI instead of -bukkit bla....
Because thats a painful process deleting images and ec2-terminate instances doesn't seem to work very well. Or is this all just ec2-tools we cant change? I guess thats probably why.. But the backend libvirt should change its annoying you can just upload an image and run it like normal. The reason i cant rely on networking is that with a SUSE vm when a new macaddr is added to the system its seen as a new device to yast and then it wont be auto-configured because SUSE sucks but i have to use it for work .purposes.
As bad as it is then you could support windows guest instances too even tho no-one should be using that ;).
Changed in eucalyptus: | |
assignee: | nobody → Dmitrii Zagorodnov (dmitrii) |
This issue is now being tracked upstream at http:// eucalyptus. atlassian. net/browse/ EUCA-2646
Please watch that issue for further updates.