Hey,
I had a thought about the builders that i wanted to run by everyone. it seems that the chroot exploded on hammer1 and as a result some files got installed in the real filesystem not in the chroot. so as a way to help things in the future. once we are done with analysis we will need to rebuild hammer1 So what i want to do is install a xen host on hammer1 and actually have hammer1 a xen guest. it would be the only guest on the system but if the same thing happened again it would be a simple task to fix. warren to do this we would need a internal only ip for the xen host.
so what id like to do as far as the xen guest goes. and we should probably do this with all production xen guests. is create a snapshot of the lvm volume once the system is up and running and ready. so if the chroot explodes again we put the snapshot in place start the instance and things are back in a matter of minutes. we would have a pre-configured system ready to roll.
thoughts? opinions? objections?
On 10/27/06, Dennis Gilmore dennis@ausil.us wrote:
Hey,
I had a thought about the builders that i wanted to run by everyone. it seems that the chroot exploded on hammer1 and as a result some files got installed in the real filesystem not in the chroot. so as a way to help things in the future. once we are done with analysis we will need to rebuild hammer1 So what i want to do is install a xen host on hammer1 and actually have hammer1 a xen guest. it would be the only guest on the system but if the same thing happened again it would be a simple task to fix. warren to do this we would need a internal only ip for the xen host.
so what id like to do as far as the xen guest goes. and we should probably do this with all production xen guests. is create a snapshot of the lvm volume once the system is up and running and ready. so if the chroot explodes again we put the snapshot in place start the instance and things are back in a matter of minutes. we would have a pre-configured system ready to roll.
thoughts? opinions? objections?
I'm kind of in the boat that we should just use large files and skip the partition all together, that way its an scp or a cp away from having another guest.
-Mike
On Friday 27 October 2006 09:02, Mike McGrath wrote:
On 10/27/06, Dennis Gilmore dennis@ausil.us wrote:
so what id like to do as far as the xen guest goes. and we should probably do this with all production xen guests. is create a snapshot of the lvm volume once the system is up and running and ready. so if the chroot explodes again we put the snapshot in place start the instance and things are back in a matter of minutes. we would have a pre-configured system ready to roll.
thoughts? opinions? objections?
I'm kind of in the boat that we should just use large files and skip the partition all together, that way its an scp or a cp away from having another guest.
-Mike
That works also. one issue we would face is storing the xen guest images. that will get big fast. But i do really like the idea. we could do something to help us manage bottlenecks also. bring up an app server image when needed and some quick configuration we have a new app server. or proxy or any of our other services.
Dennis Gilmore wrote:
On Friday 27 October 2006 09:02, Mike McGrath wrote:
I'm kind of in the boat that we should just use large files and skip the partition all together, that way its an scp or a cp away from having another guest.
That works also. one issue we would face is storing the xen guest images. that will get big fast. But i do really like the idea. we could do something to help us manage bottlenecks also. bring up an app server image when needed and some quick configuration we have a new app server. or proxy or any of our other services.
One of the downsides of file based images is I/O performance, right? Won't that have an adverse effect on the builders versus LVM?
I do like the idea of using more Xen Guests for some of the flexibility it could give us. Figuring out how to make it work well for our infrastructure is most likely worth the time.
--Jeffrey
On 10/27/06, Jeffrey Tadlock linux@elfshadow.net wrote:
Dennis Gilmore wrote:
On Friday 27 October 2006 09:02, Mike McGrath wrote:
I'm kind of in the boat that we should just use large files and skip the partition all together, that way its an scp or a cp away from having another guest.
That works also. one issue we would face is storing the xen guest images. that will get big fast. But i do really like the idea. we could do something to help us manage bottlenecks also. bring up an app server image when needed and some quick configuration we have a new app server. or proxy or any of our other services.
One of the downsides of file based images is I/O performance, right? Won't that have an adverse effect on the builders versus LVM?
Yes, though I'd like to see some numbers as to what difference we're talking about. The flexability of using a file is very appealing. Especially since our boxes aren't really that overloaded. (well except fpserv ;-)
-Mike
On Friday 27 October 2006 15:07, Jeffrey Tadlock wrote:
One of the downsides of file based images is I/O performance, right? Won't that have an adverse effect on the builders versus LVM?
Yes. I just talked w/ our Xen folks and they recommend using LVM, and doing lvm snapshots, back up the snapshot to an image file, so that in an emergency a guest could be brought up on another machine using the image file.
On 10/27/06, Jesse Keating jkeating@redhat.com wrote:
On Friday 27 October 2006 15:07, Jeffrey Tadlock wrote:
One of the downsides of file based images is I/O performance, right? Won't that have an adverse effect on the builders versus LVM?
Yes. I just talked w/ our Xen folks and they recommend using LVM, and doing lvm snapshots, back up the snapshot to an image file, so that in an emergency a guest could be brought up on another machine using the image file.
Whats the actual performance hit though? I want numbers :D
-Mike
On Oct 27, 2006, at 12:42, Jesse Keating wrote:
On Friday 27 October 2006 15:07, Jeffrey Tadlock wrote:
One of the downsides of file based images is I/O performance, right? Won't that have an adverse effect on the builders versus LVM?
Yes. I just talked w/ our Xen folks and they recommend using LVM, and doing lvm snapshots, back up the snapshot to an image file, so that in an emergency a guest could be brought up on another machine using the image file.
A variation could be having the "system partition" in a relatively static small-ish image file and the build path on an LVM partition...
- ask
On Fri, 27 Oct 2006 08:54:09 -0500, Dennis Gilmore wrote:
so what id like to do as far as the xen guest goes. and we should probably do this with all production xen guests. is create a snapshot of the lvm volume once the system is up and running and ready. so if the chroot explodes again we put the snapshot in place start the instance and things are back in a matter of minutes. we would have a pre-configured system ready to roll.
thoughts?
Sounds like a good idea.
C
infrastructure@lists.fedoraproject.org