On Sun, Nov 27, 2011 at 1:24 PM, Michael DeHaan
<michael.dehaan(a)gmail.com> wrote:
> The primary advantage is the ability to do everything through
httpd.
> This gives you the ability to bypass the layer 2 restriction of DHCP
> without having to resort to tricks like Cisco's ip-helper. This isn't
> the only advantage - gpxe lets you boot remotely over a LOT of
> different medium (AoE, iscsi, etc.). The newer versions of kvm/libvirt
> have built-in gpxe support for their virtual NIC drivers, so chain
> loading for a KVM guest isn't even required.
Thanks... that makes sense.
Seems like we *would* always be using httpd, though, in our case?. So
I guess, to userland, the advantage is "it's faster and probably can
boot more nodes at once".
Not necessarily. Is there any reason we couldn't extend cobbler to
manage iscsi exports or other methods?
This follows.
I guess what I'm curious about is *IF* it will work on all legacy
hardware, do we still need the old path around as a default? i.e.
maybe normal PXE becomes the weird fallback option?
Or does that not make sense?
Maybe there's not a good reason though, and we suspect gpxe will not
work everywhere? Not sure what is offered in terms of boot menus,
etc. Keeping those available
would be important if we tried to do that.
The issue here is the supported version of syslinux. Newer versions of
pxelinux support httpd natively, however RHEL/CentOS 5 (and 4 of
course) may not have access to those versions. This is a stop-gap for
the next 3-5 years in which we will support those distros. At some
point, we could revert back to a pure pxelinux implementation (or not,
if we do end up supporting stuff like iscsi and AoE).
(I'm also a bit curious why VMware ESX won't boot from
pxelinux.0, but
that's just for my info and unrelated)
It does - the limitation as I noted in the original email is that you
have to copy all of the ISO contents to the tftp directory. The ESXi5
install supports httpd through gpxe only. In order to get that to work
with cobbler sync and the --boot-files option, some ugly hacks had to
be made allowing for globs so that all files would be copied. That
could become a management issue too, since for older distros /tftpboot
is in the root file system and may not have a ton of space available.
>
>> + ks_mirror_name =
string.join(distro.kernel.split('/')[-2:-1],'')
>> 862
>>
>> FYI -- This seems to imply the distro must be mirrored by Cobbler.
>> It seems like the code needs to anticipate this this possibility?
>
> Yes and no - there's no reason the http server in question has to be
> cobbler. Covering that case would be a todo, but I didn't give it
> priority since I believe the number of people who use the
> --available-as option are relatively small. I'm pretty sure that
> option doesn't work with esxi4 either (since at least 6 files need to
> be copied to the tftp directory), so I figured the people who'd use
> this feature right now for esxi5 wouldn't care about that limitation
IIRC, --available-as isan option to import.
I was thinking about the manual "cobbler distro add" case, where you
might not even have a "tree" parameter at all, and the files might
live on NFS somewhere.
I'd probably attempt to make sure the path *looked* like a cobbler
path, and if not, maybe raise an error or something.
Maybe this is just the way I percieved it, but I basically view import
as a shortcut path around distro add/profile add/etc, not the path you
always have to go down.
Anyway, some way to yell at you if you're doing it wrong seemed to be
my point.
if distro.kernel.find("ks_mirror") != -1:
raise Exception("...")
or something like that
--Michael
_______________________________________________
cobbler mailing list
cobbler(a)lists.fedorahosted.org
https://fedorahosted.org/mailman/listinfo/cobbler