Xen devel - Mar 2006 - RE: Default Xen Kernel Builds (was Re: Modules in domU kernel config?)

> > These days, most of the testing is on -xen kernels rather 
> than -xen0/U.
> The default xen-unstable builds -xen0 and -xenU kernels, so 
> as far as I know, no one around these parts is testing -xen 
> kernels.  If you believe that -xen0 / -xenU are deprecated, 
> it would be good to change the build defaults.
The binary tar balls and RPMs that are built nightly (and for each
release) have all been based around the -xen kernel (for over a year
now). 

I''ve suggested changing the defult when building from source to just
build -xen before, but a number of developers complained loudly about
the build times going up masively and the inconvenience of needing to
install modules etc. Every developer would end up with a local patch to
change the default etc.

Is it time to revist this? 
NB: we got caught out by this with the 3.0.0 release whereby a config
option enabled in -xen and not in -xen0 caused the "insufficient tx
headroom" bug that effected some configurations and hastened the relase
of 3.0.0-1

Ian

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On Tue, Mar 14, 2006 at 02:57:32PM -0000, Ian Pratt
wrote:
> > > These days, most of the testing is on -xen kernels rather 
> > than -xen0/U.
> > The default xen-unstable builds -xen0 and -xenU kernels, so 
> > as far as I know, no one around these parts is testing -xen 
> > kernels.  If you believe that -xen0 / -xenU are deprecated, 
> > it would be good to change the build defaults.
> 
> The binary tar balls and RPMs that are built nightly (and for each
> release) have all been based around the -xen kernel (for over a year
> now). 
While that may be true, no one "round these parts" is using those
binary
builds or rpms for tests as far as I know.  We''re all doing
test/development
on xen-unstable out of hg.
 
> I''ve suggested changing the defult when building from source to
just
> build -xen before, but a number of developers complained loudly about
> the build times going up masively and the inconvenience of needing to
> install modules etc. Every developer would end up with a local patch to
> change the default etc.
That seems to just create a very bad model, where developers are building
and testing code on kernels that never get used by users.  All the upfront
testing may be for naught.  Developers and users should be using the same
kernel config if possible, especially folks playing with xen-unstable and
reporting bugs.

Last night, when Anthony referenced this email on #xen, it came as a big
shock to most people there.
> Is it time to revist this? 
I would think so.  I''d at least suggest it should change.  Let those
who
really need the smaller build time to go through extra effort to patch their
local tree.  But make the default xen-unstable be the same thing that gets
turned into binaries.
> NB: we got caught out by this with the 3.0.0 release whereby a config
> option enabled in -xen and not in -xen0 caused the "insufficient tx
> headroom" bug that effected some configurations and hastened the
relase
> of 3.0.0-1
	-Sean

-- 
Sean Dague
IBM Linux Technology Center                     email: japh@us.ibm.com
Open Hypervisor Team                           alt: sldague@us.ibm.com


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel