On Fri, Jun 12, 2009 at 08:05:00AM -0600, Jerry Jelinek wrote:
>> hey jerry,
>>
>> ----------
>> src/brand/image_install
>>
>> - does '-d -' actually work?  seems like it wouldn't since pkgcreatezone
>>   just mounted a zfs dataset on the zoneroot before invoking this script.
>
> It does if you first set up the dataset correctly but this is definitely
> an area that needs to be improved.  The idea is you could zfs recv the
> dataset but then you'd have to manually set up the properties to make it
> usable.  I'll file a bug to track that.
>

could you elaborate on what the final interface usage here will look
like?  (ie how will the user have to "setup the dataset correctly"?)

if using "-d -" is going to require users to manipulate zfs zbe
properties then i think that supporting this is a really bad and error
prone idea.  imho it would be better to have users zfs send/recv their
system to some randon location on the target machine and then use -d to
have the zones framework copy the zone contents from that random
location to the zone root.  that way we know that zone zbe's are always
created by the zones framework, and hence are more likely to be in a
sane state.  (rather than having them in odd states because users were
tweaking misc zbe properties till they got things working.)

>> - would it be possible to generate $ipdcpiofile and $ipdpaxfile from a
>>   single list of directories?  (that way no one would miss the other
>>   when doing bugfixes/updates.)
>
> Unfortunately these use different formats.  This is based on the
> same code we have today in nv.
>
>> - rather than manually setting up zone mounts via get_fs_info() and
>>   mnt_fs(), wouldn't it be better to do mount -f?  or ready -f?  and
>>   have the zones framework take care of this all?
>
> This is based on the nv code.  Mount -f won't work since that mounts
> under /a.  We'd first have to fix the installer to work in the
> scratch zone.  If we want to do that I'd want to fix up the original nv
> p2v installer first.  There is no ready -f option.
>

so perhaps we should introduce a "ready -f" state?  seems like it would
allow us to pitch a lot of this complex shell code in favor of well
tested zoneadm code.

ed
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to