[illumos-Discuss] Test/build setup?

[Erik Trimble]

On Thu, 2010-08-19 at 12:55 -0700, Dmitry Yusupov wrote:
> 
> On 08/19/2010 11:09 AM, Garrett D'Amore wrote:
> > On Thu, 2010-08-19 at 11:02 -0700, Dmitry Yusupov wrote:
> >
> >>> Nexenta has a system called NABAT, but its highly specific to
> >>> debian/apt, so we need something more generic.
> >>
> >> Upcoming NexentaCore 3.9 is going to be Illumos based and *is* going to
> >> be nightly tested/released by NABAT. While I agree, we need a generic
> >> framework I would suspect that we only need to focus on testing Illumos
> >> components and not its permutations or compositions. Good example would
> >> be ztest utility..
> >>
> >> For instance, NABAT is also testing NFS/CIFS/iSCSI against tens of
> >> clients such us Windows/Linux/ESX/Xen/Hyper-V/etc to ensure its
> >> interoperability. And at distro level it is easier and Illumos is not a
> >> distro.
> >>
> >
> > So Illumos is the beneficiary of this testing, which is awesome.
> >
> > But it isn't sufficient IMO for a couple of reasons:
> >
> > a) NCP replaces a number of userland components with bits from
> > GNU/Debian.  While this is fine for NCP, the Illumos community still
> > needs some basic verification of the ones that everyone else in the
> > Illumos community uses.
> >
> > b) NABAT is owned by Nexenta.  While as a Nexenta employee I have no
> > concerns about Nexenta's interests, I think its important to try to make
> > sure that we use resources that can't be taken away later by Nexenta.
> > Not that I fear this will happen, but by eliminating even the
> > possibility we can prevent any such concerns from arising elsewhere.
> >
> > c) IMO, there is great value in alternative testing.  If everyone always
> > runs the same test suites, and nothing else, then we don't get to find
> > out about bugs that are not covered by that specific test procedure in
> > the system during test. (Instead it bits us when someone is trying to
> > *use* the software, which is infinitely worse than finding the bug
> > during automated testing.)
> >
> > Hopefully this makes sense.
> 
> Make sense! And for (c), I bet Nexenta would be happy to sponsor its 
> integration into NABAT later on.


OK, So I'll get cracking on at least configuring the machines into a
usable test system base, from where we can begin to layer the build/test
system onto.

I'm also likely to go out and get a couple of IBM x3455 machines
(dual-socket, dual-core Opteron 2000), since they're soooo cheap, and it
gives us something to do virtualization on should the need arise.



I currently use several T1000 machines here at Oracle to do the Hotspot
Java VM builds, and they're noticeably faster than our US III and IV
class boxes, because the build parallel-izes so very well.  I'd hope
that we could do something similar with the 'dmake/smake' stuff, since
huge chunks of the build should independent of each other, particularly
all the userland stuff...  


-- 
Erik Trimble
Java System Support
Mailstop:  usca22-123
Phone:  x17195
Santa Clara, CA
Timezone: US/Pacific (GMT-0800)