Re: [Qemu-devel] [RFC] Future goals for autotest and virtualization tests

[Ademar Reis]

On Thu, Mar 08, 2012 at 11:03:54AM -0600, Anthony Liguori wrote:
> On 03/08/2012 10:05 AM, Ademar Reis wrote:
> >On Thu, Mar 08, 2012 at 09:14:02AM -0600, Anthony Liguori wrote:
> >>On 03/08/2012 09:07 AM, Ademar Reis wrote:
> >>>On Thu, Mar 08, 2012 at 08:56:23AM -0600, Anthony Liguori wrote:
> >>>>On 03/08/2012 08:49 AM, Ademar Reis wrote:
> >>>>>On Thu, Mar 08, 2012 at 07:36:11AM -0600, Anthony Liguori wrote:
> >>>>>>On 03/07/2012 10:00 PM, Lucas Meneghel Rodrigues wrote:
> >>>>>>>Virt/qemu tests: Minimal guest images
> >>>>>>>-------------------------------------
> >>>>>>>
> >>>>>>>In order to make development level test possible, we need the tests to 
> >>>>>>>run fast.
> >>>>>>>In order to do that, a set of minimal guest images is being developed 
> >>>>>>>and we
> >>>>>>>have a version for x86_64 ready and functional:
> >>>>>>>
> >>>>>>>https://github.com/autotest/buildroot-autotest
> >>>>>>
> >>>>>>I'm really not a fan of buildroot.  Note that in order to ship
> >>>>>>binaries, full source needs to be provided in order to comply with
> >>>>>>the GPL.  The FSF at least states that referring to another website
> >>>>>>for source that's not under your control doesn't satisfy the
> >>>>>>requirements of the GPL.
> >>>>>>
> >>>>>>Just out of curiosity, did you try to use qemu-test?  Is there a
> >>>>>>reason you created something different?
> >>>>>>
> >>>>>>I think it's good that you're thinking about how to make writing
> >>>>>>tests easier, but we have a growing test infrastructure in QEMU and
> >>>>>>that's what I'd prefer people focused on.
> >>>>>>
> >>>>>
> >>>>>You probably remember the long thread we had back in December on
> >>>>>qemu-devel on this topic. Back then our message was "we have a
> >>>>>growing test infrastructure in s/QEMU/autotest/ and that's what
> >>>>>we'd prefer people focused on". :-)
> >>>>>
> >>>>> From Dor:
> >>>>>
> >>>>>(http://lists.nongnu.org/archive/html/qemu-devel/2011-12/msg03024.html)
> >>>>>
> >>>>>"""
> >>>>>If you wish, you can challenge Lucas and Cleber w/ these type of
> >>>>>requirements and we'll all improve as a result.
> >>>>>"""
> >>>>>
> >>>>>Your response was:
> >>>>>
> >>>>>"""
> >>>>>Well consider qemu-test the challenge. It's an existence proof
> >>>>>that we can have a very easy to use framework for testing that
> >>>>>runs extremely fast and is very easy to write tests for.
> >>>>>"""
> >>>>>
> >>>>>http://knowyourmeme.com/memes/challenge-accepted ;-)
> >>>>>
> >>>>>I particularly agreed with basically everything you said on that
> >>>>>discussion regarding test simplification (I had just joined the
> >>>>>team back then). To me, autotest has been focusing on QE-level,
> >>>>>leaving the developer-level test requirements out. Now we're
> >>>>>attacking this new front, and a lot of the requirements are
> >>>>>indeed from that discussion.
> >>>>
> >>>>If you want to talk about this in terms of "requirements", my
> >>>>requirement is for "developer-level" tests to live in qemu.git and
> >>>>be integrated into make check.
> >>>>
> >>>>Just as we've been discussing and working on since the previous set of 
> >>>>discussions.
> >>>>
> >>>>>By simplifying the design and bringing barriers down, we hope to
> >>>>>reach a broader audience and help developers write and maintain
> >>>>>tests, benefiting from all the instrumentation that autotest
> >>>>>brings. It's not going to be just about qemu (check the new test
> >>>>>examples).
> >>>>>
> >>>>>We have a team fully dedicated to autotest and it's used not only
> >>>>>by Qemu but also libvirt, Google, Xen, Fedora, Twitter, etc, etc
> >>>>>(these all have code contributions in autotest)
> >>>>>
> >>>>>That said, the current qemu-tests will probably be easily
> >>>>>integrated into (the new) autotest and we hope that, given enough
> >>>>>time, autotest will be good enough to relieve qemu from the
> >>>>>framework maintenance and code duplication with other projects.
> >>>>
> >>>>autotest should not be the focal point for integration.  qemu.git should 
> >>>>be.
> >>>>
> >>>>I'd be perfectly happy to review patches submitting the test
> >>>>infrastructure from kvm-autotest into qemu.git (provided it didn't
> >>>>have unreasonable external dependencies and fit into QEMU).
> >>>>
> >>>>Developer-level tests need to live where the developers live.  The
> >>>>developers live in qemu.git.  See my other response on this thread
> >>>>for the explanation of why this is so important.
> >>>>
> >>>
> >>>Excelent, we're in the same page then. This was my number 1
> >>>requirement when I was discussing the changes with Lucas and
> >>>Cleber. For convenience, I'll repeat here what I wrote in a
> >>>previous e-mail (no qemu-devel archive available yet to use as a
> >>>reference).
> >>>
> >>>In summary, autotest is (or is going to be) a framework that
> >>>provides:
> >>>
> >>>  - A test runner, with grid/cluster support and advanced
> >>>    instrumentation
> >>>  - A devel library and set of utilities for test writers
> >>>  - A set of pre-built images (JeOS – Just Enough OS) for
> >>>    test writers
> >>
> >>I don't think autotest is the right place for this to live.  We need
> >>this directly in qemu.git otherwise we're severely limited in what
> >>tests we can write.
> >>
> >>I guess that doesn't preclude autotest having its own JeOS
> >>mechanism, but we clearly need one in qemu.git.
> >>
> >>>
> >>>(attached is a picture showing what we want to achieve)
> >>>
> >>>If a project has an internal library or set of utilities that can
> >>>be of general use, they can be submitted to autotest.git for
> >>>inclusion, thus reaching a broader audience.
> >>>
> >>>A short summary of the plans:
> >>>
> >>>  - Tests can live anywhere and each devel team implements and
> >>>    maintains their own set of tests
> >>
> >>Let me change this to:
> >>
> >>- Autotest will learn how to harness the tests that each development
> >>team creates in their respective git repository.
> >
> >Yep, given this simple requirement: "tests return 0 or an error
> >code". The output from the test runner will be a simple PASS/FAIL
> >and stdout/stderr will be properly collected by the test runner
> >and made available in a standard location. If using TAP, even
> >better.
> 
> We use gtest, so this will not be the case.  I think it would be
> best for autotest to learn how to interact with the gtest protocol.

I don't see a problem with this. It's very minor IMO, we can
probably speak gtest as well, but in the worst case, all logs
will be collected and are human-readable anyway.

> >
> >This will allow trivial tests. As the test complexity grows, then
> >developers may start using parts of the autotest library or
> >utilities, thus requiring it installed (a trivial bootstrap
> >procedure is also a requirement, see the e-mail from Lucas).
> 
> What is the value of the autotest library to a test writer other
> than being part of the "autotest framework"?

Well, that's the whole idea: there are tons of "goodies" in the
autotest library and in the utilities. You can either replicate
the code and maintain everything inside qemu.git or use what
autotest provides not just to QEMU but other projects as well.

For example, the autotest library provides ways to talk to
guests, check if they're alive, interact via ssh, open serial
channels, start/stop/install/kill/migrate VMs, record video, etc.
Why would you want to duplicate and maintain all this
functionality inside QEMU when this could be provided for free by
autotest, and shared with other projects?

Maybe we can split things a bit more and consider the creation of
a "libvirttest" (or whatever name makes sense) that would
implement all the virt testing infrastructure to be shared among
the interested parties:

 - QEMU
 - libvirt
 - spice
 - ...
 - QE teams (the ones who write complex and comprehensive tests
   in their own repo and will help a lot with the maintenance
   of such library)

But such code already exists and there are active contributors to
it in autotest.git. Our plan is to considerably improve it, but
if something is wrong with our plans, that's the moment to
improve things up by raising requirements.

> 
> >>This is where you start to lose me.  If all you're saying is, "QEMU
> >>can continue to use gtest to build out it's test infrastructure and
> >>autotest will learn how to use it", then we're in violent agreement.
> >
> >Good :-)
> >
> >Inside qemu.git, you can add as many instrumentation and helper
> >libraries as you want. As a developer it'll be up to you to setup
> >your environment to run the tests, and autotest will be just yet
> >another dependency.
> 
> I think there's an implicit assumption here that the tests in
> qemu.git will have a dependency on autotest.  I'd like to understand
> why this dependency is necessary.
> 
> Normally, autotest executes third party tests, what makes QEMU special here?

We're just offering what we already have, thanks to all the
resources invested in the "kvm-" part of "kvm-autotest" through
years. Again, we could split that part out of autotest, but in
the end it's just a naming convention.

A different question would be: what's the point of QEMU writting
code to handle interaction with VMs if such code can be shared
among different projects? Why rewrite everything if there's a
team allocated to this kind of task already, with a substancial
codebase and expertise? Why not raise requirements instead?

> 
> >QE will probably be interested in setting up a bot with several
> >tests from differente repositories, but that's their problem, not
> >yours.
> 
> Right, and this is what autotest is very good at :-)

Yep, but a lot of code can be shared and we can have a common
API, there's no need to duplicate efforts.

I clearly see difference motivations for tests written by QE and
developers, but I don't see why they can't share a common library
and why they can't exchange tests, expertise and contribute with
each other.

For example, you could get patches refactoring or fixing QEMU
tests submitted by members of a QE team. And tests could
move between the QE and QEMU repositories.

> 
> >BTW, please note that we're not trying to cover unit-tests in
> >autotest. At least not by the definition of unit-tests that I
> >know, which are code-level tests without running the application
> >itself (unit-tests test functions/methods only, not the
> >application as a whole).
> 
> There is never a clear line between unit-tests and integration tests
> as long as you're talking about a single project.

By the book, this would be the difference:

unit-test:
    void main()
    {
        my_function();
    }

integration test (or validation test):
    void main()
    {
        exec("my-application");
    }

But that's all semantics, not important for this discussion IMO.

> 
> I expect QEMU to grow tests for anything that involves launching
> QEMU directly.  Where I would not see QEMU growing tests for is
> things like launching QEMU through libvirt.

Agree. For QEMU developers, libvirt should not be on the way, the
interaction should be minimal or non-existent.

That's an area which will require some work in libautotest,
because due to previous QE requirements, it now invokes libvirt
methods instead of QEMU directly for a lot of stuff. But it can
be done and is part of the plan.

<snip>

-- 
Ademar de Souza Reis Jr.
Red Hat

^[:wq!