Re: [dpdk-dev] [PATCH v2 2/2] ci: add test suite run without hugepage

[Bruce Richardson <bruce.richardson@intel.com>]

On Fri, Mar 06, 2020 at 11:16:24AM -0500, Aaron Conole wrote:
> David Marchand <david.marchand@redhat.com> writes:
> 
> > On Fri, Mar 6, 2020 at 4:57 PM Aaron Conole <aconole@redhat.com> wrote:
> >>
> >> Ruifeng Wang <Ruifeng.Wang@arm.com> writes:
> >>
> >> >> -----Original Message-----
> >> >> From: Aaron Conole <aconole@redhat.com>
> >> >> Sent: Thursday, March 5, 2020 22:37
> >> >> To: Ruifeng Wang <Ruifeng.Wang@arm.com>
> >> >> Cc: maicolgabriel@hotmail.com; bruce.richardson@intel.com; dev@dpdk.org;
> >> >> david.marchand@redhat.com; Gavin Hu <Gavin.Hu@arm.com>; Honnappa
> >> >> Nagarahalli <Honnappa.Nagarahalli@arm.com>; juraj.linkes@pantheon.tech;
> >> >> nd <nd@arm.com>
> >> >> Subject: Re: [PATCH v2 2/2] ci: add test suite run without hugepage
> >> >>
> >> >> Ruifeng Wang <Ruifeng.Wang@arm.com> writes:
> >> >>
> >> >> >> -----Original Message-----
> >> >> >> From: Aaron Conole <aconole@redhat.com>
> >> >> >> Sent: Thursday, March 5, 2020 01:31
> >> >> >> To: Ruifeng Wang <Ruifeng.Wang@arm.com>
> >> >> >> Cc: maicolgabriel@hotmail.com; bruce.richardson@intel.com;
> >> >> >> dev@dpdk.org; david.marchand@redhat.com; Gavin Hu
> >> >> <Gavin.Hu@arm.com>;
> >> >> >> Honnappa Nagarahalli <Honnappa.Nagarahalli@arm.com>;
> >> >> >> juraj.linkes@pantheon.tech; nd <nd@arm.com>
> >> >> >> Subject: Re: [PATCH v2 2/2] ci: add test suite run without hugepage
> >> >> >>
> >> >> >> Ruifeng Wang <ruifeng.wang@arm.com> writes:
> >> >> >>
> >> >> >> > This test suite is derived from fast-tests suite. Cases in this
> >> >> >> > suite are run with '--no-huge' flag.
> >> >> >> >
> >> >> >> > The suite aims to cover as many as possible test cases out of the
> >> >> >> > fast-tests suites in the environments without huge pages support,
> >> >> >> > like containers.
> >> >> >> >
> >> >> >> > Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com>
> >> >> >> > Reviewed-by: Gavin Hu <gavin.hu@arm.com>
> >> >> >> > ---
> >> >> >>
> >> >> >> I like this much more.  Few comments.
> >> >> >>
> >> >> >> >  .travis.yml          | 10 +++++--
> >> >> >> >  app/test/meson.build | 71
> >> >> >> >  ++++++++++++++++++++++++++++++++++++++++++++
> >> >> >>
> >> >> >> You should update doc/guides/prog_guide/meson_ut.rst to include some
> >> >> >> detail about the new tests suite.
> >> >> >>
> >> >> > Thanks. Will update document in next version.
> >> >> >
> >> >> >> >  2 files changed, 79 insertions(+), 2 deletions(-)
> >> >> >> >
> >> >> >> > diff --git a/.travis.yml b/.travis.yml index b64a81bd0..eed1d96db
> >> >> >> > 100644
> >> >> >> > --- a/.travis.yml
> >> >> >> > +++ b/.travis.yml
> >> >> >> > @@ -40,7 +40,7 @@ jobs:
> >> >> >> >    - env: DEF_LIB="static"
> >> >> >> >      arch: amd64
> >> >> >> >      compiler: gcc
> >> >> >> > -  - env: DEF_LIB="shared" RUN_TESTS=1
> >> >> >> > +  - env: DEF_LIB="shared" RUN_TESTS=1 TEST_SUITES="fast-tests
> >> >> >> nohuge-tests"
> >> >> >> >      arch: amd64
> >> >> >> >      compiler: gcc
> >> >> >> >    - env: DEF_LIB="shared" BUILD_DOCS=1 @@ -63,7 +63,7 @@ jobs:
> >> >> >> >    - env: DEF_LIB="static"
> >> >> >> >      arch: amd64
> >> >> >> >      compiler: clang
> >> >> >> > -  - env: DEF_LIB="shared" RUN_TESTS=1
> >> >> >> > +  - env: DEF_LIB="shared" RUN_TESTS=1 TEST_SUITES="fast-tests
> >> >> >> nohuge-tests"
> >> >> >> >      arch: amd64
> >> >> >> >      compiler: clang
> >> >> >> >    - env: DEF_LIB="shared" BUILD_DOCS=1 @@ -101,6 +101,9 @@ jobs:
> >> >> >> >    - env: DEF_LIB="static"
> >> >> >> >      arch: arm64
> >> >> >> >      compiler: gcc
> >> >> >> > +  - env: DEF_LIB="shared" RUN_TESTS=1 TEST_SUITES="nohuge-tests"
> >> >> >> > +    arch: arm64
> >> >> >> > +    compiler: gcc
> >> >> >> >    - env: DEF_LIB="shared" BUILD_DOCS=1
> >> >> >> >      arch: arm64
> >> >> >> >      compiler: gcc
> >> >> >> > @@ -124,3 +127,6 @@ jobs:
> >> >> >> >    - env: DEF_LIB="shared"
> >> >> >> >      arch: arm64
> >> >> >> >      compiler: clang
> >> >> >> > +  - env: DEF_LIB="shared" RUN_TESTS=1 TEST_SUITES="nohuge-tests"
> >> >> >> > +    arch: arm64
> >> >> >> > +    compiler: clang
> >> >> >> > diff --git a/app/test/meson.build b/app/test/meson.build index
> >> >> >> > 0a2ce710f..162a1a76f 100644
> >> >> >> > --- a/app/test/meson.build
> >> >> >> > +++ b/app/test/meson.build
> >> >> >> > @@ -237,6 +237,60 @@ fast_test_names = [
> >> >> >> >          'thash_autotest',
> >> >> >> >  ]
> >> >> >>
> >> >> >> Shouldn't we also trim the list of fast-tests?  Otherwise, these
> >> >> >> tests will run twice.
> >> >> >>
> >> >> > I think you mean to have exclusive lists for fast-tests and nohuge-tests.
> >> >>
> >> >> That's what I was thinking.
> >> >>
> >> >> > Overlapped cases will run twice if both test suites are opted in.
> >> >> > But the two runs are not the same, one runs with hugepage and the
> >> >> > other runs in no-huge mode.
> >> >>
> >> >> Is it really so different between huge and no-huge?  Most of the libraries
> >> >> won't care - they call the rte_**alloc functions, and it returns blocks of
> >> >> memory.  Maybe I am simplifying it too much.
> >> >>
> >> >> > If fast-tests list is splited,  we will need to always run multiple
> >> >> > suites to cover all fast tests.
> >> >> > We can keep x86 to run only fast-tests suite to avoid extra test runs
> >> >> > if they are not necessary. Thoughts?
> >> >>
> >> >> I guess since most DPDK usage will be with hugepages, we should prefer to
> >> >> test with it.  I don't care too much about the color of this particular shed.  If
> >> >> you want to do it that way, it's okay by me - it gives us the coverage, and
> >> >> doesn't duplicate tests between those environments.
> >> >>
> >> >> BUT it means when we add a new test to the suite, we need to remember to
> >> >> add it in two places - fast_test and nohuge_test.  That almost guarantees we
> >> >> will miss tests (because we accidentally don't add it to nohuge).  Maybe
> >> >> there's another way, like keep a list of all the tests and some information on
> >> >> whether the test needs hugepages to run.  Then if there are no hugepages
> >> >> available, we can write that we SKIP the tests that don't support huge pages.
> >> >> In that way, we don't need two different lists - and if there are hugepages,
> >> >> we will run all the test cases.
> >> >> WDYT?
> >> >>
> >> > Yes. Agree with you that having duplicate tests in suites is error prone.
> >>
> >> Cool!
> >>
> >> > IIUC, cases in a suite is determined at build time, as well as command options to run cases.
> >> > This implies hugepage availability needs to be detected at build time if we want to run only
> >> > suitable cases in suite in an environment. It could be something we don't want.
> >> >
> >> > I'll trim fast-tests in next version to remove duplicated cases.
> >>
> >> I think it might be better to make the array something like (just a
> >> psuedo-code example):
> >>
> >>    # psuedo-code to check for hugepages
> >>    has_hugepages = check_for_hugepages()
> >>
> >>    ...
> >>
> >>    fast_test_names = [
> >>      ['acl_autotest', true],
> >>      ['alarm_autotest', true],
> >>      ['atomic_autotest', true],
> >>      ...
> >>
> >> Then in the code:
> >>
> >>    foreach arg : fast_test_names
> >>       ....
> >>       if not arg[1]
> >>          test(arg[0], ...)
> >>       if has_hugepages and arg[1]
> >>          test(arg[0], )
> >>
> >> Does it make sense?  Do you see a problem?
> >
> > I just hope meson won't create a dependency on hugepage availability...
> > Thinking about stuff like:
> > https://git.dpdk.org/dpdk/commit/?id=599d67b6a4bf96096352cc5fbc8fc28e54a1ca62
> 
> Yes, I agree.  We probably need to do some kind of similar abstraction.
>
Hopefully not, since this is for run/test targets rather than for rebuilds.
Try the simple approach first, and only add abstractions if it causes
problems, I think.