Re: [PATCH rdma-core 03/14] build: Add pyverbs-based test to the build

[Leon Romanovsky <leonro@mellanox.com>]

On Mon, Sep 09, 2019 at 11:26:35AM +0000, Leon Romanovsky wrote:
> On Mon, Sep 09, 2019 at 10:39:44AM +0000, Noa Osherovich wrote:
> >
> > On 9/9/2019 1:29 PM, Leon Romanovsky wrote:
> > > On Sun, Sep 01, 2019 at 01:30:56PM +0000, Noa Osherovich wrote:
> > >> On 8/22/2019 7:52 PM, Jason Gunthorpe wrote:
> > >>
> > >>> On Thu, Aug 22, 2019 at 01:18:24PM -0300, Leon Romanovsky wrote:
> > >>>> On Tue, Aug 20, 2019 at 01:00:47PM +0000, Noa Osherovich wrote:
> > >>>>> On 8/19/2019 4:50 PM, Jason Gunthorpe wrote:
> > >>>>>
> > >>>>>> I'd prefer run_tests to be in the tests directory..
> > >>>>>>
> > >>>>>> Jason
> > >>>>> PR was updated
> > >>>> 1.
> > >>>> IMHO, run_tests.py should be placed inside tests directory too and not
> > >>>> only installed into tests/.
> > >>> Yes, this is what I ment. The file should be in tests/ and it should
> > >>> be built into build/bin, and installed into the examples
> > >>>> 2.
> > >>>> Execution of run_tests.py produces a lot of untracked filed
> > >>>> ➜  rdma-core git:(noaos-pr-tests) ✗ git st
> > >>>> On branch noaos-pr-tests
> > >>>> Untracked files:
> > >>>>   (use "git add <file>..." to include in what will be committed)
> > >>>>
> > >>>> 	pyverbs/__init__.pyc
> > >>>> 	pyverbs/pyverbs_error.pyc
> > >>>> 	tests/__init__.pyc
> > >>>> 	tests/base.pyc
> > >>>> 	tests/test_addr.pyc
> > >>>> 	tests/test_cq.pyc
> > >>>> 	tests/test_device.pyc
> > >>>> 	tests/test_mr.pyc
> > >>>> 	tests/test_odp.pyc
> > >>>> 	tests/test_pd.pyc
> > >>>> 	tests/test_qp.pyc
> > >>> *.pyc will have to be added to the .gitignore
> > >>>> 3. run_tests.py lacks of python3 shebang
> > >>> Originally it was not installed, so this was fine, as the build/bin
> > >>> script does all the required setup, however now that it is to be
> > >>> installed it should have the #! - and it should also work without any
> > >>> trouble from it's example location.
> > >>>
> > >>> Jason
> > >> PR was updated.
> > > I tried it now and got very confusing results.
> > >
> > > On my machine there are no ib_devices, and I expected to see ALL tests
> > > marked as skipped, but got two skipped only, is it expected behaviour?
> >
> > Yes. If you recall, our previous unittests worked differently than the new ones; each
> > test would iterate over an array of all available devices and would run on each device.
> > The array can be of length 0. If no failure was found, they're marked as passed.
> > The new tests skip (the reported 's' you saw) in case a combination of device/port/GID
> > index wasn't found.
>
> array length 0 should return "skipped" and not "passed".

I run over IB device over rdma-next and get very exciting splat.

[leonro@server ~]$
/images/leonro/src/rdma-core/build/bin/run_tests.py
.....EFE..................................ss...............
======================================================================
ERROR: test_create_cq_ex (test_cq.CQEXTest)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/images/leonro/src/rdma-core/tests/test_cq.py", line 118, in test_create_cq_ex
      with CQEX(ctx, get_attrs_ex(attr, attr_ex)):
  File "cq.pyx", line 264, in pyverbs.cq.CQEX.__cinit__
pyverbs.pyverbs_error.PyverbsRDMAError: Failed to create extended CQ. Errno: 95, Operation not supported

and many more.

>
> Thanks
>
> >
> > Thanks,
> > Noa
> >
> > >
> > > _  rdma-core git:(noaos-pr-tests) ./build/bin/run_tests.py
> > > ..........................................ss...............
> > > ----------------------------------------------------------------------
> > > Ran 59 tests in 0.004s
> > >
> > > OK (skipped=2)
> > >
> > > Thanks
> > >
> > >> Thanks,
> > >>
> > >> Noa
> > >>