Re: [Qemu-devel] [PATCH v2 4/5] test: Postcopy

[Dr. David Alan Gilbert]

* Marcel Apfelbaum (address@hidden) wrote:
> On 05/03/2016 12:22 PM, Dr. David Alan Gilbert wrote:

> > > > +        QDECREF(rsp);
> > > > +        usleep(1000 * 100);
> > > > +    } while (!completed);
> > > > +}
> > > 
> > > It is possible that the migration will not finish (from some reason) ?
> > > Do you expect the test runner to stop the test?
> > 
> > The migration should always complete in postcopy; failure to complete
> > is failure of the test;   although I've not explicitly added any timeouts.
> > 
> 
> OK, so on a failure I run make-check and it never ends ? :)
> For build-bots there is no problem since they kill tests on timeout,
> but we are running it manually.
> However, we can add the test as is and if it 'behaves' is OK.

Most of the failure cases will cause an assert/crash - a hang should be
unlikely.  I'm assuming a bunch of tests can in principal hang if they go wrong;
if there's some qtest structure for timeouts I can add it.

> > <snip>
> > 
> > > > +int main(int argc, char **argv)
> > > > +{
> > > > +    char template[] = "/tmp/postcopy-test-XXXXXX";
> > > 
> > > I would not explicitly use /tmp/
> > 
> > The ivshmem-test, vhost-user-test and test-qga seem to do it this way;
> > what's your preferred fix?
> 
> You could use the P_tmpdir macro instead of '/tmp', but again,
> if all the tests assume /tmp existence maybe is not an issue.

I don't see any use of P_tmpdir in qemu at all; we do have one use of
g_get_tmp_dir (in util/memfd.c).

> > > > +    int ret;
> > > > +
> > > > +    g_test_init(&argc, &argv, NULL);
> > > > +
> > > > +    if (!ufd_version_check()) {
> > > > +        return 0;
> > > > +    }
> > > > +
> > > > +    tmpfs = mkdtemp(template);
> > > > +    if (!tmpfs) {
> > > > +        g_test_message("mkdtemp on path (%s): %s\n", template, 
> > > > strerror(errno));
> > > > +    }
> > > > +    g_assert(tmpfs);
> > > > +
> > > > +    module_call_init(MODULE_INIT_QOM);
> > > > +
> > > > +    qtest_add_func("/postcopy", test_migrate);
> > > > +
> > > 
> > > How much time does this test takes? If is too long, maybe we should not 
> > > run it
> > > automatically as part of make check, but using an environment variable.
> > 
> > 4 seconds on my laptop; it seems reasonable.
> > 
> 
> make-check takes about 1 and a half minute for x86_64 configuration, 4 seconds
> more seems reasonable indeed.

Dave

> Thanks,
> Marcel
> 
> 
> > Dave
> > --
> > Dr. David Alan Gilbert / address@hidden / Manchester, UK
> > 
> 
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK