Re: [Qemu-devel] [RFC v3 27/27] tests: qmp-test: add oob test

[Peter Xu <peterx@redhat.com>]

On Wed, Nov 15, 2017 at 10:21:16AM +0000, Stefan Hajnoczi wrote:
> On Mon, Nov 06, 2017 at 05:46:43PM +0800, Peter Xu wrote:
> > +    /*
> > +     * Try a time-consuming command, following by a OOB command, make
> > +     * sure we get OOB command before the time-consuming one (which is
> > +     * run in the parser).
> > +     *
> > +     * When writting up this test script, the only command that
> > +     * support OOB is migrate-incoming.  It's not the best command to
> > +     * test OOB but we don't really have a choice here.  We will check
> > +     * arriving order but not command errors, which does not really
> > +     * matter to us.
> > +     */
> > +    qmp_async("{ 'execute': 'dump-guest-memory',"
> > +              "  'arguments': { 'paging': true, "
> > +              "                 'protocol': 'file:/dev/null' }, "
> > +              "  'id': 'time-consuming-cmd'}");
> > +    qmp_async("{ 'execute': 'migrate-incoming', "
> > +              "  'control': { 'run-oob': true }, "
> > +              "  'id': 'oob-cmd' }");
> > +
> > +    /* Ignore all events.  Wait for 2 acks */
> > +    while (acks < 2) {
> > +        resp = qmp_receive();
> > +        if (qdict_haskey(resp, "event")) {
> > +            /* Skip possible events */
> > +            continue;
> > +        }
> > +        cmd_id = qdict_get_str(resp, "id");
> > +        if (acks == 0) {
> > +            /* Need to receive OOB response first */
> > +            g_assert_cmpstr(cmd_id, ==, "oob-cmd");
> > +        } else if (acks == 1) {
> > +            g_assert_cmpstr(cmd_id, ==, "time-consuming-cmd");
> > +        }
> > +        acks++;
> > +    }
> 
> This test is non-deterministic.  The dump-guest-memory command could
> complete first on a fast machine.
> 
> On a slow machine this test might take a long time...
> 
> Please introduce a test command that is deterministic.  For example,
> when 'x-oob-test' is invoked without 'run-oob': true it waits until
> invoked again, this time with 'run-oob': true.

Yes this sounds good.

> 
> We have similar interfaces in the block layer for controlling the order
> in which parallel I/O requests are processed.  This allows test cases to
> deterministically take specific code paths.

It's great to know that I can create a command to test it.  That
should be much easier.  Thanks,

-- 
Peter Xu