Re: [Qemu-devel] [PATCH v4 5/5] net/announce: Expand test for stopping self announce

["Dr. David Alan Gilbert" <dgilbert@redhat.com>]

* Jason Wang (jasowang@redhat.com) wrote:
> 
> On 2019/6/13 下午5:59, Dr. David Alan Gilbert (git) wrote:
> > From: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
> > 
> > Expand self-announce test to check we can stop an announce timer.
> > We set it up to send 300 packets, but after we receive
> > the first one we tell it to stop.
> > 
> > We error if:
> >     a) We receive more than 30 of the packets
> >     b) We're still receiving packets after a lot longer than the
> >        30 seconds should have arrived
> > 
> > Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
> > ---
> >   tests/virtio-net-test.c | 57 ++++++++++++++++++++++++++++++++++++++---
> >   1 file changed, 54 insertions(+), 3 deletions(-)
> > 
> > diff --git a/tests/virtio-net-test.c b/tests/virtio-net-test.c
> > index 663cf7ea7e..3b49b431dc 100644
> > --- a/tests/virtio-net-test.c
> > +++ b/tests/virtio-net-test.c
> > @@ -184,21 +184,72 @@ static void announce_self(void *obj, void *data, QGuestAllocator *t_alloc)
> >       QDict *rsp;
> >       int ret;
> >       uint16_t *proto = (uint16_t *)&buffer[12];
> > +    size_t total_received = 0;
> > +    uint64_t start, now, last_rxt, deadline;
> > +    /* Send a set of packets over a few second period */
> >       rsp = qmp("{ 'execute' : 'announce-self', "
> >                     " 'arguments': {"
> > -                      " 'initial': 50, 'max': 550,"
> > -                      " 'rounds': 10, 'step': 50 } }");
> > +                      " 'initial': 20, 'max': 100,"
> > +                      " 'rounds': 300, 'step': 10, 'id': 'bob' } }");
> >       assert(!qdict_haskey(rsp, "error"));
> >       qobject_unref(rsp);
> > -    /* Catch the packet and make sure it's a RARP */
> > +    /* Catch the first packet and make sure it's a RARP */
> >       ret = qemu_recv(sv[0], &len, sizeof(len), 0);
> >       g_assert_cmpint(ret, ==,  sizeof(len));
> >       len = ntohl(len);
> >       ret = qemu_recv(sv[0], buffer, len, 0);
> >       g_assert_cmpint(*proto, ==, htons(ETH_P_RARP));
> > +
> > +    /*
> > +     * Stop the announcment by settings rounds to 0 on the
> > +     * existing timer.
> > +     */
> > +    rsp = qmp("{ 'execute' : 'announce-self', "
> > +                  " 'arguments': {"
> > +                      " 'initial': 20, 'max': 100,"
> > +                      " 'rounds': 0, 'step': 10, 'id': 'bob' } }");
> > +    assert(!qdict_haskey(rsp, "error"));
> > +    qobject_unref(rsp);
> > +
> > +    /* Now make sure the packets stop */
> > +
> > +    /* Times are in us */
> > +    start = g_get_monotonic_time();
> > +    /* 30 packets, max gap 100ms, * 2 for wiggle */
> > +    deadline = start + 1000 * (100 * 30 * 2);
> > +    last_rxt = start;
> > +
> > +    do {
> 
> 
> while (ture) looks better here.

OK, changed.

> 
> > +        int saved_err;
> > +        ret = qemu_recv(sv[0], buffer, 60, MSG_DONTWAIT);
> > +        saved_err = errno;
> > +        now = g_get_monotonic_time();
> > +        g_assert_cmpint(now, <, deadline);
> 
> The maximum gap allowed is 1000 * 100 * 4, and we allow at most 30 packets
> that's 30 * 1000 * 100 * 4 which is 1200000.
> 
> But the deadline is 1000 * 100 * 30 * 2 which is 6000000.
> 
> Does this mean deadline is conflict with the assumption above?

I've changed deadline to match (i.e. * 4) - but it's only a worst-case
that we shouldn't hit anyway; I'm expecting it to actually stop after
1 or 2 packets worst case, the *4 in the maximum gap is there just to
deal with a busy system that takes a bit longer.


Dave

> Thanks
> 
> 
> > +
> > +        if (ret >= 0) {
> > +            if (ret) {
> > +                last_rxt = now;
> > +            }
> > +            total_received += ret;
> > +
> > +            /* Check it's not spewing loads */
> > +            g_assert_cmpint(total_received, <, 60 * 30 * 2);
> > +        } else {
> > +            g_assert_cmpint(saved_err, ==, EAGAIN);
> > +
> > +            /* 400ms, i.e. 4 worst case gaps */
> > +            if ((now - last_rxt) > (1000 * 100 * 4)) {
> > +                /* Nothings arrived for a while - must have stopped */
> > +                break;
> > +            };
> > +
> > +            /* 100ms */
> > +            g_usleep(1000 * 100);
> > +        }
> > +    } while (true);
> >   }
> >   static void virtio_net_test_cleanup(void *sockets)
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK