On Mon, Sep 11, 2023 at 11:54 AM Eelco Chaudron <[email protected]> wrote: > On 11 Sep 2023, at 11:36, David Marchand wrote: > > > On Fri, Sep 8, 2023 at 4:28 PM Eelco Chaudron <[email protected]> wrote: > >> On 6 Sep 2023, at 8:20, David Marchand wrote: > >> > >>> Integrate system-traffic.at tests as part of check-dpdk. > >>> > >>> Some tests that can't work with the userspace datapath are skipped by > >>> overriding some OVS_CHECK_* macros. > >>> > >>> ADD_VETH is implemented using the net/af_xdp DPDK driver. > >>> > >>> Signed-off-by: David Marchand <[email protected]> > >> > >> As discussed offline, this patch looks good, however I have some sync > >> issue between testpmd and OVS-DPDK which is causing the following log > >> message: > >> > >> 2023-09-08T13:15:18.160Z|00163|dpdk|INFO|VHOST_CONFIG: > >> (/root/Documents/reviews/ovs_david_dpdk_tests/ovs_github/tests/system-dpdk-testsuite.dir/005/dpdkvhostclient0) > >> vhost peer closed > >> 2023-09-08T13:15:18.160Z|00164|netdev_dpdk|INFO|vHost Device > >> '/root/Documents/reviews/ovs_david_dpdk_tests/ovs_github/tests/system-dpdk-testsuite.dir/005/dpdkvhostclient0' > >> connection has been destroyed > >> 2023-09-08T13:15:18.160Z|00165|dpdk|INFO|VHOST_CONFIG: > >> (/root/Documents/reviews/ovs_david_dpdk_tests/ovs_github/tests/system-dpdk-testsuite.dir/005/dpdkvhostclient0) > >> vhost-user client: socket created, fd: 24 > >> 2023-09-08T13:15:18.160Z|00166|dpdk|WARN|VHOST_CONFIG: > >> (/root/Documents/reviews/ovs_david_dpdk_tests/ovs_github/tests/system-dpdk-testsuite.dir/005/dpdkvhostclient0) > >> failed to connect: Connection refused > >> 2023-09-08T13:15:18.160Z|00085|dpif_netdev|INFO|Performing pmd to rx queue > >> assignment using cycles algorithm. > >> 2023-09-08T13:15:18.160Z|00167|dpdk|INFO|VHOST_CONFIG: > >> (/root/Documents/reviews/ovs_david_dpdk_tests/ovs_github/tests/system-dpdk-testsuite.dir/005/dpdkvhostclient0) > >> reconnecting... > >> 2023-09-08T13:15:18.160Z|00086|dpif_netdev|INFO|Core 58 on numa node 0 > >> assigned port 'dpdkvhostuserclient0' rx queue 0 (measured processing > >> cycles 263044). > >> > >> The test is passing, but fails due to the “WARN|VHOST_CONFIG” log message > >> being present. > >> > >> I’ll let you figure out if just adding the message is sufficient or need > >> some synchronisation. Other than that the code looks good to me! > > > > It looks like testpmd is too quick to quit, and OVS tries to reconnect > > on the vhost user port socket once, before the port is deleted from > > the bridge. > > Inverting OVS_DPDK_STOP_TESTPMD() and AT_CHECK([ovs-vsctl del-port > > br10 dpdkvhostuserclient0], [], [stdout], [stderr]) does the job. > > But other tests have the same pattern... the difference is that only > > this test 5 has a ADD_VETH() so something is changing OVS shutdown > > timing with this last patch of mine. > > > > I will be away a few days, I'll have a look again when I am back. > > Thanks David for doing the investigation. I’ll wait for a v7 once you get > back :)
With the vhost-user library, the client reconnects by itself. So it seems more natural to shutdown the client side first, and after this the server side. Even if other unit tests were not affected I would lean to do this in other (vhost-user-client) tests too. Does it make sense to you? -- David Marchand _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
