On 27.02.2016 01:53, Joe Stringer wrote: > On 26 February 2016 at 03:57, Ilya Maximets <i.maxim...@samsung.com> wrote: >> Fixes failure of test due to not yet fully destroyed ports from >> previous test if ovs-vswitchd was killed. >> >> Example: >> >> netdev_linux|WARN|br0: creating tap device failed: Device or resource busy >> bridge|WARN|could not open network device br0 (Device or resource busy) >> >> Signed-off-by: Ilya Maximets <i.maxim...@samsung.com> > > If I'm not mistaken, this will add over 20 minutes of sleeping to the > standard testsuite. Well, less if you're running in parallel, but > still not great.
Sleep will be executed only for failed or skipped after OVS_TRAFFIC_VSWITCHD_START tests, because 'kill' will fail if graceful shutdown already executed. So, this 20 minutes of sleeping may be added only if *all* tests aren't 'OK'. > Is there a way that we could detect this "device or resource busy" > error during perhaps OVS_TRAFFIC_VSWITCHD_START, for example by > querying the kernel for br0 device status first, and attempt that a > few times (maybe using OVS_WAIT_UNTIL(...)) instead? This may have sense. But issue concerns not only br0 but also all other ports. At least ovs-netdev internal port: netdev_linux|WARN|ovs-netdev: creating tap device failed: Device or resource busy dpif|WARN|datapath ovs-netdev already exists but cannot be opened: No such device ofproto_dpif|ERR|failed to open datapath of type netdev: No such device I'll prepare patch with OVS_WAIT_WHILE() for br0 and ovs-netdev because they are created inside OVS_TRAFFIC_VSWITCHD_START. Best regards, Ilya Maximets. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
