On Wed, Feb 27, 2019 at 12:45:58PM +0000, Ken Ajiro wrote: > > Hello, > > When I use ovs-ofctl on OVS 2.10.1, it failed occasionally by error: > > ovs-ofctl: br0: failed to connect to socket (Broken pipe) > > This error is possible to be reproduced with: > > # while :; do /opt/nec/pf/ds/bin/ovs-ofctl --timeout=5 dump-flows br0 > > /tmp/log || break; done; cat /tmp/log > ovs-ofctl: br0: failed to connect to socket (Broken pipe) > > Also I tried with OVS 2.4.1 and could not reproduce this error. > > I think that recently rconn.c was modified on commit > 476d2551abd2871696a64203f78d658ac2d7f32c > and the connection for ofctl was changed it begins on state S_CONNECTING > instead of S_ACTIVE. > So timeout_CONNECTING will be applied for ofctl connection, however > timeout_CONNECTING is 1 > and this is too short (e.g. if transition time of CONNCTING was > 12:00:00.999, it will be timed out > after 1ms). I think that this is cause of occasionally disconnection of > ofctl.
Thanks for the report. It seems to me that the right long-term fix is to use higher resolution (e.g. millisecond resolution) to measure the timeouts. However, that will be a relatively large fix. The following stopgap appears to fix the problem on master. I suspect it could be easily backported as far as necessary. Does it work for you too? diff --git a/lib/rconn.c b/lib/rconn.c index 48ae8c6a72e5..8ca23cbcfe90 100644 --- a/lib/rconn.c +++ b/lib/rconn.c @@ -502,7 +502,7 @@ static unsigned int timeout_CONNECTING(const struct rconn *rc) OVS_REQUIRES(rc->mutex) { - return MAX(1, rc->backoff); + return MAX(2, rc->backoff); } static void _______________________________________________ discuss mailing list disc...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-discuss