Hi OVS-team,

We had just met a problem in our customer environments, the ovs-vswitchd log 
appears in everyday like this:
WARN|system@ovs-system: lost packet on port channel 5 of handler 8

OVS version: 
# ovs-vsctl --version
ovs-vsctl (Open vSwitch) 2.9.0
DB Schema 7.15.1

>From the ovs-dpctl show command, all the port channel is Linux bond bridged in 
>OVS as our tenant network in OpenStack product.

>From the ovs-vswitchd log, we searched the OVS code, it seems the netlink 
>socket receive buffer overflowed:
https://github.com/openvswitch/ovs/blob/branch-2.9/lib/dpif-netlink.c#L2595-L2603


We noticed that the value had changed to 1024kB * 1024kB (1MB) in version v1.4:
https://github.com/openvswitch/ovs/blob/branch-1.4/lib/netlink-socket.c#L126

>From the mail list, there is a method using hping3 to generated lots of new 
>flows per second:
https://mail.openvswitch.org/pipermail/ovs-git/2012-March/012015.html

I try it in out test environment, but not see the previous warnning log.
Can someone provides more detail method of using hping3 try to reproduce the 
problem in the ovs-vswitchd log?

Thanks. 





_______________________________________________
discuss mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-discuss

Reply via email to