[ovs-dev] [BUG] broad-/multicast & SLB bonding -> FAIL

[Jesse Gross]

On Fri, Jan 25, 2013 at 10:26 AM, Markus Schuster
<ml at markus.schuster.name> wrote:
> On Thursday 24 January 2013 18:49:16 Jesse Gross wrote:
>> Can you try to isolate broadcast/multicast traffic as much as possible
>> and then run:
>> ovs-dpctl dump-flows xapi1
>> while the problem is happening?
>
> Sorry it took me a bit to respond but that command output is just too long to
> be of any use when doing it in the normal production environment so I set up a
> fresh XCP server with a test VM on it, configured it for SLB bonding and put
> it in an isolated VLAN.
>
> Some background information:
> 192.168.0.124: Laptop I used to connect to the XCP server and the VM
> 192.168.0.157: Test VM
> 192.168.0.171: XCP server

Are you configuring bonding directly through the OVS command line or
using Xen tools?  Can you also send the output of ovs-vsctl show?

> Here is the complete output of the command. I ran it multiple times, but the
> output didn't change (if you disregard counters).
>
> --- cut ---
> priority(6),in_port(0),eth(src=00:1a:64:8f:a5:9e,dst=00:16:41:55:fa:75),eth_type(0x0800),ipv4(src=192.168.0.171,dst=192.168.0.124,proto=6,tos=0x10,ttl=64,frag=no),tcp(src=22,dst=44899),
> packets:4, bytes:504, used:0.020s, actions:1
> in_port(4),eth(src=b2:76:b4:94:5d:e1,dst=ff:ff:ff:ff:ff:ff),eth_type(0x0800),ipv4(src=192.168.0.157,dst=192.168.0.255,proto=1,tos=0,ttl=64,frag=no),icmp(type=8,code=0),
> packets:104, bytes:10192, used:0.570s, actions:2,0
> in_port(1),eth(src=b2:76:b4:94:5d:e1,dst=ff:ff:ff:ff:ff:ff),eth_type(0x0800),ipv4(src=192.168.0.157,dst=192.168.0.255,proto=1,tos=0,ttl=64,frag=no),icmp(type=8,code=0),
> packets:104, bytes:10192, used:0.570s, actions:drop
> in_port(1),eth(src=00:16:41:55:fa:75,dst=00:1a:64:8f:a5:9e),eth_type(0x0800),ipv4(src=192.168.0.124,dst=192.168.0.171,proto=6,tos=0,ttl=64,frag=no),tcp(src=44899,dst=22),
> packets:7, bytes:606, used:0.020s, actions:0
> --- cut ---
>
> As you see, I was connected via SSH to the XCP server and the Test VM. I used
> the Test VM to send broadcast ping packets to 192.168.0.255.
> I used tcpdump on the XCP server listening on the two interfaces forming the
> bond and filtering for broadcast traffic and I saw the ICMP ECHO_REQUEST
> packets going out on both interfaces as on the production XCP servers.

OVS is definitely sending traffic to both ports.  I suspect that for
one reason or other it doesn't think those ports are bonded.

>> Can you also post the logs?
>
> Well, there aren't much log files - at least not much I was able to find.
>
> --- cut ---
> # ls -l /var/log/openvswitch
> total 4
> -rw-r--r-- 1 root root   0 Jan 25 13:51 ovsdb-server.log
> -rw-r--r-- 1 root root   0 Jan 25 13:51 ovs-vswitchd.log
> -rw-r--r-- 1 root root 160 Jan 25 13:52 ovs-xapi-sync.log
> # cat /var/log/openvswitch/ovs-xapi-sync.log
> Jan 25 13:52:19|1|reconnect|INFO|unix:/var/run/openvswitch/db.sock:
> connecting...
> Jan 25 13:52:19|4|reconnect|INFO|unix:/var/run/openvswitch/db.sock: connected
> --- cut ---

It's also odd that the ovs-vswitchd.log file is empty.  Can you check
the command line to make sure that it is supposed to be logging?