Hi

I have just upgraded to the tip of master, and some of my regression
tests are failing, but only when run as a group - individually they
pass.

Not done with my investigations yet, but it looks as though Ethan
Jackson's recent commit might have something to do with it.

https://github.com/openvswitch/ovs/commit/43b2f131a229238619ed9ae9ee64092fcae092ca

In a nutshell, flows created for earlier tests are being kept alive
and having their actions modified, but then the modification as a
result of a later test causes that later test to fail.

Here's the flow in the first test:

recirc_id(0),in_port(1),eth_type(0x0800),ipv4(proto=17,frag=no),
packets:0, bytes:0, used:never, actions:2

It undergoes a few mods over the next few tests:

recirc_id(0),in_port(1),eth_type(0x0800),ipv4(proto=17,frag=no),
packets:1, bytes:104, used:3.754s,
actions:userspace(pid=0,slow_path(controller))

recirc_id(0),in_port(1),eth_type(0x0800),ipv4(proto=17,frag=no),
packets:1, bytes:104, used:8.065s, actions:drop

But ends up as:

recirc_id(0),in_port(1),eth_type(0x0800),ipv4(proto=17,frag=no),
packets:3, bytes:304, used:10.135s, actions:2

when the rule becomes:

duration=2s, n_packets=0, n_bytes=0,
priority=1000,vlan_tci=0x0000/0x1fff,actions=output:2

Disregarding the flow's age, any UDP packet sent, VLAN or not, will
result in output:2, but this is not the correct action for all packets
(those with a VLAN tag).

My guess is that this is fairly deep and beyond my limited experience
with OpenVSwitch. However, I will continue to work this issue, but
thought you might like to know. And any advice much appreciated.

Tony
_______________________________________________
discuss mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/discuss

Reply via email to