L2GW CSIT


Victor Pickard <vpickard@...>
 

Hi all,
I'm looking at a 3-node run fro L2GW CSIT on ocata and nitrogen. There is 1 failure out of 14 tests, but what is surprising to me is that more tests did not fail.

If you look at [1] the test results, you will see that test case 5 failed. This is where the L2GW and connections are created, which failed. There is no tunnel on OVS or in the HWVTEP emulator. 

But, somehow, the subsequent ping tests pass! Which is a bit odd considering there are no tunnels.
It makes me suspect that we are somehow getting l2 connectivity from VMs to control node other than the tunnel (which is not there).

I'll continue to look, may be an issue with how the tests are configured, or infrastructure connectivity, not sure yet.

If anyone on the list has other ideas or time to investigate, please take a look and share.

Thanks,
Vic


Vishal Thapar <vishal.thapar@...>
 

Hi Vic,

 

What I suspect is that, like most such issues, tunnel is not there when we look for it but shows up later. Likely test script returns failure after just one attempt.

 

To confirm we need ovs logs of the node running hwvtep emulator or output of ovs-vsctl show on the hwvtep/emulator, which is missing in teardown debugs [would be good to add].

 

Regards,

Vishal.

 

From: Victor Pickard [mailto:vpickard@...]
Sent: 25 September 2017 18:58
To: <ovsdb-dev@...> <ovsdb-dev@...>; K.V Suneelu Verma <k.v.suneelu.verma@...>; Vishal Thapar <vishal.thapar@...>
Subject: L2GW CSIT

 

Hi all,

I'm looking at a 3-node run fro L2GW CSIT on ocata and nitrogen. There is 1 failure out of 14 tests, but what is surprising to me is that more tests did not fail.

 

If you look at [1] the test results, you will see that test case 5 failed. This is where the L2GW and connections are created, which failed. There is no tunnel on OVS or in the HWVTEP emulator. 

 

But, somehow, the subsequent ping tests pass! Which is a bit odd considering there are no tunnels.

It makes me suspect that we are somehow getting l2 connectivity from VMs to control node other than the tunnel (which is not there).

 

I'll continue to look, may be an issue with how the tests are configured, or infrastructure connectivity, not sure yet.

 

If anyone on the list has other ideas or time to investigate, please take a look and share.

 

Thanks,

Vic

 


Victor Pickard <vpickard@...>
 

Vishal,
Thanks for taking a look. 

I think you are correct. In the same run, I looked at tc06 (first tc following failed tc), and indeed see tunnels in the hwvtep emulator db.

This is a 3 node run, and as you said in IRC, it takes a little longer for config to be applied.

I'll add a timeout/retry to this test, see how that goes. Also, will see about capturing ovs logs from hwvtep emulator.


Thanks,
Vic


On Mon, Sep 25, 2017 at 9:59 AM, Vishal Thapar <vishal.thapar@...> wrote:

Hi Vic,

 

What I suspect is that, like most such issues, tunnel is not there when we look for it but shows up later. Likely test script returns failure after just one attempt.

 

To confirm we need ovs logs of the node running hwvtep emulator or output of ovs-vsctl show on the hwvtep/emulator, which is missing in teardown debugs [would be good to add].

 

Regards,

Vishal.

 

From: Victor Pickard [mailto:vpickard@...]
Sent: 25 September 2017 18:58
To: <ovsdb-dev@....org> <ovsdb-dev@....org>; K.V Suneelu Verma <k.v.suneelu.verma@ericsson.com>; Vishal Thapar <vishal.thapar@...>
Subject: L2GW CSIT

 

Hi all,

I'm looking at a 3-node run fro L2GW CSIT on ocata and nitrogen. There is 1 failure out of 14 tests, but what is surprising to me is that more tests did not fail.

 

If you look at [1] the test results, you will see that test case 5 failed. This is where the L2GW and connections are created, which failed. There is no tunnel on OVS or in the HWVTEP emulator. 

 

But, somehow, the subsequent ping tests pass! Which is a bit odd considering there are no tunnels.

It makes me suspect that we are somehow getting l2 connectivity from VMs to control node other than the tunnel (which is not there).

 

I'll continue to look, may be an issue with how the tests are configured, or infrastructure connectivity, not sure yet.

 

If anyone on the list has other ideas or time to investigate, please take a look and share.

 

Thanks,

Vic