TREX API connection lost in TREX used as VM

[hakl...@gmail.com]

Hi,
I am using TREX as a VM for some network convergence measurements. The TREX version used is 2.39.

TREX is started in stateless mode: ./t-rex-64 -i (maybe there is a better way avoiding the API connection loss?)

I am connecting with python API from a baremetal server. When simulating a network problem (link down or switch failure). The failure is sometimes more than 3 seconds. My problem is that this makes API connection lost (as also my mgmt interface and not only traffic ports get hit from the convergence test). Is there a way to change the 3 second API timeout to something longer?

Below is warning message that I get when network outage is more than 3 seconds:

2018-10-16 14:23:11 - [local][warning] - Connection lost - Subscriber timeout: no data from TRex server for more than 3 seconds


Later in my script (due to the earlier connection loss):
******

Exception stack (most recent call last):

#1 nfvi-850.py:349 - 'c.stop(ports = [first_port,second_port])'
#0 trex_stl_client.py:1193 - 'raise STLError("'{0}' - connection to the server had been lost: '{1}'".format(func_name, client.conn.get_disconnection_cause()))'

Summary error message:

'stop' - connection to the server had been lost: 'Connection lost - Subscriber timeout: no data from TRex server for more than 3 seconds'

/Håkan

[Yaroslav Brustinov]

Hi,

It's suggested to have different networks for control and data planes.

Thanks,

Yaroslav.

--
You received this message because you are subscribed to the Google Groups "TRex Traffic Generator" group.
To unsubscribe from this group and stop receiving emails from it, send an email to trex-tgn+u...@googlegroups.com.
To post to this group, send email to trex...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/trex-tgn/16091e5b-f5fe-429f-886f-72400884d6bd%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[hakl...@gmail.com]

OK, thanks for answering.

this suggetion is however not possible in this case. Trex VM is using different Openstack neutron networks (two for traffic and one for management) but same DataCenter underlay (switches and Links). This will usually be the case with VMs.


Thanks/Håkan

[Matt Callaghan]

I would like to toss in some counter to your statement that "usually will be the case with VMs";

If you're building traffic simulator devices into your lab network (usually in order to test data plane devices), you SHOULD be separating data and management planes;  Most shops will have a dedicated physical network for all management; and for the DUTs (hosts/guests) that require "data plane", they'll deploy additional physical network to those.

This guarantees product/test cannot overload or break management accessibility.

[hakl...@gmail.com]

Hi,
yes this is OK in some cases, maybe I was generalizing too much ;-).

In my case however, testing in an Openstack datacenter environment, there are for this case no options to have control and data plane of VM not sharing faith (all VM virtio interfaces leave Compute on same physical interfaces). Seems only thing I can do in this case is to run TREX API from VM locally and collect results afterwards.

Just wanted to know if tere was some way maybe setting API connection loss timer to something higher than 3 seconds. I think answer is at least for now that there is no such options.

Thanks anyway for a great project.


Thanks/Håkan