Re: [dpdk-dev] Duplicating traffic with RTE Flow

[Jan Viktorin <viktorin@cesnet.cz>]

Hello Asaf,

it is a while we were in touch regarding this topic. Finally, I am
again trying to get work this feature. I've seen that sampling is
already upstreamed which is great. However, I am not very successful
with that. There is nearly no documentation, just [1], I found no
examples, just commit logs...

I tried:

 > set sample_actions 0 port_id id 1 / end
 > flow validate 0 ingress pattern end actions sample ratio 1 index 0 / drop / end
 port_flow_complain(): Caught PMD error type 1 (cause unspecified): port id action is valid in transfer mode only: Operation not supported
 > flow validate 0 ingress transfer pattern end actions sample ratio 1 index 0 / drop / end
 port_flow_complain(): Caught PMD error type 1 (cause unspecified): (no stated reason): Operation not supported

Using CentOS 7, DPDK 20.11.0, OFED-5.2-1.0.4.
NICs: MT2892 Family [ConnectX-6 Dx] 101d (fw 22.28.1002), MT27800 Family [ConnectX-5] 1017 (fw 16.27.2008).

My primary goal is to be able to deliver exactly the same packets both
to DPDK and to the Linux kernel. Doing this at RTE Flow level would be
great due to performance and transparency.

Jan

[1] https://doc.dpdk.org/guides/prog_guide/rte_flow.html#action-sample

On Fri, 18 Sep 2020 14:23:42 +0000
Asaf Penso <asafp@nvidia.com> wrote:

> Hello Jan,
> 
> You can have a look in series [1] where we propose to add APIs to DPDK20.11 for both mirroring and sampling for packets, with additional actions of the different traffic.
> 
> [1]
> http://patches.dpdk.org/project/dpdk/list/?series=12045
> 
> Regards,
> Asaf Penso
> 
> >-----Original Message-----
> >From: dev <dev-bounces@dpdk.org> On Behalf Of Jan Viktorin
> >Sent: Friday, September 18, 2020 3:56 PM
> >To: dev@dpdk.org
> >Subject: [dpdk-dev] Duplicating traffic with RTE Flow
> >
> >Hello all,
> >
> >we are looking for a way to duplicate ingress traffic in hardware.
> >
> >There is an example in [1] suggesting to insert two fate actions into the RTE Flow
> >actions array like:
> >
> >  flow create 0 ingress pattern end \
> >      actions queue index 0 / void / queue index 1 / end
> >
> >But our experience is that PMDs reject two fate actions (tried with mlx5). Another
> >similar approach would be to deliver every single packet into two virtual
> >functions:
> >
> >  flow create 0 ingress pattern end \
> >     actions vf index 0 / vf index 1 / end
> >
> >Third possibility was to use passthru:
> >
> >  flow create 0 ingress pattern end \
> >      actions passthru / vf index 0 / end
> >  flow create 0 ingress pattern end \
> >      actions vf index 1 / end
> >
> >Again, tried on mlx5 and it does not support the passthru.
> >
> >Last idea was to use isolate with passthru (to deliver both to DPDK application
> >and to the kernel) but again there was no support on mlx5 for passthru...
> >
> >  flow isolate 0 true
> >  flow create 0 ingress pattern end actions passthru / rss end / end
> >
> >Is there any other possibility or PMD+NIC that is known to solve such issue?
> >
> >Thanks
> >Jan Viktorin
> >
> >[1]
> >https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdoc.dpdk
> >.org%2Fguides%2Fprog_guide%2Frte_flow.html%23table-rte-flow-redirect-
> >queue-5-
> >3&amp;data=02%7C01%7Casafp%40nvidia.com%7C1a46005bec5245e729e708d
> >85bd24caf%7C43083d15727340c1b7db39efd9ccc17a%7C0%7C0%7C6373603060
> >73519816&amp;sdata=EOF%2Fz62crvBZK8rwzwKIWxj5cVlfPVnU3FLmcL9X2w0%3
> >D&amp;reserved=0