[dpdk-users] link bonding driver, lacp mode

[dpdk-users] link bonding driver, lacp mode

[Alex Kiselev]

Hello.

I am facing another issue with LACP LAG port.

Jun 29 13:24:40 h4 the_router[24361]: PMD: Failed to allocate LACP
packet from pool

There are thousands of such messages in the syslog.
But my app was working ok at the same time.
The mbuf pool that is used to setup rx queues (rte_eth_rx_queue_setup)
was ok too.

So, is it a mbuf leak in the bonding driver?


P.S.
Does anybody have a success story working with LACP bonding ports?


--
Alex Kiselev

Re: [dpdk-users] link bonding driver, lacp mode

[Kyle Larose]

Hey Alex,

> -----Original Message-----
> From: users [mailto:users-bounces@dpdk.org] On Behalf Of Alex Kiselev
> Sent: Thursday, June 29, 2017 6:50 AM
> To: users
> Subject: [dpdk-users] link bonding driver, lacp mode
> 
> Hello.
> 
> I am facing another issue with LACP LAG port.
> 
> Jun 29 13:24:40 h4 the_router[24361]: PMD: Failed to allocate LACP packet
> from pool
> 
> There are thousands of such messages in the syslog.
> But my app was working ok at the same time.
> The mbuf pool that is used to setup rx queues (rte_eth_rx_queue_setup) was
> ok too.
> 
> So, is it a mbuf leak in the bonding driver?
> 
> 
> P.S.
> Does anybody have a success story working with LACP bonding ports?
> 

I've been able to get it working, but I was always transmitting. I'm not sure if I'd run into the issue you mentioned in your later email.

> 
> --
> Alex Kiselev

Re: [dpdk-users] link bonding driver, lacp mode

[Alex Kiselev]

Hi Kyle.

2017-06-29 16:06 GMT+03:00 Kyle Larose <klarose@sandvine.com>:
> Hey Alex,
>
>> -----Original Message-----
>> From: users [mailto:users-bounces@dpdk.org] On Behalf Of Alex Kiselev
>> Sent: Thursday, June 29, 2017 6:50 AM
>> To: users
>> Subject: [dpdk-users] link bonding driver, lacp mode
>>
>> Hello.
>>
>> I am facing another issue with LACP LAG port.
>>
>> Jun 29 13:24:40 h4 the_router[24361]: PMD: Failed to allocate LACP packet
>> from pool
>>
>> There are thousands of such messages in the syslog.
>> But my app was working ok at the same time.
>> The mbuf pool that is used to setup rx queues (rte_eth_rx_queue_setup) was
>> ok too.
>>
>> So, is it a mbuf leak in the bonding driver?
>>
>>
>> P.S.
>> Does anybody have a success story working with LACP bonding ports?
>>
>
> I've been able to get it working, but I was always transmitting. I'm not sure if I'd run into the issue you mentioned in your later email.

Yes, the bug "Failed to allocate LACP packet" appears only after some
idle time (about half an hour in my case).
I've just applied Robert Sanford's patch, so in half in hour I will
have figured out if it works or not.

>
>>
>> --
>> Alex Kiselev


--
Alex Kiselev

[dpdk-users] link bonding driver, lacp mode

[Alex Kiselev]

Hi.

I have a couple of questionts abount the link bonding driver.

I am using Link Aggregation 802.3AD (mode 4) of the driver.
Bonding ports are created using the EAL Command Line parameters.
A bond port works fine for some time (about one day), then it stops
forwarding packets.
Recreating aggregation group on the switch doesn't help. The only way that helps
in that situation is restarting my application. Last messages in the syslog are:

Jun 26 21:48:35 router the_router.lag[4917]: PMD: Bond 3: slave id 1
distributing started.
Jun 26 21:49:01 router the_router.lag[4917]: PMD: Bond 3: slave id 0
distributing started.
Jun 28 16:19:00 router the_router.lag[4917]: PMD: Bond 3: slave id 1
distributing started.
Jun 28 16:33:18 router the_router.lag[4917]: PMD: Bond 3: slave id 0
distributing started.
Jun 28 16:52:38 router the_router.lag[4917]: PMD: Bond 3: slave id 1
distributing started.
Jun 28 17:14:50 router the_router.lag[4917]: PMD: Bond 3: slave id 0
distributing started.
Jun 28 17:15:30 router the_router.lag[4917]: PMD: Bond 3: slave id 1
distributing started.
Jun 28 17:33:04 router the_router.lag[4917]: PMD: Bond 3: slave id 0
distributing started.
Jun 28 17:35:18 router the_router.lag[4917]: PMD: Bond 3: slave id 1
distributing started.
Jun 28 17:38:50 router the_router.lag[4917]: PMD: Bond 3: slave id 1
distributing started.
Jun 28 17:57:25 router the_router.lag[4917]: PMD: Bond 3: slave id 1
distributing started.
Jun 28 18:16:51 router the_router.lag[4917]: PMD: Bond 3: slave id 0
distributing started.
Jun 28 18:18:54 router the_router.lag[4917]: PMD: Bond 3: slave id 0
distributing started.
Jun 28 18:20:52 router the_router.lag[4917]: PMD: Bond 3: slave id 0
distributing stopped.
Jun 28 18:20:52 router the_router.lag[4917]: PMD: Bond 3: slave id 1
distributing stopped.
Jun 28 18:21:47 router the_router.lag[4917]: PMD: Bond 3: slave id 0
distributing started.
Jun 28 18:21:59 router the_router.lag[4917]: PMD: Bond 3: slave id 1
distributing started.
Jun 28 13:28:22 router the_router.lag[4917]: PMD: Bond 3: slave id 0
distributing started.

So my questionts are:

1) Is the bond driver supposed to automatically recover a bonding link
in a situation like that when pysycal links are ok?

2) What could cause a failure of a bonding link? LACPDU timeouts or drops?

3) The dpdk documentation indicates that rte_eth_tx_burst and
rte_eth_rx_burst functions should be called with intervals period of
less than 100ms.

Does it mean that the function should be called even there are no
packets to send?
What are the correct parameters for rte_eth_tx_burst function in a
case when there is no
packets to send?

Can it be called with the NULL parameter?
rte_eth_tx_burst(portid, queueid, NULL, 0);

Thank you.

Alex Kiselev.