On 2/3/25 11:51 AM, Felix Huettner wrote:
> On Thu, Jan 30, 2025 at 11:31:16AM +0100, Dumitru Ceara wrote:
>> On 1/30/25 11:21 AM, [email protected] wrote:
>>> On Tue, 2025-01-28 at 17:42 +0100, Felix Huettner wrote:
>>>> Hi everyone,
>>>>
>>>> i still need to review the actual patch, but i wanted to join the
>>>> discussion.
>>>>
>>>> On Tue, Jan 28, 2025 at 04:19:37PM +0100, Frode Nordahl wrote:
>>>>> Hello all,
>>>>>
>>>>> Apologies for the tardy response, I've been pursuing an interesting
>>>>> bug related to the "add_route" NAT option and routing IPv4 over
>>>>> IPv6
>>>>> next hop, and wanted to see that to an end before chiming in.
>>>>> Patches
>>>>> imminent.
>>>>>
>>>>> On Tue, Jan 28, 2025 at 1:50 PM Dumitru Ceara <[email protected]>
>>>>> wrote:
>>>>>>
>>>>>> On 1/28/25 1:33 PM, [email protected] wrote:
>>>>>>> Hi Frode/Numan/Dumitru,
>>>>>>> Thanks for the review and interest in this patch, I'll put my
>>>>>>> two cents
>>>>>>> to the discussion below.
>>>>>>>
>>>>>>
>>>>>> Hi Martin,
>>>>>>
>>>>>>> On Mon, 2025-01-27 at 11:57 +0100, Dumitru Ceara wrote:
>>>>>>>> On 1/26/25 4:49 AM, Numan Siddique wrote:
>>>>>>>>> On Sat, Jan 25, 2025 at 5:52 PM Frode Nordahl
>>>>>>>>> <[email protected]>
>>>>>>>>> wrote:
>>>>>>>>>>
>>>>>>>>>> Hello,
>>>>>>>>>>
>>>>>>>>>> Note that this email is sent at a time convenient to me,
>>>>>>>>>> please
>>>>>>>>>> don't
>>>>>>>>>> feel obliged to read nor respond to it until a time
>>>>>>>>>> convenient
>>>>>>>>>> for
>>>>>>>>>> you.
>>>>>>>>>>
>>>>>>>>>> The timing of this bit of time skewed work is because I
>>>>>>>>>> anticipated
>>>>>>>>>> the output to be a new patch set for which we would have
>>>>>>>>>> a desire
>>>>>>>>>> to
>>>>>>>>>> try to get within soft freeze. It became apparent that it
>>>>>>>>>> would
>>>>>>>>>> likely
>>>>>>>>>> fit within the context of the patch set you diligently
>>>>>>>>>> posted on
>>>>>>>>>> friday, so I'll leave the result of it as a review
>>>>>>>>>> comment
>>>>>>>>>> instead.
>>>>>>>>>>
>>>>>>>>>> On Thu, Jan 23, 2025 at 3:00 PM Martin Kalcok
>>>>>>>>>> <[email protected]> wrote:
>>>>>>>>>>>
>>>>>>>>>>> This change builds on top of the new "dynamic routing"
>>>>>>>>>>> OVN
>>>>>>>>>>> feature
>>>>>>>>>>> that allows advertising routes to the fabric network.
>>>>>>>>>>> When LR
>>>>>>>>>>> option
>>>>>>>>>>> "dynamic-routing" is set on the router, following two
>>>>>>>>>>> new LRP
>>>>>>>>>>> options
>>>>>>>>>>> become available:
>>>>>>>>>>>
>>>>>>>>>>> * redistribute-nat - When set to "true", ovn-controller
>>>>>>>>>>> will
>>>>>>>>>>> advertise
>>>>>>>>>>>                      routes for external NAT IPs valid
>>>>>>>>>>> for the
>>>>>>>>>>> LRP.
>>>>>>>>>>> * redistribute-lb-vips - When set to "true", ovn-
>>>>>>>>>>> controller
>>>>>>>>>>> will advertise
>>>>>>>>>>>                          host routes to LB VIPs via the
>>>>>>>>>>> LRP.
>>>>>>>>>>>
>>>>>>>>>>> Co-authored-by: Frode Nordahl <[email protected]>
>>>>>>>>>>> Signed-off-by: Frode Nordahl <[email protected]>
>>>>>>>>>>
>>>>>>>>>> Thanks! Good to have these options resurrected and
>>>>>>>>>> adapted to the
>>>>>>>>>> current state of the fabric integration patch epic.
>>>>>>>>>>
>>>>>>>>>> After having read up a bit on the current state of review
>>>>>>>>>> of our
>>>>>>>>>> sibling options "dynmic-routing", "dynamic-routing-
>>>>>>>>>> connected" and
>>>>>>>>>> "dynamic-routing-static", I wonder if we need to make the
>>>>>>>>>> nat and
>>>>>>>>>> LB
>>>>>>>>>> options have similar prefixes?
>>>>>>>
>>>>>>> Yeah, "dynamic-routing-redistribute-lb-vips" is kinda mouthful,
>>>>>>> but if
>>>>>>> that's the prefix for every other related option, we should
>>>>>>> follow it.
>>>>>>>
>>>>>>
>>>>>> Alternatively we could try to change the dynamic-routing-
>>>>>> redistribute-*
>>>>>> options and instead adding a boolean for each "redistribute"
>>>>>> option we
>>>>>> could change "dynamic-routing-redistribute" to be a list of
>>>>>> entities we
>>>>>> want to redistribute, e.g.:
>>>>>>
>>>>>> LR.options.dynamic-routing-redistribute="connected;nat;lb"
>>>>>
>>>>> On the topic of NAT and LB, as discovered during review, there is
>>>>> an
>>>>> oddly named get_nat_addresses() function that can retrieve both NAT
>>>>> and LB addresses.
>>>>>
>>>>> To save us the work of refactoring it, perhaps we could use a
>>>>> single
>>>>> keyword for both NAT and LB addresses? I mean, would you ever want
>>>>> one
>>>>> without the other?
>>>>
>>>> "dynamic-routing-redistribute" looks like the shorter version.
>>>> However i see an advantage in the individual "dynamic-routing-
>>>> static",
>>>> etc. options as they could allow us to use more values than just
>>>> "true"
>>>> and "false". E.g. they could in the future be used to filter for a
>>>> given
>>>> prefix and only announce that.
>>>>
>>
>> In my opinion if we want to add support for route filtering (route-maps
>> or whatever we call them) a single string will not be enough anyway to
>> express potentially complex conditions.
>>
>>>> But that is just an idea on a potential benefit. I don't actually see
>>>> a
>>>> reason to currently built that.
>>>
>>> It occurred to me that we could also use `dynamic-routing-
>>> nat`/`dynamic-routing-lb-vips` names. This would be both shorter and
>>> kept the door open for non-boolean values in the future.
>>>
>>
>> My vote still goes to a single option with value the list of
>> protocols/objects we want to redistributed.  But I'm not going to block
>> the per-object implementation if that's what the majority prefers.
> 
> Hi Dumitru,
> 
> since on IRC on Thursday there was no heavy opinion in either way I will
> implement the "dynamic-routing-redistribute" for the next version of the
> patches. Then everyone can take a look how it feels.
> 

Sounds good to me, thanks!

> Thanks a lot,
> Felix
> 

_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to