Rare project developers

[mc36 <>]

hi,

On 3/15/22 14:59, Alexander Gall wrote:
> On Tue, 15 Mar 2022 14:21:57 +0100, mc36 <> said:
>
> > right now, imho the forwarder could blindly follow the commands....
>
> I see. It makes sense to stick to that principle. Then we accept that
> freerouter may need to be aware of certain restrictions that apply
> when updating certain elements in the data plane. E.g. it may have to
> make sure that messages are sent in a particular order.
okkk then, let's do this...


> > state id 0/1 is just to shut/noshut
> > port_add/del/mod id speed etc... are only sent on (no) export-port, or change 
> > in the parameters...
> > that is, these messages should add/remove from the ports table in tofino...
> > in the case of change, you get the mod, in case of no export, you
> > get the del, in case of freshly added port, you get the add
>
> Ok.
>
> > ... if you wish, i can dissect the mod to a del followed by an
> > add...
>
> Yes, I think that would put us on the safe side in case some changes
> can't be made unless some other condition holds (like changig FEC
> requires the port to be disabled).
okk, then i'll get rid of port_mod, and do the 
state0/port_del/port_add/state1 stanza....


> > in any case, the state and mtu messages are sent in the order i thing the 
> > tofino would need:
> > shut before ports_del, and noshut followed by a ports_add, and ports_mod 
> > surrounded by 2 state messages...
>
> There can't be anything before ports_add, I think, since the port
> simply does not exist before it's added. But maybe I misunderstand.
my bad wording, so the port_add, then the state, obviously... :)

regards,
cs