Re: load balancing with rdomains

On Mon, Dec 18, 2023 at 01:53:50PM +0100, Marko Cupać wrote:
> On Sat, 16 Dec 2023 18:53:29 +0100
> Petr Ročkai <m.v0adgys6@fixp.eu> wrote:
>
> > Hi,
> >
> > On Sat, Dec 16, 2023 at 06:37:54PM +0100, Marko Cupać wrote:
> > > pass in on em0 from (em0:network) to <public> probability 50%
> > > rtable 1 pass in on em0 from (em0:network) to <public> probability
> > > 50% rtable 2
> >
> > IIUIC these two only add up to 75% probability – you presumably want
> > probability 50% on the second of the two (the first one then being a
> > match for everything that the later rule doesn't take up).
>
> Thank you, I can confirm that your solution:
>
> pass in on em0 from (em0:network) to <public> rtable 1
> pass in on em0 from (em0:network) to <public> probability 50% rtable 2
>
> ... results in what I was trying to achieve - it load balances over both
> uplinks without any blocked packets as long as both uplinks are active.
>
> What OpenBSD FAQ https://www.openbsd.org/faq/faq6.html#Multipath says
> for a bit different scenario applies to some extent for this one as
> well:
>
> "It's worth noting that if an interface used by a multipath route goes
> down (i.e., loses carrier), the kernel will still try to forward
> packets using the route that points to that interface. This traffic
> will of course be blackholed and end up going nowhere. It's highly
> recommended to use ifstated(8) to check for unavailable interfaces and
> adjust the routing table accordingly."

Uhm. This is not accurate. The kernel tracks interface state on routes and
will not select a multipath route that is not considered UP. There is a
smaller issue when there is no other multipath route. The lookup will
select the route and not fall back to a less specific one that is still
up.

Could please someone update the FAQ?

> ...except - if I'm not mistaken - ifstated should in this case adjust
> pf ruleset instead of routing table.
>
> If so, would using anchors be the best way? Any working examples to
> share? I used some simple ifstated rules but it is hard to wrap my head
> around probability percentages for all the use cases - first link up,
> second down and vice versa.
>
> Thank you in advance,
>
> --
> Before enlightenment - chop wood, draw water.
> After enlightenment - chop wood, draw water.
>
> Marko Cupać
> https://www.mimar.rs/
>

--
:wq Claudio