[j-nsp] IPv6 multicast -- stuck in spt-pending

[Daniel Roesen]

On Wed, Mar 02, 2005 at 08:54:43AM +0100, Josef Buchsteiner wrote:
>  >> once  the  shared  tree  is  equal  to  the spt we will stay in
>  >> spt-pending  state which is perfectly ok and not a bug.
> 
> DR>  Uhm, how is that perfectly OK? It hangs in a state which should only
> DR>  persist until the first packet is received from the (S,G) via the
> DR>  correct RPF interface.
> 
>      It is normal/OK in the sense that it is implemented this way
>      since day one.

That was NEVER a good argument to support something as being correct. :-)

>      Since the RPF interface does *not* change the
>      (S,G) entry with the information spt-pending is indicating that
>      this entry is willing to switch once there is really a different
>      RPF interface and this why we save state.

Docs say, that spt-pending means "The software has initiated the switch
to the shortest-path tree for this source.". Fact is, that the router
is already on the SPT (=RPT). Fact is too, that it says that it's
waiting for something to happen which already DID happen. This is just
irritation (not just to me, but other Multicast folks who were looking
with me on the same alledged problem).

Can you please replace "spt-pending" with something like "spt=rpt" (or
just the most obvious "spt") for the case SPT=RPT? This would clear
things up and be unambiguos. People think there's something wrong that
the router (looks as if it) got stuck in a state which should be only
very transient.

>      It is not causing interop. problems and it is not against the
>      RFC...

It costed me two hours poking around what's going wrong there. And docs
are quite clear and support my interpretation:

spt-pending: The software has initiated the switch to the shortest-path
tree for this source.
spt: Entry is on the shortest-path tree for the source.

We just expected the state to switch immediately from "spt-pending" to
"spt".


>      Since you know now what it means it should not be irritating
>      anymore ;-)

Now _I_ know. Others _will_ waste time on that too. Docs don't describe
what you do describe here.


Best regards,
Daniel

-- 
CLUE-RIPE -- Jabber: dr at cluenet.de -- dr at IRCnet -- PGP: 0xA85C8AA0