[Xorp-hackers] Patch to fix issue with interface removal and multicast.

[Ben Greear]

Pavlin Radoslavov wrote:
> Ben,
> 
> XORP has to be able to handle interfaces that disappear, including
> pure software-created interfaces like tunnels and VPNs.
> Hence, please submit a bug about the issue including a little bit
> more detailed instructions how to reproduce it. E.g., please include
> the starting configuration, and your monitoring script.
> 
> I am afraid that the patch itself cannot be accepted for
> the following two reasons:
> * It hides the real problem instead of fixing it.
> * Performing error checking by testing the content of the error
>   message itself is not acceptable: it is error prone and is not
>   efficient.
> However, the patch could be useful to point toward the problem,
> hence feel free to include it in the Bugzilla entry.

I agree the patch is a hack.  I'll open a bug with more info,
but I'm afraid I can't really give you a simple way to reproduce this:

It requires not only my own program but a partner's program
that is driving mine to easily reproduce this.  I'm quite sure
I can get permission to install their stuff on a system (for no charge)
if you want to test with our full package.  With everying enabled,
this is 100% reproducible with a few simple commands.

I think the basic problem is that pieces of xorp expect that it
knows the state of the OS, when in fact it can only know the *preferred*
state at any given time.

For instance, if I tell xorp to enable multicast on eth4, and eth4
does not currently exist, it should not fail the commit but remember
the config and apply it if eth4 ever does come into existence.  The
only time a commit should fail is if the requested config is logically
incorrect regardless of the current state of the OS's interfaces.

That is at least my opinion, and what I believe I need to drive
it programatically in a dynamic environment.

Thanks,
Ben

-- 
Ben Greear <greearb at candelatech.com>
Candela Technologies Inc  http://www.candelatech.com