[Babel-users] babeld flushes routes and they never come back until restarted

[Harshal Vora]

Hi Juliusz,

Thanks a lot.
We have been experimenting with Batman-adv as well and are noticing 
similar issues.

Also, we have identified few issues with the Hardware where the chip-set 
just gets stuck and never comes back up.
We were able to overcome the DHCP issues using some hacks.. one of the 
being what you mentioned.

Will keep you posted if and when we succeed and get a stable environment.

Regards,


On 05/17/2014 06:26 AM, Juliusz Chroboczek wrote:
>> We have moved to version 1.4.3 and still issues persist.
> Sorry for the delay, Harshal, I was very busy.
>
> You appear to be having two unrelated issues, and it's very difficult
> to work out what's going on from your somewhat confused descriptions.
> I would suggest first getting a pure mesh to work (without gateways),
> and only then to try adding your DHCP/NAT-ing stuff.
>
>> Once I restart the master machine (where I run babeld with "-C
>> 'redistribute metric 128'") either all the routes come back on all the
>> devices or all the devices get negative routes and they are never
>> fixed. (waited for half hour).
> That's most probably a DHCP issue.  Most DHCP clients don't install
> a default route if there's already one, and will not install a new
> default route if the previous default route disappears.
>
> The basic issue is that that's not what DHCP is designed for -- it's
> designed for configuring end hosts, not routers.  The right solution
> would be to have a statically configured address and route on your
> gateway; if you insist on running DHCP on the gateway, you could
> partially work around it by doing
>
>    in ip 0.0.0.0/0 le 0 deny
>
>> Every few minutes, we keep getting negative routes in the routing
>> table and then routes correct themselves.
> Have you definitely ruled out your interfaces going down-up because of
> hardware problems?  Babeld will flush all routes that go through
> a downed interface, and will take up to a dozen seconds in its default
> configuration to recover.
>
> You can work around the issue by reducing the hello interval.  But
> there's only so much that can be done with unreliable hardware.
>
> Sorry I cannot be of more help, but there's only so much that can be
> debugged remotely.
>
> -- Juliusz