BGP session flapping

[gplhost]

Hi,

It seems that quite often, we get BGP session flapping. The result is quite a disaster, as Quagga seems to flush all the v6 routing tables until it gets some fresh ones from Tunnelbroker.

How to prevent this?

[kcochran]

When a BGP session drops, it's supposed to pull all the learned routes.  Anything else would cause a blackholing of traffic.

That said, what do you have in your logs showing the cause of the session drop, and for which tunnel?  Feel free to fire that information off to ipv6@he.net.

[kasperd]

When a BGP session drops, it's supposed to pull all the learned routes.  Anything else would cause a blackholing of traffic.

Dropping the only route you have to a destination isn't going to get the traffic there either. And depending on why the BGP session dropped, the routes you learned from it might still work.

When stuff breaks, it is always a good idea to look for multiple ways to avoid the problem. If you can find only one fix, it means there is only one thing that has to go wrong before it breaks again.

The following sounds like a good list to me:

• Find out why the BGP session drops and do something about it.
• Run multiple BGP sessions for redundancy.
• Find out if the software can be configured to keep the routes for a little longer, if there is no other path.

Are there any other providers, which offer BGP tunnels similar to HE suitable for a redundant setup? Wikipedia lists local providers in Ukraine and Indonesia, but no provider with a global presence apart from HE.

[gplhost]

It turned out that doing:
echo "net.ipv6.route.max_size = 40960" >>/etc/sysctl.conf
sysctl -p

solved the problem. The default max in the Linux kernel is 4096, which is by far not enough to hold all the HE routes.

Thanks for the help,

Thomas