On 2017-01-02 15:02, Ondrej Zajicek wrote:
On Fri, Dec 30, 2016 at 04:53:52PM +0100, Michal wrote:
On 2016-12-30 13:30, Michal wrote:
On 2016-12-30 13:26, Ondrej Zajicek wrote:
On Fri, Dec 30, 2016 at 01:13:59PM +0100, Michal wrote:
Stupid... increasing read buffer helped of course
sysctl -w net.core.rmem_default=1064960
I see, the bird is not increasing the read buffer on it's netlink socket. Would it be bad idea to create an configuration option for that?
Yes, that is probably a good idea.
Ok, let's see if can pull up some C skills.
So here comes the ugly patch. I can confirm that rising rcvbuf size to something bigger fixes my problem.
Hi
Thanks. Am i understand it correctly that rising rcvbuf size by this patch (without the sysctl change) fixed the problem? We will probably merge the patch after some cleanups, Jan Matejka is doing some netlink experiments to see if there are some other ways to fix the netlink socket to misbehave.
Yop, tested in production.