unstable BGP sessions after debian 10 -> 11 upgrade

Fran fatal at mailbox.org
Sun Oct 24 17:23:51 CEST 2021


Hey,

I ran the following combinations without problems:

Debian 10, kernel 5.8 backports, bird 2.0.7
Debian 10, kernel 5.10 backports, bird 2.0.8 (from suse repo)
Debian 11, kernel 5.10, bird 2.0.8 (from Debian Bookworm repo).

sysctl values (unchanged since ~ 2 y):

net.core.rmem_max=67108864
net.core.wmem_max=67108864
net.core.wmem_default=8388608

only relevant syslog messages is
"Kernel dropped some netlink messages, will resync on next scan"
on bird (re)start.

For context:

1 iBGP and 3 eBGP sessions

bird> show route count
2903596 of 2903596 routes for 852313 networks in table master4
520418 of 520418 routes for 130386 networks in table master6
239213 of 239213 routes for 239213 networks in table ROA4
46527 of 46527 routes for 46527 networks in table ROA6
Total: 3709754 of 3709754 routes for 1268439 networks in 4 tables

Best,
f

On 24.10.21 16:23, Christoph wrote:
> Hello,
> 
> we upgraded our debian BGP routers from debian buster to bullseye.
> 
> On debian 10 they used the repo https://bird.network.cz/debian/
> now they use the bird2 package directly from the official debian repos,
> both repos contain BIRD version 2.0.7.
> 
> 
> After upgrading and rebooting we noticed that iBGP sessions are constantly dying every few minutes.
> 
> 
> The logs also show multiple of these log entries:
> 
> <WARN> Kernel dropped some netlink messages, will resync on next scan.
> <WARN> Event ... took 31415 ms
> <WARN> I/O loop cycle took 31529 ms for 9 events
> <WARN> Netlink: File exists
> 
> 
> Especially the warning "Kernel dropped some netlink messages" was not new but we previously solved them with these sysctl.conf settings:
> 
> 
> # solves the "Kernel dropped some netlink messages, will resync on next scan." BIRD Warnings
> # https://bird.network.cz/pipermail/bird-users/2017-September/011542.html
> # (we did not change the wmem_default since this is apparently not needed)
> net.core.rmem_max=4194304
> net.core.rmem_default=4194304
> net.core.wmem_max=4194304
> 
> 
> The most obvious change is the linux kernel version:
>   4.19.208 (buster)
>   5.10.70 (bullseye)
> 
> To avoid constantly dying iBGP sessions we increased the keepalive and hold timers.
> Are you observing similar issues on your debian 11 systems?
> Does the newer kernel need higher rmem values?
> 
> best regards,
> Christoph


More information about the Bird-users mailing list