Hi Vincent,
As per this pull request:
https://github.com/acassen/keepalived/pull/2750
It looks like keepalived subscribes to gratuitous ARP/NDISC packets, which
may generate an enormous load on one single core.
In our case, on OpenStack network nodes, keepalived was getting absolutely
all ARP packets from all of the subnets connected to it. Imagine this with
thousands of VMs on the cluster, and you can easily imagine that this is
a real disaster: we naively had to scale the number of network nodes to
handle the (single core) load.
You'll find the patch we came with attached to this bug report, which is
the same as what we forwarded upstream, and was merged a few days ago.
I would love to get this fixed in Unstable first, and then to Trixie and
Bookworm. If you do not have time to manage either of them, please let me
know: I can attempt to deal it with the Stable release team, and hopefuly
make it happen.
Let me know,
Cheers,
Thomas Goirand (zigo)