Hi,
I have been building VyOS images from source regularly to test on an appliance a setup with 3 ISPs each in a VRF. I noticed starting approximately the start of April the CPU usage spiked up from 1-2% to approximately 20%; the cause seems to be vyos-netlinkd consuming 100% of a CPU core constantly:
As an example, here is a CPU graph from the appliance; at approximately Tue 11:30 I rebooted VyOS to switch from an image I built on 2026-03-29 to one I had built 30 mins prior (2026-05-12) without any configuration modifications.
I found this topic that was recently started: High CPU usage on 2026.04.10-0031-rolling - #3 by c-po
However the problem appears to be different; if I strace the process for any exec calls I don’t see anything. I had a look through the logs/journal and couldn’t find anything indicating a problem/loop. If its useful
I took a strace for a couple of seconds to see what was happening: strace.txt (4.5 MB)
Because the setup from my ISPs isn’t what I typically encounter (DHCP/DHCPv6 required with 5 minute renewals) and/or because I am taking multiple full routing tables I put the same software image on an existing VyOS firewall used for IPSEC connectivity which has a much simpler setup - there is no BGP in VRFs, a tiny routing table, no DHCP etc. - that firewall has the same issue.
I had a couple of ideas of some things to test to see if the issue is also encountered there to maybe narrow down the cause a bit:
- Remove the VRFs entirely to see if the issue occurs. The IPSEC firewall does have a single VRF configured for management (no routing protocols configured) which is something in common.
- Remove BGP completely to see if the issue occurs with/without VRFs.
I wanted to see if anyone else had experienced the same issue or had any better ideas of where to start to troubleshoot this to save myself some time.
