Network Manager 100% CPU usage for large number of routing entries
I recently installed a RHEL8 box with FRR 7.4 in order to perform BGP peering with our ISP. We requested that the ISP send us the complete IPv6 routing table from the Internet which, at the time of this writing, equated to a little under 95,000 routes.
As soon as the routes started being populated by FRR, CPU utilization of Network Manager rose to 100% CPU usage and stayed there permanently. nmcli
also began to exhibit slowdown, sometimes taking tens of seconds before returning after having issued a command. And since nmcli
by itself typically shows all interfaces and their associated routes by default, the listing of 95,000 routes took a while. Ultimately, I had to disable Network Manager and go back to using the legacy network-scripts, even though they are considered deprecated in RHEL8, as Network Manager in this shape was unusable.
I noticed that the systemd journal showed entries such as these:
NetworkManager[733]: <info> [1603969070.2556] platform-linux: netlink: read: too many netlink events. Need to resynchronize platform cache
Would this behavior be considered a bug? Or would it simply be considered as operating parameters that are outside the number of routes that Network Manager was designed to realistically support?