I’ve been working on an OpenStack lab, and one of the (many) issues I encountered was that my instances (VMs) wouldn’t get their DHCP addresses from the neutron controller.
This ended up being due to the fact that I’d built CentOS 7 off a security-focused kickstart which set the IPTables FORWARD policy to “DROP” (a bad idea for an OpenStack node, along with SELinux and umask changes).
Before I reached this conclusion, in the process of debugging, I discovered the following disturbing error messages from stevedore.named in linuxbridge-agent.log:
Could not load neutron.agent.linux.interface.BridgeInterfaceDriver Could not load neutron.agent.linux.iptables_firewall.IptablesFirewallDriver
These had me concerned for a while (I wasted several hours debugging), since they seemed directly related to the fault I was experiencing.
Turns out, they’re completely unrelated, and they’re harmless.
As I discovered, provided the error is a WARNING, it can be ignored. It’s simply an artifact of the fact that stevedore (whatever that is) can’t load the class by alias, since we are supplying a class path. If the error was serious, it would be a CRITICAL error, which would stop the DHCP agent from loading altogether.
So, if you encounter this error, but your DHCP agent is still running per
neutron agent-list, ignore it and look for your problem elsewhere :)