From: Brian Kantor Sent: Thursday, January 24, 2019 3:58 PM
I agree.
It seems to me that testing with almost-valid data (well formed, but with disallowed values) as well as fuzz-testing are essential parts of software quality control.
To be frank, Have blasted packets at the platforms from ixias and sipernts to see if they break gracefully, loaded millions of routes and thousands of VRF and BGP sessions to see what happens, even designed the backbones with separate Internet and VPN RRs, and enabled enhanced error handling, but have I ever sat down and generated BGP packets with slight deviations to see how the BGP session, process or whole RPD copes with these? I've got to say no, never. And judging from the overly positive (or even negative) responses to the BGP Experiment I'm not alone in this. Otherwise everyone would be like, nah I don't care as I have all my bases covered and I know how my BGP behaves processing exceptions. adam