On Fri, Jun 24, 2016 at 04:13:56PM +0200, Conrad Hoffmann wrote: > Yeah, I was pondering the same thing. DNS servers not capable of that > extensions (very few, I think) would ignore it, so always adding the OPT > record would be safe indeed.
OK good. > I just wasn't sure about the value itself. I > guess always setting it rather high would be an option. I didn't quite > finish thinking about whether there are scenarios where the server actually > sending such a big packet it might not make it through your network (e.g. > if the server would set the IP don't fragment bit). In our setup, IP > fragmentation kicks in and it works nicely. This is a very important concern indeed. It can definitely cause some trouble to people using encapsulation between DCs with incorrect PMTU discovery. It's also the same for people mixing jumbo frames and ethernet frames in their infrastructure since certain servers are on 9K and will not fragment. Over TCP they never notice such issues thanks to the MSS, they often start to report NFS issues in fact. If in order to work around this we need a directive, better ensure the default value is always safe then, instead of using an arbitrary value. Thus I guess your approach was the best one in the end. I'll wait for Baptiste to take a look and confirm he's OK with this. Thanks! Willy
