On 25 Mar 2015, at 11:02, Paul Vixie <p...@redbarn.org> wrote: > are you sure? because if a server is going to close connections prematurely, > it may do so based on instantaneous load spikes, such as the number of active > connections reaching 80% of quota, where the premature closure is designed to > avoid connection overload rather than signalling actual connection overload. > if in the UDP case we would respond to an unanswered query by repeating the > same query to the same server, then in the TCP case we might want to do > likewise. > > it could also be that the traditional UDP retry mechanism is seriously > flawed, and that even in that case we should move on to the next server, and > only return to the timed-out server if all other servers also time out. > > this bears discussion.
Yes, it does :) > if 100% of the installed base had not at one time in the history of DNS > behaved that way, i could agree. but because the specification's guidance is > demonstrably light in this area, is there any harm in extending that guidance > -- "for the avoidance of confusion or doubt” ? That’s fair - I guess it wouldn’t do any harm. >>> but since we must also guide the initiator to not leave a tcp session idle, >> >> We must? > > perhaps you've got me procmailed. see below: No, I saw those. I just didn’t agree with the conclusion :p I will however re-read those messages and see whether my position changes. Ray _______________________________________________ DNSOP mailing list DNSOP@ietf.org https://www.ietf.org/mailman/listinfo/dnsop