Re: Domain check in draft-obispo-epp-idn-00.txt
Andrew Sullivan <[email protected]>
| Newsgroups | gmane.ietf.provreg |
|---|---|
| Message-ID | <[email protected]> |
On Wed, Jan 04, 2012 at 04:19:52PM -0500, Michael Young wrote: > Well a domain check is a read only transaction in registry, in a > create, you need to hold that create while you validate it in case > someone else is racing for the same unique string. Given load > balancers and multiple app servers you either time synch at the app > server or the DB. Most would go for the DB to lock then you face a > rollback if the validation fails. There are, of course, other possible implementations, some of which would not face this exact problem. (And indeed, the effects of rollback might be implementation-dependent as well.) Therefore, > Way back when domain drops weren't rate limited registries faced > "add storms". the add storm problem was at least partly an issue related to implementation, and not the protocol itself. While it is clearly a good thing that a client attempt to validate its input as well as possible before submission to an EPP server, it's equally obviously true that sometimes a client is going to be in a race to create an object. The protocol cannot rely on the client having checked that a creation meets registry policy prior to submission. So, while I certainly have no objection to including extensions to the check command (and think it's a good idea), I sure want to make sure nobody is planning to rely on check commands to save us from expensive validation of submitted name objects at create time. Best, A -- Andrew Sullivan <[email protected]> _______________________________________________ provreg mailing list [email protected] https://www.ietf.org/mailman/listinfo/provreg