On Wed, 2009-03-11 at 14:46 +0100, Iñaki Baz Castillo wrote: > Hi, imagine a domain "myomain.org" with a SRV _sip._udp register > pointing to two hosts: > > sip1.mydomain.org (priority 1) > sip2.mydomain.org (priority 2) > > Clients MUST contact sip1.mydomain.org since this entry has higher > priority. In case sip1.mydomain.org is down, clients MUST inmediately > use sip2.mydomain.org. > But when should they try to use sip1.mydomain.org? > > Basically I want that clients to come back to the first host and leave > using the second host when sip1.mydomain.org becomes reachable again. > > What should be the client behaviour? > What is the real behaviour of current clients supporting DNS SRV?
There are two critical principles that modify the specified behavior: - The two servers are *equivalent*, in that any request sent to one must have the same effect as the same request sent to the other. - Due to random failures in the network, any requester might see either of the servers as unreachable at any moment, in a way that is uncorrelated with whether the servers are functioning or the ability of other requesters to reach the servers. The second principle means that you can never, from the outside, certainly determine that an element is sending its messages to the wrong server, as the network may have lost messages to the higher-priority servers. I don't know what your specific situation is. But for instance, if you are talking about UACs registering, you have to be prepared for a UACs sequence of REGISTER requests being delivered at random between the two servers. That is, the two server have to jointly maintain a *common* registration database. Dale _______________________________________________ Sip-implementors mailing list [email protected] https://lists.cs.columbia.edu/cucslists/listinfo/sip-implementors
