Ideally you are correct… Unity subscribers homed on other servers in the site that are still up should still be able to get in and snag their messages while other servers are off line. Unity 2.4.5 does a reasonable job of this using a MAPI “Traffic Cop” model that keeps track of which servers are up/down and will not allow subscriber to try and connect to their mailstore on a box we know is down… the reason for this is the MAPI threading model in SP3 and earlier leaves a bit to be desired and more or less all the users on the box at any given time are sharing the same thread. If one subscriber gets stuck trying to get their messages, it chews up the thread for everyone else on the box regardless of which Exchange server they’re on. The really bad part is Exchange has a time out of somewhere around 10 minutes for the attempt to access the mailbox to fail and return control to us… ugly.
Outside callers have their own MAPI thread such that we can always process calls and take messages regardless of if any or all of the other Exchange servers in the site are up or down… since outside callers never log into their mailbox, this thread is always cool. The MTA on our local box queues these messages up and will get them to the appropriate home store when they're back in action.
Naturally there is a small window where subscribers can slip in to get their messages before we know the exchange server they’re homed on is down (we aren’t polling every second… that’d be a pretty bit hit). Problem. In this case we will kill that thread and start a new one after we realize it’s hosed, however this timeout could still be a few minutes which will back up everything for all users behind that guy.
All in all, kind of a pain. One of the reasons we require SP4 for the upcoming 2.4.6 release of Unity is to resolve just this problem. As you noted in another post out here, Microsoft suggested you upgrade to SP4… they’ve cleaned up some of their MAPI threading stuff such that we can now have a separate MAPI thread for each user in the system without it having a resource lockup situation. In this case only the one subscriber can slip in under the radar and get stuck trying to check for messages on a down system… it will not affect the other callers in the system. As such if they’re homed on servers that are still up, they should have unrestricted access to their messages.
In short, 2.4.6 with SP4 on the box should completely resolve this issue (finally!).
Jeff Lindborg
Unity Product Architect
Active Voice
jlindborg@activevoice.com
http://members.home.net/jlindborg
