Configure it Out
A couple of days ago, I reviewed how our test teams have been doing to extend out the configurations we validate as we move closer to the next beta. Beyond per-feature validation, the teams collaborate to produce a “config of the week” that combines several different topology elements into a deployment that’s then the basis for both focused and ad-hoc validation that week. Each week the team rolls a new config, with the previous week’s config left up for a while so the developers can debug, test fixes, etc.
In the earlier phases of the project, the testing was very directed – essentially try one thing and make sure it works– ADFS support or SQL auth for instance. Then it’s a case of “Lather, rinse, repeat” for existing and new config options. Once the stuff works in isolation, we move on to simple combinations, then more complex ones, etc. Last week’s “config of the week” was an extranet deployment with fully qualified domain names, ssl, simultaneous vanilla LDAP & Windows auth (on separate URL namespaces, both hosting the same base content), inter-farm federation (between two medium farms), a mix of 32-bit and 64-bit builds, all running a non-English language. This week they switched to an English build and added multiple shared service providers (our new way of hosting different search, user profile, and other services for different portals all on the same farm) so one portal on the farm got its user profile info from an LDAP-based directory while another synchronized with the Active Directory. Finally, they drastically dialed back the privileges for all the service accounts to lowest privilege level.
Now as a program manager on the team, I am of course supremely confident in our engineering methodology and dev/test prowess. But even I would have given even odds that the first time all this stuff was turned on at once, it would have taken a couple days to get pages to render reliably. (On my more cynical days, I might have predicted needing a haz-mat crew to wipe down what was left of the lab.) We did find a bunch of bugs, but all the pages on all the portals in both farms rendered in both auth contexts. They’d never admit it, but I think the test guys were a little disappointed. Obviously, there are still miles to go on config scale-up/scale-out, but I like the progress.
--Jonathan Kauffman, Group Program Manager, SharePoint Portal Server
