[OpenAFS] Windows client-related issues

[Ken Hornstein]

>3) Whenever we lose an AFS DB server (we have 3, each on a separate subnet)
>Windows authentications take inordinately long, at least by end user
>perceptions.  After a recent prolonged loss of a DB server to hardware failure
>we had to resort to temporarily changing CellServDB info on the Windows clients
>until the replacement hardware could be brought on-line.  Do I have something
>woefully mis-configured, or did I oversell everyone's expectations here of
>redundancy and fault-tolerance in AFS?

The problem here is that the while the cache manager "remembers" that a
DB server is down, the userspace applications (like the authentication
widget) don't, and they invariably try the down one (why, I have no idea;
IIRC they pick a DB server at random and then loop through the list).

I sometimes think there should be an interface between userspace and the
cache manager that let's the userspace code ask the cache manager if a DB
server is down ...

--Ken