Cyrus active-active cluster checklist

[Andre Nathan]

Hello

I've been keeping a list of configuration changes to setup Cyrus in an
active-active two-node cluster (I'm using DRBD and OCFS2 for that).

Here's the list of things I have so far:

* Don't use BerkeleyDB (set all databases which default to berkeley to 
  skiplist);
* Under configdirectory, move the "log", "proc", "db" and "socket" 
  directories to local storage and symlink (or mount -obind) them to
  avoid conflicts between the two cyrus instances;
* Change the append_newstage() function in imap/append.c to add some 
  host information to the "stage." file names, to avoid PID clashes 
  between the two cluster nodes;

Dave McMurtrie has also suggested me modifying the mmap() call in the
map_refresh() function to use MAP_PRIVATE instead of MAP_SHARED for
performance reasons. Is anyone aware of any side effects that this could
cause?

Any input on this list would be really appreciated. I think it's
important to keep a list of things like this that people thinking about
a similar setup can refer to (or give up on the idea :)

Thanks,
Andre