CTDB - the 'Mainz' plan for clustered Samba

[tridge at samba.org]

Jim,

 > The LMASTER change shouldn't be a problem, as this is just algorithmically
 > determined.  The LMASTER can't migrate like the DMASTER can, so there is no
 > penalty for moving it, and the LMASTER isn't guaranteed to know the true
 > DMASTER anyway; it can only point to what it knows as the last DMASTER
 > (which could in turn point to a new DMASTER, and so on).  There's no
 > optimzation lost by this change.

This is correct for version 0.1 of the plan, but in version 0.2 I
changed it so that the LMASTER does in fact always know who the
DMASTER is. That's the reason behind the rather strange 3 node message
during a dmaster transfer.

I decided to change this as I realised the practical cost of having
the LMASTER always know who the DMASTER is turns out to be very small,
and it makes analysis much simpler!

Cheers, Tridge