CTDB API

[tridge at samba.org]

Volker,

 > This logic (essentially the share_mode_conflict loop) has to
 > be coded in the DMASTER, and it needs to be aware about the
 > DB semantics. Not nice and clean, but fast :-)

yep. It's not just the round trip saving that is key to this
though. It is the fact that a a record is never locked while waiting
for a network operation. That is really the key!

As soon as you have a record locked 'remotely' then you get all the
worst nasties of cluster computing coming to ensnare you. A delay on a
network link between nodes A and B suddenly becomes a problem for node
C, and you start getting feedback and a catastrophic slow down of the
entire system. You also open up the possibility of deadlocks.

That's why I'm really hoping that we can get rid of
CTDB_REQ_FETCH_LOCKED. 

 > P.S: We could encode the rules in Javascript in the DB or
 > so....
 > 
 > P.P.S: This was a JOKE! :-)

:-)