Re: clusters triggers ?

On Mon, 02 Apr 2007 00:35:54 -0700, sybrandb wrote:

> But assuming you mean RAC by cluster, and RAC enhances scalability, why

RAC is primarily high availability option.

> do you intend to cripple RAC by running code on a specific instance?

Because of something called "functional partitioning". If I have several nodes accessing the same database, I will try always forcing DML to the same objects to be done on the same nodes. Global locks are much slower then the local ones and you can't diagnose them from V$LOCK because other instances normally use other undo segments and you can't find out who is locking what. In other words, the fabled "closed graph" algorithm doesn't work if the locks are global, so deadlocks are entirely possible. I've seen situations when the only solution to deadlock problem was to bounce an instance or two. Locality of reference is a good principle and should be obeyed, RAC or no RAC.

-- 
http://www.mladen-gogala.com