[DISCUSS] Evicting dirty instances

This topic was raised by Christiaan http://mail-archives.apache.org/ 
mod_mbox/db-jdo-user/200708.mbox/%3c12232225.post@...%3e

Before the JDO spec introduced the concept of application-defined  
flush(), the evict method could be used to make an instance hollow  
and therefore subject to garbage collection.

When we introduced flush(), it became possible for all of the  
application state in the cache to be erased from the cache, leaving  
only the life-cycle state of the instance in memory for future  
reference.

My proposal is to require that the JDO implementation treat flushed  
dirty instances the same as hollow and persistent-nontransactional  
instances, allowing them to be garbage collected. The uniqueness  
guarantee only applies if the application itself holds a strong  
reference to the instance.

This proposal would allow a well-designed program that creates or  
updates a large transaction set of persistent instances to flush  
periodically and not run out of memory.

If an application constructs a large number of instances and flushes  
them, the only artifact remaining in the cache would be the memory  
that the instance was new. That is, if the instance were subsequently  
queried or obtained by id, the instance would still be persistent-new  
although the instance retrieved might be a different Java instance  
from the instance created and flushed.

We could still use the evict method to remove even the memory of the  
life cycle state of the instance. So if the persistent-new instance  
were flushed, it would still be persistent-new, but if it were then  
evicted, it would be subsequently retrieved as a hollow or persistent-
nontransactional instance.

Please comment.

Craig Russell
Architect, Sun Java Enterprise System http://java.sun.com/products/jdo
408 276-5638 mailto:Craig.Russell@...
P.S. A good JDO? O, Gasp!