Grant Maxwell wrote:
You see Brian the restart situation/node going down is already taken care of. It's only a problem if the class record exists but the item record does not.
That would only take care of the case when *all nodes* are restarted at the same time. If you just restart a node that happens to hold the item record but not the one that happens to hold the class record, you'd still be in trouble.
The way to deal with this is to write them all to the cache as a single object (perl hash list probably). That actually delivers exactly the guarantee we need. And it's really simple.
Anyway, it looks like you've come up with a solution to your problem, so my point above it kind of moot :-)
Dean.
