Thursday, October 21, 2004

OK, it's days late, but I've had other things to contend with for the last few days. So here's an overview of Thursday...

I thought I'd finish testing the remote resolver, but testing was very difficult. Whenever I attempted a remote query, it would deadlock. This turned out to be due to a bug in DatabaseSession.findModelResolverFactory(). This method looked in the system model for the model type of a URI given to it. If it was found, then it knew it was local, and it used a local resolver facotry. If it wasn't found then it checked the protocol for the type of resolver to look for.

The problem was that the code assumed that if the protocol was "rmi" then it should have been local, so it threw an exception. The fix was to compare the non-fragment part of the URI to the current server name, and then throw an exception if it matched (since a match meant that the model should have already been found in the system model). If there was no match, then the model appears on another server, and it returns a remote resolver factory.

After finding and fixing this, I then ran into deadlocking problems. This was because the local system was obtaining a lock, and then asking for data from a remote resolver, which was actually on the same machine and required the same lock. That was where the day finished, and I took the bug up the next day.

No comments: