Oliver Zeigermann <[EMAIL PROTECTED]> writes:
Ah, I see, it works now. Shall I check it in for you (I assume you have not been granted commit access, yet)?
Yes please. Needs a lot of cleanup, but its better than no tests on Unix at all.
Done :)
For now I get 105 errors for 215 tests using the default Domain.xml (TxFileStores). Lots of internal server errors, because transactions are rolled back. Do not yet know, why they are rolled back :-(.
Ooops! I have only the 14 described before, but I know why :) I have only one thread running, you have 10. I'd propose to set the default to 1 anyway...
The tx file store has *no real deadlock* detection, but simply assumes there was a deadlock when transactions time out. This still is a TODO!
When you have 10 threads running, transactions take longer and finally time out. For now I will add a timeout parameter to the tx file store...
Any volunteers for deadlock detection? I promise it is fun and I will help as much as I can!
Basicly it works like this: If a lock can not be acquired, check which thread owns it and see if it waits for this first thread. If so rollback the thread that tried to acquire this new lock as a deadlock victim. More formally you will have to search for a cycle in the lock dependency graph, but who cares for formal stuff ;)
Oliver
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
