This content has been marked as final.
Show 2 replies
-
1. Re: EJBTHREE-1142 - Incomplete fix - Generic problem
wolfc Dec 3, 2007 10:51 AM (in response to adrian.brock)I was thinking about putting a TODO there.
I don't want to disassociate when there is an error on rollback or commit. In that case we have bigger problems. The whole MessageInflowLocalProxy should be invalidated then. Hmm, except for heuristic exceptions.
We then should resume the suspended tx. Oh bugrit, this will get complicated with the tidy up hack in there. What do we do if we have a suspended and a 'wrong' tx? -
2. Re: EJBTHREE-1142 - Incomplete fix - Generic problem
adrian.brock Dec 3, 2007 11:03 AM (in response to adrian.brock)"wolfc" wrote:
We then should resume the suspended tx. Oh bugrit, this will get complicated with the tidy up hack in there. What do we do if we have a suspended and a 'wrong' tx?
Are you talking about// Suspend any bad transaction - there is bug somewhere, but we will try to tidy things up if (currentTx != null && currentTx.equals(transaction) == false) { log.warn("Current transaction " + currentTx + " is not the expected transaction."); tm.suspend(); tm.resume(transaction); }
The log message should really be something like:log.warn("SOMEBODY HAS PLAYING WITH THINGS THEY DON"T UNDERSTAND!");
So there are more serious problems anyway.
This has nothing to do with the endpoint's state so there is no need to invalidate it.
A rollback exception when you attempt commit() is expected
e.g. optimistic failure in the database, etc.