On 03/02/11 13:59, Benson Margulies wrote:
This is very interesting. It's a synchronization bug of mine which
never happens with 0.8.9 and is perfectly reproducible with the
snapshot. I have a 'background sync' thread.
So, the steps are something like:
load a new graph into the database with some ontology.
try to run my query.
in parallel, the sync thread calls TDB.sync, with no synchronization
between the thread running the query and the sync.
*blam*
I assume that this is entirely my problem unless you tell me
otherwise, in which case I'll attach the small test case I just
concocted.
Yes - it's the application's problem.
The system can't do the locking sensibly. It could do MRSW around every
operation but all it's protecting is the database RDF consistency, not
the application. An app might add a triple, look in the DB, add a
second triple, and want that under a single lock. The system can't know.
Transactions give you that. And sometime ... sometime ...
Andy
