Re: TDB Dataset.commit

["Dr. André Lanka" <ma...@dr-lanka.de>]

Hello Andy,

sorry for my very, very late answer. We rolled out many new features on 
our platform in the last weeks, so I was too busy to integrate our TDB 
0.8.10 changes to TDB 0.9. Yesterday I started again.
Our unit tests print out a ConcurrentModificationException if we iterate 
over the model's statements (read transaction) and update the model in 
another thread (write transaction) concurrently. I could imagine that is 
is an intended functioning, but I'm not sure. We generally operate on 
the model using the following pattern:

public static <T> T doWork(int id, final boolean writeAccess, 
StoreWork<T> work) throws Exception {
     Dataset set=newDataset(id);
     set.begin(writeAccess ? ReadWrite.WRITE : ReadWrite.READ);

     try {
       T result=work.execute(id, set.getDefaultModel());
       set.commit();
       return result;
     } catch (Exception e) {
       set.abort();
       throw e;
     } finally {
       set.end();
     }
   }

where id is the ID of the graph we want to access and work.execute does 
the actual work on the model. I hope we use the transaction framework 
correctly.

Do you know if the iterator behaviour above is intended?

Best wishes
André



On 13.11.2012 21:49, Andy Seaborne wrote:
> On 13/11/12 12:26, "Dr. André Lanka" wrote:
>> Hi Andy,
>>
>> On 12.11.2012 18:21, Andy Seaborne wrote:
>>> Is each store using named graphs?  Or is one store, one graph?
>>
>> One store is one default graph without any special named graphs along.
>> The reason for having multiple stores is that we need write access to
>> each graph in parallel. Serialising the write access would lead to poor
>> performance for our application. By having multiple stores, we can write
>> to each store independent of the current state of the other stores. I
>> assume that this would also be a problem when we use dynamic datasets.
>> Or is there something I missed again?
>
> No - that's a reasonable setup, in fact one of the few I can think of
> for multiple stores each with one graph.
>
> I look forward to the patch,
>
>      Andy
>
>>
>> Best wishes
>> André
>>
>>
>

-- 
Dr. André Lanka  *  0178 / 134 44 47  *  http://dr-lanka.de