Morning has broken on the American East Coast, but apparently not my app,
yet!
This new documentation is much better. Largely on its account, I'm going
to demur posting an operative example, but I will write about what I did
that should be noted by multiple readers of this thread.
I discard a dataset (close it, and obtain a new one in that thread if/when
needed) after it has been commit()/end()'ed for a WRITE operation. If this
isn't necessary, let me know (previous email suggested that it was), but if
it really is, that fact should be added to the documentation section you
just improved, as well.
Also, in my own practice, I use try {} finally {} to close datasets I have
created. Perhaps this is no longer important (because of transaction
ACIDity); if that be so, please say so (perhaps Java garbage collection and
active destroy-handlers now suffice), or else, your examples should close
datasets properly. You should say what, if anything, happens to you if you
fail to "close" a dataset all of whose WRITE transactions have been
successfully committed.
I do my work by means of a "Wrapper"/proxy object around "dataset". The
"Wrapper" contains a "dataset" variable, initially null. There are two
operations on the Wrapper:
1. Dataset *getMaybeCreateDataset*(){ if (dataset != null) return data;
else {dataset = TDBFactory.createDataset(staticStringContainingLocation);
return dataset;}
2. void *shutdown *() {if (dataset!= null) dataset.close();
dataset=null;}
The important thing is that I create one of these wrappers per thread, or
more precisely, in my server environment where request handlers are invoked
from a thread pool, I code
Wrapper dsw = null;
try {
Wrapper = new Wrapper();
.... all that was there ... now using Wrapper()...
} finally {dsw.shutdown();}
in each handler that expects to interact with Jena. I pass this object
around the call stack; it is not static (although I suppose the pointer
could be ThreadLocal, Occam's razor cuts against it). When I need a dataset
for a transaction, I call my Wrapper's getMaybeCreateDataset. If the
transaction was a WRITE transaction, I call its shutdown() after I call
end() on what I obtained.
Bernie
p.s., add "graphs" to the list of object-types to participate in the new
diagram set.
On Fri, Mar 30, 2012 at 5:22 AM, Andy Seaborne <[email protected]> wrote:
> On 29/03/12 19:51, Bernie Greenberg wrote:
>
>> I have seemingly gotten this to work (but I said that last time, too) and
>> endure our fullest battery of tests, under the assumption of per-thread
>> "dataset"s, discarding them after a commit/end and creating a new one for
>> the thread when needed.
>>
>> What the documentation needs, besides clear design guidelines for and a
>> minimal code example of multithreaded, transactional use, is a family of
>> diagrams explaining the scope, extent, uniqueness, and ownership of all
>> the
>> kinds of data objects you can get (e.g., maps, datasets ...) with respect
>> to each other, databases, and threads.
>>
>> Thanks again ----
>> Bernie
>>
>
> Hi Bernie,
>
> There's a partial, quick change to the documention so it's at least been
> vaguely helpful,
>
> http://incubator.apache.org/**jena/documentation/tdb/tdb_**
> transactions.html#multi-**threaded-use<http://incubator.apache.org/jena/documentation/tdb/tdb_transactions.html#multi-threaded-use>
>
> Andy
>
