Allright, I'm pretty sure that it's all about the usage of
tx.failure() (which is very seldom needed btw). In the first code
snippet you call tx.failure() inside the for-loop and continues. That
seems a little odd to me since tx.failure() ensures that the
transaction can't be committed, allthough I see that
LuceneIndexService seems to ignore that fact. Then you continue to
insert artists knowing that they won't be committed anyway. That's
probably not right.
In your second snippet you won't be needing the tx.failure() since a
transaction is by default not successful until you call tx.success().
2008/5/11 Philip Jägenstedt <[EMAIL PROTECTED]>:
> On 5/11/08, Mattias Persson <[EMAIL PROTECTED]> wrote:
>> Hmm allright, great that it's working though. Also it would be nice to
>> know which structural changes you made to your code to make it work.
>
> What this code does is to take a bunch of ids and for each of them (a)
> replicate the musicbrainz entity with that id into the neo store (b)
> if the entity has a wikipedia relationship, extract a text snippet
> from wikipedia and store along with it. The musicbrainz webservice
> throws a 503 if you make more than 1 request per second. Before, I
> just threw a ServletException when that happened. Everything wrapped
> in a single transaction. Like this:
>
> Transaction tx = Transaction.begin();
> try {
> for (UUID id : artistIds) {
> MushArtist artist = maf.getArtistById(id);
> if (artist == null) {
> // replicate data from MusicBrainz WebService
> try {
> Query q = new Query();
> Includes inc = new Includes();
> inc.include("url-rels");
> Artist mbArtist = q.getArtistById(id, inc);
> artist = maf.copyArtist(mbArtist);
> } catch (WebServiceException e) {
> // FIXME: log this error!
> //tx.failure();
> //continue;
> throw new ServletException(e);
> }
> }
>
> try {
> artist.updateWikipediaBlurb();
> } catch (WikipediaException e) {
> // FIXME: log this error!
> tx.failure();
> continue;
> }
> }
> tx.success();
> } finally {
> tx.finish();
> }
>
>
> Now, I have one transaction for every id, so that successfully
> replicated data is commited immediate. If there is an exception, I
> explicitly call tx.failure() continue with the next id. However, in
> the code below I throw an exception and it works just the same (i.e.
> in the next request I'm not getting
> org.neo4j.impl.core.NotFoundException: Node[40] not found)
>
> for (UUID id : artistIds) {
> Transaction tx = Transaction.begin();
> try {
> MushArtist artist = maf.getArtistById(id);
> if (artist == null) {
> // replicate data from MusicBrainz WebService
> try {
> Query q = new Query();
> Includes inc = new Includes();
> inc.include("url-rels");
> Artist mbArtist = q.getArtistById(id, inc);
> artist = maf.copyArtist(mbArtist);
> } catch (WebServiceException e) {
> // FIXME: log this error!
> //tx.failure();
> //continue;
> throw new ServletException(e);
> }
> }
>
> try {
> artist.updateWikipediaBlurb();
> } catch (WikipediaException e) {
> // FIXME: log this error!
> tx.failure();
> continue;
> }
>
> artists.add(artist);
> tx.success();
> } finally {
> tx.finish();
> }
> }
>
> As you can see, the only structural change is moving the transaction
> inside the loop. It's a bit tricky to reproduce because the webservice
> only fails after a certain number of requests and there are long
> delays involved with actually accessing it.
>
> Philip
> _______________________________________________
> Neo mailing list
> User@lists.neo4j.org
> https://lists.neo4j.org/mailman/listinfo/user
>
_______________________________________________
Neo mailing list
User@lists.neo4j.org
https://lists.neo4j.org/mailman/listinfo/user
