Re: objects in cache after transaction abort

Armin Waibel Fri, 02 Jan 2004 15:09:20 -0800

Hi Jair jr,

Jair da Silva Ferreira J�nior wrote:

Hi Armin,

Thank you for your fast reply.

hmm, this should not happend. Do you use checkpoint() or flush()
in your code?
This only could happens when the object was already in DB. Only the
PersistenceBrokerImpl and RsIterator class push objects to cache.
If you abort the tx, the PB instance is not aware of locked objects.

    I use flush() and run an OJB query before aborting the transaction. I
don't use checkpoint().


If you call flush() all objects hold by ODMG-api are written to DB and
thus they are passed to the cache. If you now abort the tx the invalid
objects still in cache. Think this is a bug.
Thanks for the test case, reproduce the problem. Will check in a fix ASAP.

regards,
Armin

Interesting problem, please let me know if you can reproduce it.

    I was able to reproduce the problem in the above test case. I hope it
helps:

  Transaction t=implementation.newTransaction();
  t.begin();
  PersistenceBroker broker=((HasBroker)t).getBroker();
  Student s=new Student();
  t.lock(s,t.WRITE);
  s.setName("student");
  s.setEmail("[EMAIL PROTECTED]");
  s.setPassword("abcd");

((TransactionExt)t).flush();

  Criteria crit=new Criteria();
  crit.addEqualTo("_name",s.getName());

QueryByCriteria query=QueryFactory.newQuery(Student.class,crit);

  Iterator it=broker.getIteratorByQuery(query);
  assertSame(s,it.next());
  assertFalse(it.hasNext());
  t.abort();

  t=implementation.newTransaction();
  t.begin();
  broker=((HasBroker)t).getBroker();
  QueryByIdentity query2=new QueryByIdentity(s);
  boolean cacheOk=(broker.getObjectByQuery(query2)==null);
  System.out.println("cacheOk: "+cacheOk); //cacheOk==false always
  t.commit();

Thanks,
    Jair Jr

Jair da Silva Ferreira J�nior wrote:
Hi,
   I am using ojb1.0_rc5, ODMG api with OJB queries, mysql4 (innodb

tables) in Linux Red Hat 7.3 (kernel 2.4.20-20.7).

I moved from rc4 to rc5 recently and I noticed that sometimes the


objects persisted inside an aborted transaction are still in cache when
another transaction is started. Please, take a look at the above code to
better undestand what I am saying:

  Transaction t=implementation.newTransaction();
  t.begin();
  Person e=new Person();
  t.lock(e,t.WRITE);
  e.setName("some exam");
   t.abort();

   t=implementation.newTransaction();
  t.begin();
   PersistenceBroker broker=((HasBroker)t).getBroker();
  QueryByIdentity query=new QueryByIdentity(e);
  Person loaded=(Person)broker.getObjectByQuery(query);
  boolean cacheOk=(loaded==null); //here cacheOk==false sometimes
  t.commit();


   This problem does not happen everytime. So, sometimes cacheOk==true.
   I wasn't able to reproduce this problem in a test case, but it does

happen sometimes.

This problem has never happened when I used rc4.

Here's my cache configuration in OJB.properties:

ObjectCacheClass=org.apache.ojb.broker.cache.ObjectCacheDefaultImpl

descriptorBasedCaches=false
Thanks,
   Jair Jr


---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]


---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Re: objects in cache after transaction abort

Reply via email to