Basic doubt on cache transaction when modifying cache value by reference

[sudheerk84]

Since infinispan cache works by identify, I understand that changing the value of the object reflects in changing the value of the cache.

Example

Person p = new Person("Sudheer");  //Assuming person is a custom object with a single strring as its content.

cache.put("key1",p)

Now when i execut ethe following code

Person person = (Person) cache.get("key1");

person.setName("newName");  //This value is also update in teh cache since it uses object identity, that it  cache.get("key1") - now returns ->  Person("newname").

Now what is the expected behaviour when the following happen inside the transaction ,assuming the cache already contains ("key1", Person("Sudheer") )

1.   ut.begin()

    Person person = (Person) cache.get("key1");

    person.setName("newName");

    ....... other statements

     ut.commit();

2. 

   Person person = (Person) cache.get("key1");

    ut.begin()

    person.setName("newName"); -- ?/is this statement considers as cache operation.????

    ....... other statements

     ut.commit();

Are 1 and 2 same ?

Infinispan doc says - http://community.jboss.org/wiki/Infinispantransactions

"On every cache operation Infinispan does the following:" .......  My doubt is,  Is operation on the cache  value(person.setName() above)  considered as cache operation ?

If yes -  In case 2 above is a local copy of Person maintained inside transaction  to ensure that changes inside transaction are not reflected to objects outiside ?

If no - Is there any other way to ensure that value objects can be modified within scope of a transaction.?

.

[galder.zamarreno]

Whenever you update the object, you should call cache,put() or equivalent, otherwise, even if the changes might be visible in the local machine, these won't be replicated/distributed to other nodes.

[sudheerk84]

Hi Galder,

i did observe that behavour. The changes were reflected locally but not distributed.

So as a guideline - we are planninng to make all objects(POJO) in cache as immutable(i .e is cache.get always returns immutable POJOs) . This will force the developer to ensure that he doesnt make pojo modifications(by calling setter methods) by mistake and expect it to get distributed across cluster.

Is this a good guideline when cache objects are POJO's and we are heavily relying on transactions.