OpenJPA
  1. OpenJPA
  2. OPENJPA-1692

Add post creation callback to BrokerFactory

    Details

    • Type: Bug Bug
    • Status: Closed
    • Priority: Major Major
    • Resolution: Fixed
    • Affects Version/s: 2.0.0, 2.1.0
    • Fix Version/s: 2.1.0
    • Component/s: kernel
    • Labels:
      None

      Description

      There have been a couple instances (that come to my mind) where there was the need to do some work after creating the broker factory. Sometimes there is a necessity to ensure single threadedness @see (PersistenceProviderImpl.postBrokerFactoryInitialization(...)), other times we need to perform some additional configuration / initialization after the BrokerFactory has completed (openjpa.InitializeEagerly=true).

      I recently ran across a problem with the second case. In AbstractBrokerFactory.ctor(...) we attempt to create a broker even though the subclassed BrokerFactory hasn't able to execute it's constructor.

      Example pseudo code:

      abstract class AbstractBrokerFactory {
      AbstractBrokerFactory(Config c)

      { // setup if(InitializeEagerly==true) this.newBroker(); <-- BAD! The subclass hasn't executed it's constructor but it may be asked to create a new broker. }


      }

      class ConcreteBrokerFactory extends AbstractBrokerFactory {
      Config _conf;
      ConcreteBrokerFactory (Config c)

      { super(c); // setup _conf = c; }

      newBroker()

      { _conf.getSomething(); <-- _conf IS NULL because the constructor hasn't fully executed yet. }

      }

      I propose that we need a postCreationCallBack method on the BrokerFactory interface. This will be a single place that Bootstrap can drive this 'after creation ish' work.

        Activity

        Hide
        Pinaki Poddar added a comment -

        This is an useful suggestion. The other usage of such post-initialization method hooks can be leveraged to construct some critical internal data structures (e.g. MetaDataRepository) that are often immutable after initialization. One can initialize these structures with thread guards and then drop the guards for entire lifetime. This pattern will provide an opportunity to scale/perform faster in multi-core environment.
        JPA does provide indicators such as <metadata-complete> etc to make such pattern.

        Show
        Pinaki Poddar added a comment - This is an useful suggestion. The other usage of such post-initialization method hooks can be leveraged to construct some critical internal data structures (e.g. MetaDataRepository) that are often immutable after initialization. One can initialize these structures with thread guards and then drop the guards for entire lifetime. This pattern will provide an opportunity to scale/perform faster in multi-core environment. JPA does provide indicators such as <metadata-complete> etc to make such pattern.
        Hide
        Rick Curtis added a comment -

        My previous commit will regress OPENJPA-1491. Refactoring some code right now to take that problem into account.

        Show
        Rick Curtis added a comment - My previous commit will regress OPENJPA-1491 . Refactoring some code right now to take that problem into account.
        Hide
        Rick Curtis added a comment -

        Update fix version.

        Show
        Rick Curtis added a comment - Update fix version.
        Hide
        Rick Curtis added a comment -

        @Pinaki – Great point, but I want to keep this JIRA focused on properly designing the infrastructure for post creation callbacks and moving existing dependent code.

        I opened OPENJPA-1694 as a defect for us to be more intelligent about when we can eagerly load metadata. Perhaps when doing this work we will discover that there are other cases where it makes sense to remove some of the laziness of OpenJPA for the sake of scalability.

        Show
        Rick Curtis added a comment - @Pinaki – Great point, but I want to keep this JIRA focused on properly designing the infrastructure for post creation callbacks and moving existing dependent code. I opened OPENJPA-1694 as a defect for us to be more intelligent about when we can eagerly load metadata. Perhaps when doing this work we will discover that there are other cases where it makes sense to remove some of the laziness of OpenJPA for the sake of scalability.
        Hide
        Rick Curtis added a comment -

        Reopening this issue while I investigate some test failures.

        Show
        Rick Curtis added a comment - Reopening this issue while I investigate some test failures.
        Hide
        Rick Curtis added a comment -

        Closing this JIRA as the original issue has been resolved. Will handle problem with the DataCacheManager initialization with OPENJPA-1705.

        Show
        Rick Curtis added a comment - Closing this JIRA as the original issue has been resolved. Will handle problem with the DataCacheManager initialization with OPENJPA-1705 .

          People

          • Assignee:
            Rick Curtis
            Reporter:
            Rick Curtis
          • Votes:
            0 Vote for this issue
            Watchers:
            0 Start watching this issue

            Dates

            • Created:
              Updated:
              Resolved:

              Development