Doctrine 2 - ORM
  1. Doctrine 2 - ORM
  2. DDC-2757

Manual transcation handling not possible when transaction fails, documentation gives wrong example

    Details

    • Type: Bug Bug
    • Status: Resolved
    • Priority: Major Major
    • Resolution: Fixed
    • Affects Version/s: 2.4
    • Fix Version/s: 2.5
    • Component/s: ORM
    • Security Level: All
    • Labels:
      None

      Description

      The current version of documentation states in section:

      13.1.2. Approach 2: Explicitly

      that one can use a custom transaction handling. One of the described things is that one has a responsibility for closing the entity manager, after exception is thrown.

      However, the fact is that the unit of work will close itself regardless, so closing the EM is not up to the user (documentation is wrong there). In fact, there is no way, one can say: Don't close the entity manager, I know what I am doing, just clear the thing and let me continue somehow (which is another topic/feature).

      What is an issue however is that once you reset an entity manager/obtain a new one, the old objects fetched are not usable, even though one just needs some information from them (for example, their primary keys) so that a new object can be created. When I try using them, there is a PHP Notice that spl_object_hash of the given object cannot be found inside entityIdentifiers (line 2852, UnitOfWork.php), and merging previously selected entities into the new entity manager does not work for some reason – the entity stays in detached state. So the only possibility is to refetch all data. Which means, scenarios such as these:

      foreach ($data as $record) {
          // update the entity
          // persisting new $record fails, as some constraint gets violated (unique, for instance)
          // catch the exception, obtain new EM, continue
      }
      

      ... won't work, or all highly problematic to say the least. I am not able to figure out, how to handle such scenarios properly using Doctrine. Just to make the scenario clear, there is no way I can check beforehand, that given constraint will be violated, short of complete exclusive table lock.

        Activity

        Hide
        Benjamin Eberlei added a comment -

        Yes, if a constraint fails it is usually necessary to start all over. However it should be possible for you to check for NULL, unique key or foreign key constraints in your code before calling flush.

        Show
        Benjamin Eberlei added a comment - Yes, if a constraint fails it is usually necessary to start all over. However it should be possible for you to check for NULL, unique key or foreign key constraints in your code before calling flush.
        Hide
        Benjamin Eberlei added a comment -

        I updated the documentation accordingly, as this is a documentation error.

        Show
        Benjamin Eberlei added a comment - I updated the documentation accordingly, as this is a documentation error.
        Hide
        Tomas Plesek added a comment -

        Okay, just wanted to point out that checking for constraints beforehand may not be possible. When at least two concurrent clients try to insert the same thing with the same unique constraint for example, there is nothing one can do to check for the constraint short of pessimistic lock on the table (because one process may insert conflicting data in between the check and the flush), which is not feasible in all scenarios. What I was trying to look up in the documentation was a way around this. And interestingly, obtaining new EM and merging entities (existing ones, fetched from database and not modified at all) back into it resulted in problems as well (the mentioned NOTICE problem).

        Show
        Tomas Plesek added a comment - Okay, just wanted to point out that checking for constraints beforehand may not be possible. When at least two concurrent clients try to insert the same thing with the same unique constraint for example, there is nothing one can do to check for the constraint short of pessimistic lock on the table (because one process may insert conflicting data in between the check and the flush), which is not feasible in all scenarios. What I was trying to look up in the documentation was a way around this. And interestingly, obtaining new EM and merging entities (existing ones, fetched from database and not modified at all) back into it resulted in problems as well (the mentioned NOTICE problem).

          People

          • Assignee:
            Benjamin Eberlei
            Reporter:
            Tomas Plesek
          • Votes:
            0 Vote for this issue
            Watchers:
            2 Start watching this issue

            Dates

            • Created:
              Updated:
              Resolved: