README

This repository represents a collection of alternative components with tweaks and/or changed behaviour for the doctrine2-project.

UnitOfWork Save-State

Addiks\DoctrineTweaks\UnitOfWork\UnitOfWorkSaveState

This represents a save-state of a doctrine unit-of-work instance. Using this, you can always revert the state of a unit-of-work instance back to a previous state.

This can be useful for bulk-operations: Create a save-state at the beginning of the bulk-process and then restore it after every flush. That way the unit-of-work will never contain objects that were already flushed, saving memory and (execution-)time.

This differs from a simple entity-manager clear call because it leaves all objects in the unit-of-work that were in it when the save state was created. Keeping these objects in it can (and often will) help with keeping the state of the unit-of-work consistent with the rest of the application.

Remember to also call "gc_collect_cycles()" when restoring to actually free the memory.

Example:

<?php

use Addiks\DoctrineTweaks\UnitOfWork\UnitOfWorkSaveState;

$saveState = new UnitOfWorkSaveState($entityManager->getUnitOfWork());

foreach ($thingsToImport as $counter => $thing) {

    $importer->import($thing); # Import a thing, add it to a collection or something

    if ($counter % 100 === 0) {
        $entityManager->flush(); # Send new entities to database
        $saveState->restore(); # Remove imported entities from unit-of-work
        gc_collect_cycles(); # Clean up memory
    }
}

Transactional Entity-Manager:

Addiks\DoctrineTweaks\TransactionalEntityManager (Alternative for Doctrine\ORM\EntityManager)

Manages entities for persistence via ORM.

This entity-manager poses an alternative to doctrine's own entity-manager. In contrast to doctrine's entity-manager, this entity-manager never closes. Instead, on rollback it rolls the managed part of all managed entities back to the point of when the transaction was created.

It does this by managing not only one UnitOfWork, but a stack of UnitOfWork-instances. There is one UnitOfWork per open transaction plus the root-UnitOfWork. Each UnitOfWork in this stack contains the state of managed entities from the time when the next transaction started. The top UnitOfWork on the stack is always the one currently used. When a transaction begins, the topmost UnitOfWork is cloned and the clone put on top of the stack becoming the new current UnitOfWork.

When a transaction get's committed, the secont-topmost UnitOfWork get's removed from the stack, replaced by the current and topmost UnitOfWork. There is also an alternative method to commit called "commitAndDetachNewEntities" that also detaches all entities that were not known at the beginning of the transaction, which may be what you want to do for every iteration in batch-processing in order to prevent having an enourmous amount of entities managed.

When a transaction get's rolled back, the topmost UnitOfWork get's discarded and it's previous UnitOfWork (which still contains the state of the entities of when the transaction begun becomes the new topmost and current one.

There is also an alternative method for rolling back a transaction called "rollbackEntities" that rolls back the managed part of the state of all managed entities using the UnitOfWork that still contains the state of the managed entities when the transaction begun (resulting in an expensive rollback!).

The process described in the paragraph above allows for a meaningful workflow using transactions. If a process fails and causes a rollback, not only the state in the database get's rolled back, but also the state of all managed entities. This allows for the executed and failed process to be re-tried again or to continue with the next process using the same entity-manager. This can be done because the state of the runtime is known, it is the same as at the beginning of the transaction. This follows the meaning of a "rollback" => The return from a faulty state into a well- known state. If correctly used, this entity-manager could save some developers that care deeply about transactions a big headache.

Please note that this component is currently only roughly tested and should be considered more like a proof-of-concept. If you actually do make the decision to use this in production environment, please test everything beforehand and look for any impact on performance. Please provide feedback if you actually do use this anywhere or even better, fork and improve it.