Jira issue originally created by user gcaseres:
I've been reading Doctrine2 Batch Processing documentation.
I've a simplified the code and made a sample where I'm using a Repository inside the loop:
for ($i=1; $i<=10000; <ins></ins>$i) {
$user = $this->_em->getRepository('some user class')->find($i);
$this->_em->clear();
//Clear variables to ensure garbage collections
unset($user);
$user = null;
}
I expect that this script will consume some constant memory in all iterations, but what happens is that every iteration raises memory consumption (more iterations, more memory).... i think that the clear method has some sort of memory leak.
In my production environment (with complex script), i reach a memory limit exception even with 600MB limit... but if I clear the EntityManager on every iteration, shouldn't memory be freed?
Sorry for my bad english.
Comment created by @beberlei:
You are probably running symfony2 in debug mode? Is the SQL logger enabled? This is probably not a Doctrine problem but something in your code / Symfony that keeps increasing the memory.
Comment created by gcaseres:
I've executed the script in debug and prod mode, but I had the same problem in both modes.
I don't think it's a Symfony problem because I had measured memory consumption only before, after and inside the for loop (no symfony methods involved).
About my code, I'm using simple clases with no business code, only simple Doctrine mappings (and standard repository).
Have you tested a similar code? I don't understand why memory consumption continues raising if I'm "destroying" the objects.
I tried with gc_enable and gc_collect_cycles but no success... every iteration increases memory consumption like if the previous loaded objects weren't destroyed... maybe the repository is instancing other objects in every find call that are not destroyed?
Comment created by @beberlei:
are you using lifecycle listeners? access global state or something?
Comment created by @beberlei:
Can you generate an xdebug trace for some of the $i's ? say 100 and 1000 with xdebug_start_trace("/tmp/loop".$i); and xdebug_stop_trace(); and upload them? Maybe you can compare yourself, where in the loop the memory increases and if clear even empties it or not.
Comment created by @ocramius:
Any news about this one? There's been more than one case where the Symfony data collector (for debug) caused problems like this one... Imo this is not a ORM issue.
Comment created by @beberlei:
No feedback given.
Comment created by mvrhov:
I've been debugging a similar issue today. And Yes, the culprit is the Symfony's data collector. Running the command with --no-debug worked like a charm.
Comment created by pourquoi:
same issue here with 2.2.3, php 5.4 & symfony 2.1
have a symfony command running as deamon with --no-debug and no listeners
while(true) {
$q = $this->em->createQueryBuilder()->select()...->getQuery();
$results = $q->getResult(AbstractQuery::HYDRATE_ARRAY); // commenting this line resolve the memory leak
$this->em->clear();
gc_collect_cycles(); // with or without does not change the issue
}
the consecutive traces shows that memory does not reduce after clear()
Comment created by @ocramius:
[~pourquoi] please check this in insulation (without Symfony2 if possible)
Comment created by @beberlei:
this may be array hydrator related, not sure that may not cause problems.
Comment created by pourquoi:
without symfony:
<?php
use Doctrine\ORM\Tools\Setup;
use Doctrine\ORM\EntityManager;
$loader = require*once __DIR_*.'/../app/autoload.php';
$loader->add('Acme\\CoreBundle', *_DIR_*.'/../src/Acme/CoreBundle/');
$isDevMode = true;
$config = Setup::createAnnotationMetadataConfiguration(array(*_DIR_*."/../src/Acme/CoreBundle/Entity"), $isDevMode, null, null, false);
$conn = array(
'driver' => 'pdo_mysql',
'host' => 'localhost',
'dbname' => 'dbname',
'user' => 'root',
'password' => ''
);
$em = EntityManager::create($conn, $config);
$d = new \DateTime();
while(true) {
echo memory*get_usage() . PHP*EOL;
$qb = $em->createQueryBuilder()
->select('c')
->from('Acme\\CoreBundle\\Entity\\Consultation', 'c')
->where('c.date > :date')->setParameter(':date', $d)
->orderBy('c.date', 'ASC');
$q = $qb->getQuery();
$results = $q->getResult();
foreach($results as $c) {
echo $c->getDate()->format('H:i:s') . PHP_EOL;
}
$q->free();
$em->clear();
gc*collect*cycles();
echo memory*get_usage() . PHP_EOL . PHP*EOL;
}
output:
7978568
7978568
7978568
7978568
7978568
11:51:27
11474520
11474520
11473368
11473368
11473368
11473368
11473368
11473368
11473368
Am I missing something?
Comment created by @ocramius:
Memory usage here seems quite constant (the change from 7978568 to 11474520 may well be because of metadata and hydrators). The output doesn't seem to be conforming your snippet though.
Comment created by pourquoi:
yes sorry the above output is for:
foreach($results as $c) {
echo $c->getDate()->format('H:i:s') . PHP_EOL;
$d = $c->getDate();
}
with
foreach($results as $c) {
echo $c->getDate()->format('H:i:s') . PHP_EOL;
}
the output is:
3489864
12:22:27
13502680
13502680
12:22:27
13515496
13515496
12:22:27
13528328
13528328
12:22:27
13541144
13541144
12:22:27
13553976
....
74513560
12:22:27
74526520
74526520
12:22:27
74539520
74539520
12:22:27
74552560
and goes on
Comment created by stof:
Do you have bidirectional relations in your user entity ? If yes, you will still have some references to the object after clearing the EntityManager (in the related object, itself reference by the user)
Comment created by brunolemos:
Fixed it by doing the following:
$this
->_em
->getConnection()
->getConfiguration()
->setSQLLogger(null);
Issue was closed with resolution "Incomplete"
Most helpful comment
Comment created by brunolemos:
Fixed it by doing the following:
$this
->_em
->getConnection()
->getConfiguration()
->setSQLLogger(null);