[5.6] Potential race condition with model factory after callbacks

[Indemnity83]

• Laravel Version: 5.6.12
• PHP Version: 7.1.14
• Database Driver & Version: sqlite in memory (testing)

Description:

It looks like I introduced an unexpected bug/race condition with the behavior of after callbacks in PR #23495. In that the model returned can become out of sync with the database state that was created inside the factory()->create() call.

Steps To Reproduce:

Assuming you have users and teams:

class User extends Model
{
    public function team()
    {
        return $this->belongsToMany(Team::class);
    }
}

class Team extends Model
{
    public function users()
    {
        return $this->belongsToMany(User::class);
    }
}

Example after callback for a user, who must be a member of a team to be a valid user:

$factory->define(App\User::class, function (Faker $faker) {
    return [
        // valid attributes ...
    ];
});

$factory->afterCreating(\App\User::class, function (\App\User $user, Faker $faker) {
    $team = factory(\App\Team::class)->create();
    $team->users()->attach($user);
});

In a test, the first test assertion will unexpectedly fail sometimes, but the second (after calling fresh()) will always pass.

$user = factory(User::class)->create();

$this->assertCount(1, count($user->teams));
$this->assertCount(1, count($user->fresh()->teams));

[Indemnity83]

False alarm, after simplifying my particular code for the example above it looks like its working, must just be a bug in my slightly more complex code.

[Indemnity83]

Re-opening ... looks like this might be a race condition. I'm getting varied results even when just re-running the same tests repeatedly.

I'm not sure if I can put together a test that is 100% reproducible though. Does this make sense to others that a race could be happening here?

I'm running tests with in-memory sqlite.

[unstoppablecarl]

It might be a sqlite quirk can you try mysql?

[ChrGriffin]

I've just encountered an issue myself that seems very similar.

Consider the following test:

$language = factory(Language::class)->create();
$story = factory(Story::class)->create();
factory(Story::class)->create();
factory(Clip::class)->create([
    'story_id' => $story->id,
    'language_id' => $language->id]
);

$stories = Story::notEmpty()->get();
$this->assertFalse($stories->isEmpty());
foreach($stories as $story) {
    $this->assertFalse($story->clips->isEmpty());
}

The intent of the test is fairly self-evident. I've written a custom scope called scopeNotEmpty(), that only returns Stories (Categories) that have media clips attached.

public function scopeNotEmpty($query)
{
    return $query->whereHas('clips');
}

This test fails or passes randomly, on the assertion: $this->assertFalse($stories->isEmpty());. Randomly and without apparent cause, sometimes there are Stories in the database, and sometimes there aren't.

I'm running my tests on an in-memory SQLite database, just like @Indemnity83. Notably, I have an entire test suite that runs predictably. This is the first time I've encountered this behaviour. Unlike @Indemnity83, I do not have an existing entity that I can call fresh() on to get a fresh copy out of the database.

PS: As a truly blunt debug, I tried adding sleep(5) before the assertion, in case it was somehow a race condition with SQLite not writing the data fast enough. No change, still randomly fails or succeeds.

PPS: Despite the random behaviour of my custom scope, Story::all() returns consistent, predictable results. So the problem must be in my scope, somehow (that one line doesn't seem to leave a lot of room for error). I'm not ruling out that it's unrelated to a race condition in the factories, since the Clip, which is central to the concept of the scope, is created in a factory.

PPPS: After switching from using the factory to just using Clip::create(), the behaviour was immediately solved. Consistent, predictable results every time. I have to conclude the issue is with using the Factory.

[cbj4074]

@tillkruss Why was this closed?

[edwinvdpol]

Same problem here. I have to refresh() the model after creating and afterCreating is ran.
When I do a GET request afterwards without the refresh, it works.
When I do a PUT request afterwards without the refresh, it does not work.
When I do a PUT request afterwards with the refresh, it does work.

[TwilightDuck]

I know the issue is closed, but I still wanted to mention that I'm getting this behaviour now.