Segmentation Fault in Phalcon 1.3.x

[klaussilveira]

I have an intermittent problem with Phalcon 1.3.x, running on several different stack combinations:

• PHP 5.5.9, PHP 5.4.4
• Nginx 1.4.6, Nginx 1.6.0

Nginx gives me the following:

[error] 5467#0: *38 recv() failed (104: Connection reset by peer) while reading response header from upstream, client: 127.0.0.1, server: bla.com, request: "GET /my-route HTTP/1.1", upstream: "fastcgi://unix:/var/run/api.sock:", host: "bla.com"

I have attached GDB to the PHP-FPM pool, and got this backtrace on the segfault:

#0  0x000000000070c479 in gc_zval_possible_root ()
#1  0x00000000006fadb8 in zend_hash_destroy ()
#2  0x00000000006ec22b in _zval_dtor_func ()
#3  0x00000000006dd300 in _zval_ptr_dtor ()
#4  0x00000000006f9775 in ?? ()
#5  0x00000000006faf78 in zend_hash_graceful_reverse_destroy ()
#6  0x00000000006dd9e6 in shutdown_executor ()
#7  0x00000000006ed1e2 in zend_deactivate ()
#8  0x000000000068d62d in php_request_shutdown ()
#9  0x0000000000463524 in main ()

I'm really sorry, but i can't provide any further details on this. I have tried to isolate the problem, but the thing is: it's happening on almost all routes, in a 1/10 proportion. Sometimes it segfaults, sometimes it does not. That trace back there, i got it after making the very same request 3 times. The 4th time i got the segfault.

I have analyzed the process with Valgrind and got something like 20k+ uninitialized variable warnings and many leak errors. I have compiled both PHP and Phalcon in debug mode so i could make my life easier in stepping the code.

Want to back this issue? Post a bounty on it! We accept bounties via Bountysource.

[klaussilveira]

I also have noticed that your test suite is pretty unstable. Take this, for example: https://travis-ci.org/phalcon/cphalcon/builds/26888915

It's a simple CHANGELOG fix, but it broke the whole thing on random PHP versions. This happens on other commits as well. Most are mysterious segfaults.

[dreamsxin]

Could you tell me this request execution which operations? Or use which Phalcon components, you can rule out under which components are out of the question.

[pgasiorowski]

@klaussilveira do you, by any chance, use Memcache/Libmemcached session adapter and/or Session bags in your app?

I've managed to isolate current segfaults a little, which led me to SessionTest and SessionBagTests.
Tracking this down further I left:

• SessionTest::testSessionMemcache(), testSessionLibmemcached()
• SessionBagTest::testGetSet()

It's interesting that running these two separately does not segfault, but with one after another I can repro this.

Here's full gdb:
https://travis-ci.org/WooDzu/cphalcon/jobs/28132592

My guess is that there is something wrong with how sessions are destroyed.

[bezl]

Hello! I have the same problem with Phalcon - stable 1.3.2. I tried with several combinations on my Mac:

• Nginx 1.4.6, Nginx 1.6.0 + php-fpm (php 5.4.30, php 5.3.28)
• Nginx 1.6.0 + Apache 2.2.26 + php5_module (php 5.3.28)
• php54 web server + php 5.4.30

but got crashes with 502 Bad Gateway.

After Phalcon\Http\Response::send() I check Phalcon\Http\Response::isSent() and successfully log all Phalcon\Http\Response::getContent() with headers.

But sometimes with .core, sometimes without it I have the following diagnostic reports:

https://github.com/bezl/api/blob/master/php-fpm_2014-07-04-134157_BezlMac.crash
https://github.com/bezl/api/blob/master/httpd_2014-07-04-163237_BezlMac.crash

I don't use sessions in any case.

[bezl]

Hello again! The problem doesn't appear on Phalcon v.1.2.6.
I checked with current stable 1.3.2 on FreeBSD 9.2 and it was also failed. So I decided to downgrade Phalcon to v1.2.6 and got stable positive result.

[cdekok]

We where having problems too with segfault when upgrading from 1.2.4 to 1.3.2 :(

[swen100]

At me too; downgraded from 1.3.2 to 1.2.6 and the error went away. But now i dont have the full list of methods which are available at 1.3.x :-(

[bezl]

The only method I removed was Phalcon\Http\Response::getHeaders()->toArray() but for me it was not critical. Also I believe that all new methods from 1.3.x can be written with 1.2.6 and logic from 1.3.x source code. But I don't know how to tell guys from Phalcon about current issue with stable segfault... I can't freeze Phalcon version for my project development with v1.2.6.

[dreamsxin]

@phalcon
I use the machine for installation test, the system environment:
Machine A:
PHP 5.4.30
Phalcon1.3.2

Machine B:
PHP 5.3.10 upgrade to PHP 5.4.30
Phalcon1.3.2

Machine C:
PHP 5.3.10 upgrade to PHP 5.4.30
Phalcon1.3.2

Machine D:
PHP 5.4.30
Phalcon1.3.2

The machine B、C is happen error, A、D is fine. The upgrade is cover installation.

The following code will appear this error
Code example 1:

$session->set('data', 1);
$apps = Robots::findFirst(array(
             'conditions' => 'id = :id:'
             'bind' => array('id' => $session->get('data'))
 ));

Code example 2:

$json = json_encode(array(1, 2, 3));
$arr = json_decode($json);
$builder->from('Robots')->inWhere('id', $arr)->getQuery()->executue();

My guess is that PHP array operation when happen this problem.