opcache with Phalcon 1.3.2

[kbtz]

Not sure if it's related with #2239 and #2512, not sure if it's even a Phalcon bug. I'm just providing more details from my context in hope that it lead for some insight for solving this issue if it's related to Phalcon somehow...

My Environment:
Apache/2.2.15 (CentOS)
PHP 5.5.9
Phalcon 1.3.2 777fc83
Zend Engine v2.5.0
Zend OPcache v7.0.3
Xdebug v2.2.4

Reproduction & Errors:
From my testings, I'd say that's the reproduction is to use these extensions I've listed above. The code executes in the first time after a web server restart then fails on 2nd time (using opcached version of the script?), but that's just for some actions, so it may be an error introduced in the bytecode version of the script.

Some controllers actions called via AJAX (GET method) shows in javascript console a specific error code of Google Chrome called net::ERR_INCOMPLETE_CHUNKED_ENCODING. I wasn't able to find any more details of this error code.

Here's a sample of the response header when it occurs:

HTTP/1.1 200 OK
Date: Mon, 09 Jun 2014 19:22:25 GMT
Server: Apache/2.2.15 (CentOS)
X-Powered-By: PHP/5.5.9
Expires: Thu, 19 Nov 1981 08:52:00 GMT
Cache-Control: no-store, no-cache, must-revalidate, post-check=0, pre-check=0
Pragma: no-cache
Connection: close
Transfer-Encoding: chunked
Content-Type: text/html; charset=UTF-8

At server side, I've got some zend_mm_heap corrupted at /var/log/httpd/error_log. I've double checked all my unsets but nothing changed.

Fun facts:

• If a Xdebug breakpoint is set on the problematic php file, the error never occurs. I believe the breakpoint vanishes any cache for that file.
• However, not loading Xdebug didn't change the symptoms.
• By not loading the opcache extension the issue is gone and Xdebug works like a charm.

---

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

[tmatei]

@cvsguimaraes I had that error when I had xdebug on, and that was due to the memory not being enough. If you change your php.ini output_buffering = Off then you will probably get the same error I had in #2512

[kbtz]

@tmatei which error more specifically? net::ERR_INCOMPLETE_CHUNKED_ENCODING or zend_mm_heap corrupted? Both?

[tmatei]

@cvsguimaraes zend_mm_heap corrupted

[oh-ren]

Just curious, do you guys still have these issues? Did you, by any means, have found a workaround, or got it working in the meantime?
I haven't tried to run my app with OPcache enabled yet, but it's kind of a biggie when you have to disable it??
Have you perhaps tried using APC instead of OPcache? (though that's the way backwards I guess, and not really the proper way to go)

So it doesn't seem Xdebug related?

[kbtz]

For now I'm fine without APC and OPCache, so I haven't tried any workaround than disabling these modules.

I believe Xdebug is related but passively, it isn't the source of the symptoms but is affected by them. After disabling OPCache the zend_mm_heap corrupted error was gone without applying the tweak suggested by tmatei.

[tmatei]

I don't know how to debug this any further, the problem I was seeing was due to opcode cache being turned on. Temporary workaround is to turn off opcode cache, but I haven't tried it with APC. This issue may be isolated to linux (centos) environment, but I am not sure. If someone is able to take a look at it, I can try to come up with an example and replicate this.

[maxowar]

I dont believe the problem is Opcache so disabling it isn't a solution (like removing exit/die for me, see #2564) because we still don't know what is the real cause.

What i've seen is that it happens that there are some memory errors, that are then "stored" on opcache that cause the crash of every next request, but it's just a consequence of a previous error.
Disabling Opcache make it works and the error still firing.

@tmatei it's really hard to debug because even gdb exit with error :( what i've seen with strace is that we arrive at the RSHUTDOWN phase...don't know how to help you to help us! :)

[kbtz]

Still not sure about what caused this issue but seems that it's gone anyway.
#2564 seem's to be resolved too.