Re: Re: APC-3.0.10 issue
| From: | Sara Golemon | Date: | Tue, 23 May 2006 18:32:35 +0000 |
| Subject: | Re: Re: APC-3.0.10 issue | ||
| References: | 1 2 3 4 5 | Groups: | php.pecl.dev |
| Request: | Send a blank email to pecl-dev+get-3798@lists.php.net to get a copy of this message | ||
> Just to remove our application from the problem
> I am re-producing the issue with a simple test script.
>
Okay, well, the opcodes I'm generating on this side match what the gdb
results your producing in every way that matters so whether it's a cache
miss and APC is recompiling the script or it's a cache hit and it's pulling
it out of shared memory, all signs point to "it should just work". So
rather than pound my head into the desk any longer, I'm going to try taking
a wild shot in the dark.
The backtrace seems to be pretty adamant about there being a write violation
when trying to copy that constant value into the result node so that says to
me that the storage space for that result value is unusable for some reason.
The one thing quirky about this storage space is that it's on the stack and
there *could* be a celestial misalignment between fastcgi, apc, and your
webserver when it comes to allocating dynamic chunks of memory on the stack.
What I'd like you to do is rebuild your PHP install with that feature
disabled:
1. cd /path/to/phpsrc/
2. edit Zend/zend_execute.c changing the following line (should be at or
around line 56):
#define TEMP_VAR_STACK_LIMIT 2000
to a value of zero
#define TEMP_VAR_STACK_LIMIT 0
3. make all install
Functionally, this core won't be any different from the one you've got now,
but it'll allocate those temp vars on the heap rather than the stack. I
admit I'm reaching here, but since I can't get this issue to reproduce
locally I'm gonna have to poke the black box until something falls out...
By the way, do you have apc.optimization turned on? If so, try turning it
off and see if that makes a difference... (again, grasping at straws)
-Sara