lighty's life: One more opcache for php (Preview)

One more opcache for php (Preview)

moo — Tue, 04 Apr 2006 04:36:00 +0000

I'm sure you use lighttpd because of performance and scalability, and many of you run php too. I'm please to introduce you another opcode cache for php.

Yes, another.

The preview version is deprecated. please check http://trac.lighttpd.net/xcache/wiki/GettingSource to get newer source.

and report problems at http://trac.lighttpd.net/xcache/newticket

Introducing

I myself, have used apc, mmcache/ea, ZendOptimizer, The only 3 choices before xcache. ab cache is too old and dead, saidly, before i use it. Every one of u write php script know them, and have one if them boost your php performance, because they're good.

But "What? Why you just introduce another opcode cacher?" It take some time to explain ... the short term: it give you one more choice. the long term: mmcache/ea is the most optimized php-opcache. ZendOptimizer isn't pre-built on all platform -- they release binary not source file. apc was stable on php4 but not works with php5, and now, is said works with php5 but php4 is left to desert...

Highlight spot of xcache

cacher, similar as other opcaches.
opcode disassembler, to see how your scripts looks after they're compiled into opcode.
grows as php grows, using automated discover scheme by the devel, and the new changes to xcache to catch up new php version is brought to you easily.
php4.x lastest is stably supported, php5.x lastest just works, php6-devel is supported for your further vision.
many more ...

New function for script

$ grep proto *.c

mixed xcache_get(string name)
bool  xcache_set(string name, mixed value [, int ttl])
mixed xcache_isset(string name)
bool  xcache_unset(string name)
int   xcache_inc(string name [, int value [, int ttl]])
int   xcache_dec(string name [, int value [, int ttl]])

How to install

$ wget http://blog.lighttpd.net/files/xcache-preview.tar.gz
$ tar -zxf xcache-preview.tar.gz
$ cd xcache-preview
$ phpize
$ cat ./conf
$ ./configure --help
$ ./configure --enable-xcache-disassembler
$ make
$ su
# make install

take care about the output of make install append the following to php.ini:

[xcache]
xcache.size = 64M
xcache.mmap_path = "/tmp/xcache"
xcache.cacher = 1
zend_extension="/usr/.../extensions/..../xcache.so"
# or
extension="xcache.so"

Testing

restart your php (fastcgi or mod_php), check output for xcache status. if you see it there, you're done.

and finally test the opcode dumper:

$ ./phpdop.phpr ./phpdop.phpr
$ ./phpdc.phpr ./phpdc.phpr

Patch for hardened source (not tested)

Index: processor.m4
===================================================================
--- processor.m4    (revision 27)
+++ processor.m4    (working copy)
@@ -575,6 +575,9 @@

    /* reserved */
    DONE(reserved)
+ #if defined(HARDENING_PATCH) && HARDENING_PATCH
+   DISPATCH(zend_bool, created_by_eval)
+ #endif
    } while (0);
 ')
 dnl }}}

it is suggested to modify processor/processor.m4 directly as there might be spaces lost in blog.

"One more opcache for php (Preview)" by mOo

Fri, 19 May 2006 05:30:04 +0000

we have setup trac system at http://trac.lighttpd.net/xcache/, you may report bugs with http://trac.lighttpd.net/xcache/newticket include system version `uname -a`; php version `php -v` or check out phpinfo(), gcc version `gcc -v`

"One more opcache for php (Preview)" by null

Sun, 14 May 2006 16:26:02 +0000

fly# make /bin/sh /usr/xcache-preview/trunk/libtool --mode=compile gcc -I. -I/usr/xcache-preview/trunk -DPHP_ATOM_INC -I/usr/xcache-preview/trunk/include -I/usr/xcache-preview/trunk/main -I/usr/xcache-preview/trunk -I/usr/local/include/php -I/usr/local/include/php/main -I/usr/local/include/php/TSRM -I/usr/local/include/php/Zend -I/usr/local/include/php/ext -DHAVE_CONFIG_H -g -O2 -c /usr/xcache-preview/trunk/xcache.c -o xcache.lo gcc -I. -I/usr/xcache-preview/trunk -DPHP_ATOM_INC -I/usr/xcache-preview/trunk/include -I/usr/xcache-preview/trunk/main -I/usr/xcache-preview/trunk -I/usr/local/include/php -I/usr/local/include/php/main -I/usr/local/include/php/TSRM -I/usr/local/include/php/Zend -I/usr/local/include/php/ext -DHAVE_CONFIG_H -g -O2 -c /usr/xcache-preview/trunk/xcache.c -fPIC -DPIC -o .libs/xcache.o In file included from /usr/xcache-preview/trunk/xcache.c:9: /usr/include/malloc.h:3:2: #error " has been replaced by " In file included from /usr/xcache-preview/trunk/xcache.c:21: /usr/xcache-preview/trunk/align.h:16:1: warning: "ALIGN" redefined In file included from /usr/include/sys/param.h:109, from /usr/local/include/php/main/php.h:206, from /usr/xcache-preview/trunk/xcache.c:12: /usr/include/machine/param.h:79:1: warning: this is the location of the previous definition /usr/xcache-preview/trunk/xcache.c:24:23: processor.h: No such file or directory /usr/xcache-preview/trunk/xcache.c: In function `xc_entry_store_dmz': /usr/xcache-preview/trunk/xcache.c:149: warning: assignment makes pointer from integer without a cast *** Error code 1 Stop in /usr/xcache-preview/trunk.

"One more opcache for php (Preview)" by peterb

Sat, 06 May 2006 21:57:36 +0000

Cannot get to run with NuSphere dbg.so extension with apache2 and php-5.1.4. Error report says DBG extension must be loaded before any other. Does not seem to matter what order I place the extension commands in php.ini. Anybody any experience of this?

"One more opcache for php (Preview)" by moo

Fri, 28 Apr 2006 07:54:44 +0000

i simply don't have enough experience on those compatibility. once the site is up, any patch is welcomed.

"One more opcache for php (Preview)" by wedge

Mon, 24 Apr 2006 15:15:33 +0000

yes, a bsd-compatible version would be very nice.

"One more opcache for php (Preview)" by Jedi

Sat, 22 Apr 2006 15:22:16 +0000

The m4/sed/awk scripts used to create the processor use non-standard GNU extensions. Even the Makefile requies GNU Make. It's why the package is tricky to build on BSD systems.

"One more opcache for php (Preview)" by Rajib Roy

Tue, 18 Apr 2006 16:55:05 +0000

Great man. Working fullproof with php, Thanks for this

"One more opcache for php (Preview)" by mOo

Sun, 16 Apr 2006 05:07:41 +0000

autoconf, phpize, m4 is required. ctags is recomended but optional

"One more opcache for php (Preview)" by ericj

Fri, 14 Apr 2006 21:32:27 +0000

After exec phpize it really made "configure" and then I exec ./configure --enable-xcache-disassembler and make Some error message like this happened. /root/src/xcache-preview/processor.c:1:28: processor_real.c: No such file or directory why? My OS is FreeBSD 6.1 RC#1

"One more opcache for php (Preview)" by ericj

Fri, 14 Apr 2006 21:23:21 +0000

sorry , I saw that ! I should use phpize ..

"One more opcache for php (Preview)" by ericj

Fri, 14 Apr 2006 21:21:16 +0000

there is no "configure" such file in the xcache-preview.tar.gz?

"One more opcache for php (Preview)" by moo

Wed, 12 Apr 2006 04:13:17 +0000

created_by_eval is introduced by harded php patch. and congratulation! you're the first user (not devel) protected by xcache's struct detection :) it just stop instead of making a buggy xcache that is not for your version of php.

"One more opcache for php (Preview)" by foo

Wed, 12 Apr 2006 00:48:20 +0000


$ phpize
$ ./conf
$ make
(...snip...)
==== calc zend_op_array =================
expected:  type function_name scope fn_flags prototype num_args required_num_args arg_info pass_rest_by_reference return_reference refcount opcodes last size vars last_var size_var T brk_cont_array last_brk_cont current_brk_cont try_catch_array last_try_catch static_variables start_op backpatch_count done_pass_two uses_this filename line_start line_end doc_comment doc_comment_len reserved created_by_eval
missing :  created_by_eval
==== store HashTable =================
==== store zval =================
==== store zend_op_array =================
==== restore HashTable =================
==== restore zval =================
==== dprint HashTable =================
==== dprint zval =================
==== dprint zend_op_array =================
==== dasm HashTable =================
==== dasm zval =================
==== dasm zend_op_array =================
==== asm HashTable =================
==== asm zval =================
==== asm zend_op_array =================
make: *** [processor.out] Error 1

System info: Slackware 10.0.0, GCC 3.3.4, PHP 5.1.2

"One more opcache for php (Preview)" by mOo

Tue, 11 Apr 2006 13:08:50 +0000

run phpize to build configure. it is only a preview package. thank u guys for your time trying it out.

"One more opcache for php (Preview)" by refactored.net

Tue, 11 Apr 2006 09:29:27 +0000

Sorry, I'm not very up-to-speed with linux. How would I create the ./configure file? I thought it was a script which ran through the process of switching on/off options for "make"?

"One more opcache for php (Preview)" by moo

Tue, 11 Apr 2006 07:04:41 +0000

./conf is just a example, so cat to read it. the u have to do ./configure yourself with options you decided. according one of the above comments, coverage have to be enabled due to a bug.

"One more opcache for php (Preview)" by refactored.net

Mon, 10 Apr 2006 17:51:11 +0000

Tried to install using instructions above, but with no luck. "cat ./conf" simply echo's to screen. Am I possibly missing a package?

"One more opcache for php (Preview)" by moo

Sun, 09 Apr 2006 13:47:31 +0000

well, apc isn't the real reason i write another opcode cacher. i wrote it before i can't start it up. and i've used apc for a long time, already. and i've contribute many bug reports/patches to apc/ea too. and yes, apc WAS stable for php4. but some idea can't be done by simply writing patches. it might need a whole struct change, sometimes.

"One more opcache for php (Preview)" by -HR

Sun, 09 Apr 2006 10:21:28 +0000

Mmh, if you can't get APC to start on your setup why not file a bug/try to fix it rather that spawning a whole new project? I think this bug has already been reported: http://pecl.php.net/bugs/bug.php?id=7141 A good opcode cache is a hell lot of work to maintain and since Yahoo uses APC I don't see it being left unmaintained so I see no compelling reason for creating a new one. Maybe your work would be more useful if you contributed to APC instead?

"One more opcache for php (Preview)" by moo

Sat, 08 Apr 2006 03:21:27 +0000

always run phpize to get configure file. undefined symbol: xc_coveragedump_dir ... thanks for your feedback :) one of the most important reasons to release (not only the preview version) and setup a site, is to attach more testers, make it more stable. there's too much left not released yet, such as a administration frontend, an coverage reader (read the dump and view in browser), an reallife testcase runner, and document! about all the functions (not just php function) in xcache (for user) and internal doc (for devel who's interested in the source code). i can't get apc start up(warm up) on fastcgi+php4+smp, strace -p `pid of php` gives me bunch of flock/fcntl-lock

"One more opcache for php (Preview)" by User

Wed, 05 Apr 2006 17:56:35 +0000

When loading the extension (compiled exactly as in sample) got such PHP Warning: Unknown(): Unable to load dynamic library '/usr/local/lib/php/extensions/no-debug-non-zts-20020429/xcache.so' - /usr/local/lib/php/extensions/no-debug-non-zts-20020429/xcache.so: undefined symbol: xc_coveragedump_dir in Unknown on line 0 When I compile the ext with --enable-xcache-coverage the module loads fine just segfaults on any code execution.. Any notes on this?

"One more opcache for php (Preview)" by Mogrol

Wed, 05 Apr 2006 10:03:41 +0000

There doesn't seem to be any configure file in the archive, or am I missing something?

"One more opcache for php (Preview)" by foo

Wed, 05 Apr 2006 03:42:36 +0000

APC works fine with both PHP 4 and PHP 5.

"One more opcache for php (Preview)" by refactored.net

Tue, 04 Apr 2006 15:20:56 +0000

Another quick question - does this work with Zend Optimizer?

"One more opcache for php (Preview)" by moo

Tue, 04 Apr 2006 12:20:37 +0000

me, moo, one of the lighttpd developers, wrote it, and have it online stable for a long time until i decide to release it. i'm going to setup a website for it, this is way the preview is here. the reference to http://xcache.sourceforge.net/ was wrong. although i've live with this name for a long time, i might change it if there's a better one.