WinCache ACM backend

[ChrisTX]

Hello,

I'm not exactly sure where to post this, but I made a ACM backend for Windows Cache for PHP (wincache) and wanted to share it. I wasn't able to test this on anything bigger than my own newly-setup forum, but it seems to be loaded fine, and I couldn't spot any issues with it so far. Since WinCache works pretty similar to the APC/XCache, I don't expect there to be any either.

Hoping it might be useful for somebody.

Chris

[bantu]

Hello,

I think this could even go into 3.0.9-RC1 if it is working.

Thanks
- bantu

[ChrisTX]

Hello,

I think this could even go into 3.0.9-RC1 if it is working.

That'd be great.
Is there any way I can test it more deeply than just being able to tell that my own forum can run it without any errors? I took a look into the tests at phpBB's git, but I couldn't see any ACM-specific tests there. But as I said, WinCache works very similar to APC (of which's acm I also forked this), the only real difference being that wincache_ucache_get() ( apc_fetch()'s equivalent ) returns an empty array on failure, and not FALSE, thus I implemented _isset ( wincache_ucache_exists() ) as well.

- Chris

[ToonArmy]

I have an implementation of this sat in my repository, I've just never got around to testing it. https://github.com/cs278/phpbb3/compare ... ncache-acm

[ChrisTX]

I have an implementation of this sat in my repository, I've just never got around to testing it. https://github.com/cs278/phpbb3/compare ... ncache-acm

Our implementations are pretty much the same, only you made _read return false on failure, like APC would do. I don't think however that you need to implement _isset in that case ( at least judging by how I understood acm_memory's interna, might be wrong ).

EDIT: Actually modifying _read should be better than implementing _isset, considering it won't require WinCache to lookup the $key value twice. I updated my pull request accordingly. (also tested this patch, it works fine, too.)

EDIT2: I missed the most obvious way of testing the ACM. By using the wincache.php that comes with WinCache, you can view statistics on the cache usage.

[igorw]

Here is the diff between the implementations. a is by ChrisTX, b is by ToonArmy.

Code: Select all

$ diff -u a.txt b.txt 
--- a.txt	2010-11-24 15:02:10.000000000 +0100
+++ b.txt	2010-11-24 15:02:16.000000000 +0100
@@ -3,7 +3,7 @@
 *
 * @package acm
 * @version $Id$
-* @copyright (c) 2010 phpBB Group
+* @copyright (c) 2005, 2009, 2010 phpBB Group
 * @license http://opensource.org/licenses/gpl-license.php GNU Public License
 *
 */
@@ -23,13 +23,18 @@
 }
 
 /**
-* ACM for WinCache
+* ACM for Wincache
 * @package acm
 */
 class acm extends acm_memory
 {
 	var $extension = 'wincache';
 
+	function acm()
+	{
+		parent::acm_memory();
+	}
+
 	/**
 	* Purge cache data
 	*
@@ -37,9 +42,10 @@
 	*/
 	function purge()
 	{
-		wincache_ucache_clear();
-
+		// Run before for XCache, if admin functions are disabled it will terminate execution
 		parent::purge();
+
+		wincache_ucache_clear();
 	}
 
 	/**
@@ -52,7 +58,9 @@
 	function _read($var)
 	{
 		$success = false;
+	
 		$result = wincache_ucache_get($this->key_prefix . $var, $success);
+
 		return ($success) ? $result : false;
 	}
 
@@ -81,6 +89,18 @@
 	{
 		return wincache_ucache_delete($this->key_prefix . $var);
 	}
+
+	/**
+	* Check if a cache var exists
+	*
+	* @access protected
+	* @param string $var Cache key
+	* @return bool True if it exists, otherwise false
+	*/	
+	function _isset($var)
+	{
+		return wincache_ucache_exists($this->key_prefix . $var);
+	}
 }
 
-?>
+?>
\ No newline at end of file

They are almost identical now. Since ToonArmy's appears to be a tiny bit more complete, we might want to just go with that. Some further testing would be great though.

In terms of unit tests I don't think it's possible to test this cleanly in 3.0, because every cache class has the same name. There is a refactoring planned for 3.1 (I've started some work on it: https://github.com/igorw/phpbb3/compare ... m-extended) that will allow to load different cache plugins in order to test them.

[ChrisTX]

As I said, it is less efficient to implement _isset under these conditions, because that will cause serveral doubled queries, i.e. in destroy() of acm_memory, the _exists() function ( which refers to the _isset one ) will be called before _delete() and if it fails - and only then - it will skip the rest. However, wincache_ucache_delete() does the lookup anyway, and if it fails there (= it doesn't exist), FALSE is returned. Similarly, get() would be made less efficient the same way. I didn't implement _isset() on purpose, or actually, I first had it, but then realised it being like this. Implementing _read with a return of the success status ( opposed to the empty array it would return otherwise ), is more efficient, since the actual lookup has only to be made once, and not in wincache_ucache_exists() and the function you call then.

Considering testing, it would only be necessary to verify that the actual ACM code has no issues. When implementing that reflected change yesterday, I was a bit distracted and made this typo, which would have led to make wincache's stats show properly ( since the request was made) but then to not actually use the cache. The rest (deletions, requests, writing, purging) is done properly, which can be verified - at least for wincache, for other systems I do not know - using the stats site.

[igorw]

Okay, I see. What about the "Run before for XCache, if admin functions are disabled it will terminate execution" comment in purge() though? I'm not sure what Chris (ToonArmy) meant exactly, since it is kind of vague. But that may still need adjusting.

[ChrisTX]

I do, he implemented the ACM similary to me, that is by branching an existing backend. I started off acm_apc, he started off acm_xcache. If you'd look inside the file, you can also see the comment there, it's a left-over he forgot to remove. Same goes to the call to parent::acm_memory() in the constructor, in acm_xcache, that has a purpose; however, here it's not needed.

That's also the reason why the order of our parent::purge / wincache_ucache_clear(); calls in the purge is inverted, acm_apc and acm_xcache do it differently ( though it doesn't matter for wincache ).

[igorw]

I am really failing today, it seems.