Improved caching with eAccelerator/turck mmcache

+1 in principle... however, rather than see this get committed directly, I'd rather see a "caching framework" evolve...

we've also had other talks about making drupal more squid friendly, in addition to using memcached, etc.

i'm willing to help put some brainpower into this...

- The variable names are not consistent with our coding style (no camel casing).

- The code won't work against HEAD.

- This needs to be benchmarked before it can be committed. I don't think it will buy us much.

- I agree with James that the squid changes might be worth looking at too (forgot the URL).

Quite the contrary. I'd like to see this patch evolve and work on this together.

Please keep in mind that your patch is one of many others in the queue... we can't shower everyone who submits one with praise, it is not a good use of our limited time. And in spite of you not submitting your code in the most ideal form possible (a diff), you had 4 replies already. Your issue hasn't even been up for a whole day. Most of us do this in our spare time, so this is a succesful response.

The fact is, Drupal has high standards to keep up. If people point out how your patch does not conform to those, it means they would probably accept it if it did. It is a good sign, trust me.

Your change automatically uses eaccelerator/mmcache if they are present, without providing an option to the administrator for this. Is this a reasonable thing to do? I'm not sure how widespread they are, but it is these sort of things which can cause sites to suddenly break when the host's configuration changes.

Now as far as the patch goes:

• You still call db_encode_blob() and db_decode_blob() even when using the alternate storage mechanism. This is a specific form of encoding suited for the current database type, so it is unnecessary.
• All data passed to cache_set must be a string already, so serialization is unnecessary.
• You convert CACHE_TEMPORARY to 1 (second?). Is there a reason for using this and not 0 (that would seem more logical to me).

The code does not conform to the code style in much more ways than just one name. It looks messy and unorganised.

• You do the check for 'function_exists' twice. You can easily re-arrange things to avoid this.
• The if statements for checking the type of $expire passed would look much cleaner expressed as a switch statement.
• We put spaces after control flow statements (if, switch, while, etc).
• We put spaces around the concatenation operator, except between a space and a quote. This is a rule which many forget, but it strikes the perfect balance between keeping code readable, and keeping code compact.
• We prefer single quoted strings to double quoted strings, except when it would reduce readability (e.g. when there are single quotes inside).
• We prefer proper typography for comments (ending with a period if it is a full sentence) and a space between the comment's marker and its contents.
• Having a clean code style keeps the code easy to read, easy to scan visually and easy to search programmatically.

By respecting these ideas, any future contributions you make will have a higher chance of being accepted. Please take this as constructive criticism, not a sign that we dislike you or your contribution.

First, don't think that Dries does not want you to contribute, but this has been discussed already.

However, I have some concerns. Natrak installed eAccelerator on Drupal at one point and it segfaulted Apache quite often. Natrak, if you read this please comment!

chx: I have been using PHP5 + eAccelerator 0.9.2a (and just upgraded to 0.9.3) for a few months on a couple of my Drupal sites without any trouble. My sites certainly aren't as large or as high-traffic as Drupal.org, though. (I'm not sure whether it matters, but I'm using lighttpd instead of Apache.) Anyway, it's possible that a lot of eAccelerator bugs have been fixed recently.

eAccelerator has problems with sites that have a high load and at some point will screw up on randon pieces of PHP code and cause segfaults till apache is restarted. From what I last heard they are making progress and probably has something to do with locking and eAccelerator not being fully thread safe. Zend Optimizer seems to work without problems, so using that till something better comes up.

Also just quickly looking at this code wouldn't the cache size theoretically be small than when using MySQL? Admins can set a limit on shared memory storage, from the docs on eaccelerator_put it seems that once its full you can't store any new data (function returns false). On Drupal.org I usually see the cache table to be anywhere from 20 to 80 megs, not sure what the default values of eAccelerator are.

KJ - true, but we're seeing much better behaviour out of eaccel 0.9.3 ..

Yerlec - please don't be put off by such comments, call it "constructive criticism". I'm interested in working on this as well, but agree that making this eaccel / turck specific is probably less desirable as there are several different options for caching , and ideally we'd like a framework to accomodate as many as possible.

Yrlec:

Dries was being constructive. By posting his comments, he was implying that he wanted you to continue working on this patch and gave you concrete suggestions that would help your patch get integrated into core. One of the reasons Drupal's code is clean and consistent quality control from the core team. Instead of having the coding style of hundreds of developers, there's one consistent code style for all of core.

In other words, let's get this cleaned up and into core!

dummy comment to subscribe to this.

We have a new cache system in the HEAD.
Probably this will work as an "alternative cache option" in this context.
What do you think?

we now have pluggable cache.inc files in D5. this can and should be a contrib module. perhaps announce the module here if you create it.