Skip to content
This repository

HTTPS clone URL

Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP

A hackish module to avoid many cache wipes/flushes/clears in Drupal.

branch: master

Fetching latest commit…

Octocat-spinner-32-eaf2f5

Cannot retrieve the latest commit at this time

Octocat-spinner-32 README.md
Octocat-spinner-32 cache_delayed.inc
Octocat-spinner-32 cache_delayed.info
Octocat-spinner-32 cache_delayed.install
Octocat-spinner-32 cache_delayed.module
Octocat-spinner-32 cache_delayed.override_all.diff
README.md

Delayed Cache module

Provided by Dutch Open Projects Developed by Bèr Kessels

About

A simple module that allows certain cache_clear_all() calls to be delayed. Requires core hacks. only for certain For experienced Drupaleers only :) (it has severe side-effects, such as a much better working cache, or a broken site).

Cases:

  • A site with commenting only for authenticated users will wipe cache for anonymous users every-time a comment is saved.
  • A site where you want to enforce cache for a longer period, despite you (or other authenticated users) writing articles.
  • Any other site where you have problems building up a proper cache for blocks, pages and such, because it gets flushed by several (core) actions.

Installation and usage:

  1. Install the module in the regular way. It will not do anything yet.
  2. Include the cache_delayed.inc file in settings.php (needed to have functions available on lower bootstrap moments) require_once('./sites/all/modules/custom/cache_delayed/cache_delayed.inc');
  3. Create, or think up a token. Secure tokens can be made (on *nix) with the command uuidgen.
  4. Add a token to your settings.php variables. This token grants access for cron. Keep it secret. $conf = array( ...other conf might be set already. 'cache_delayed_token' => 's3cr3t-h3r3', );
  5. Use this token in a cron call. Every time this cron is ran (or more accurate: the URL is called) the queue is walked trough and all caches are cleared. You could set this to run every 10 minutes, for example. curl --silent --compressed http://example.com/cache_delayed_remove/s3cr3t-h3r3
  6. In the places where you want to avoid a cache_clear_all() , replace with cache_delayed_clear_all(). Calling the latter, puts a new to-be-cleared in the queue, which will be actually cleared on the next cron run.

How it works:

Every time a cache_clear_all is invoked, instead of clearing the cache, it will add an entry to a worker queue. When calling a (secured) URL, we walk trough the queue, select those entries "old enough" and wipe them from your cache (and remove them from the queue).

Why it was made :

For a certain site, we saw over 1000 queries to build pages. One quick-fix was to use the cache better. Drupals caching system however, is brokennot very smart. And will flush on every comment posted, node created, poll-voted and so on. We wanted to avoid these wipes, by enforcing the cache to remain un-flushed for a little longer. Even if Drupal is stubborn and wipes it anyway.

The other option was to build the site in a proper framework :) And the last option, is to simply fix that ridiculous amount of queries. But that is the long route.

Core hacks

You have two options. 1. Hack the actual cache_clear_all in core, so that /all/ cache_clear_alls go via the queue. This will have many side-effects. So know what you are doing. 1. Hack certain places where cache_clear_all is called, for example comment_save(), to avoid a full cache-wipe only in that place. Has little side-effects.

Todo

Cherry-picking: Allow URL-parameters to wipe cache with a CID, in one table or with a wildcard only.

Features?

Few. We opt for speed, not flexibility.

Gotcha's

We insert, delete and update with delayed and low_priority. Thus only works on databases that support that. And the resulting status can most often not be used at all. After all: mysql may wait with inserting a broken or duplicate record, but never report this brokenness or duplicateness back to our code. Keep an eye on the logs. Use the Logs Luke. Use the Logs.

Something went wrong with that request. Please try again.