Occasionally, about 1 out of 5 times, a node's cache isn't cleared when it is updated. Even going to view the revision that the update created or editing the node doesn't show the changes. Minimum cache lifetime is set to 1 minute but even after waiting for a minute and running cron it isn't automatically cleared.
It seems that this problem happens more during the site's busier times and we haven't been able to reproduce the issue in our development environment (which typically only has 1 user on it at a time). The cache tables are all InnoDB so I'm told this means we shouldn't need to worry about locking.
A similar problem with the chosen answer saying that this is just an issue of properly setting the minimum cache lifetime. But, I found this Drupal bug that implies that on node updates the cache should be cleared regardless of that setting. Though, its status is still "needs work" which maybe means a solution was never implemented and this is still the case.
Update
Building off a patch in the bug and using hook_node_update I tried the following
function MYMODULE_node_update($node)
{
// Forcibly wipe this node's page view from the cache because, if a
// minimum cache lifetime is in effect, cache_clear_all() will not
// clear any page cache entries younger than that minimum.
//
// Page cache cids are absolute URLs. url() in absolute-mode will
// get the aliased path if there is one, and we also wipe the
// absolute version of the node/nid URL just in case.
// Copied from http://drupal.org/node/256416#comment-838118
global $base_url;
cache_clear_all( url('node/'.$node->nid, array('absolute' => true))
, 'cache_page' );
cache_clear_all($base_url .'/node/'. $node->nid, 'cache_page');
}
This should work around the minimum cache limit and only clear the cache for that node. Unfortunately, this did not work.
Update 2
After digging further I discovered that running cache_clear_all('field:node:<nid>', 'cache_field') (where <nid> is replaced with the relevant node ID) cleared the correct cache when the problem cropped up. I tried putting this in a MYMODULE_node_update() hook but it did not resolve the issue.
In discovering this and following the path of the code I also discovered that the APC module is taking care of the caching and none of this is being stored in the database as I originally thought.
Knowing that the caching is being handled by APC and considering the intermittent nature of the issue it brings me back to a possible locking issue with APC. Would that even make sense? Preliminary searches have not revealed anything.
