Support » Plugin: Redis Object Cache » Question about wp-admin concurency

  • Resolved Presslabs


    Hello Till, my name is Robert and I’m facing a strange problem and I’d appreciate any insight or ideas on the matter.

    I’ve been using Memcache for some time but I’ve experienced inconsistencies when coming to save widgets / activate plugins in WordPress, inconsistencies as the DB being updated but the cache not.

    I’ve dropped Memcache since the problem seemed to arise more frequently than acceptable, so I’ve switched to Redis (2.8.x for now, considering an upgrade to 3.0) and stumbled upon your plugin.

    In the tests I’ve run the improvements are huge (10 ms of exec time in Redis down from 200+), but there is a catch the issue still seems to be present but only under heavy load. At this point I’m thinking that it’s not the caching mechanism (also tried APCU), but due to my nature I’ve tried once more with the version your plugin is forked from. Curiously enough I couldn’t reproduce the issue in that version but as always there is a catch, the execution time was more than 5x more than your fork.

    At least now I’ve got something to go on, I’ve made a diff of the 2 object-cache.php implementations and I couldn’t see something that would cause the issue to appear (I’ve checked and both versions are using the PECL package).

    One thing to note is that after removing the internal cache from this version and forcing the plugin to retrieve data directly from Redis fixed the issue, but increased the execution time even on a local Redis server.

    This got me thinking what could have changed between the 2 versions related to how they handle the internal cache, all I see are inverted if statements nothing out of place. Maybe I’m mistaken and the problem might not be here at all.

    My question is where should I look next or what may have caused this issue to arise in the first place.

    What I think is happening is that a process A reads / stores in the internal cache, another process B reads/ stores in the internal cache then does an update / store while A still is running with stale data and if that data is written it may cause these inconsistencies.

    Any thoughts are welcome, I myself have ran out of ideas.
    Thanks for taking your time to read this post :D.

Viewing 1 replies (of 1 total)
  • Plugin Author Till Krüss


    I’m not sure if I can help you with that issue, I’d recommend to consult an expert on scaling WordPress on this.

    I personally would probably see how HHVM performs. Are you sure my plugin is using the PECL extension and not Predis?

Viewing 1 replies (of 1 total)
  • The topic ‘Question about wp-admin concurency’ is closed to new replies.