On Thu, Oct 25, 2012 at 10:00 AM, Johannes Weberhofer <jweberhofer(a)weberhofer.at> wrote: http://www.mediawiki.org/wiki/Manual:How_to_debug#Profiling Enable profiling and it will break down timings on what is taking how long.

On Thu, Oct 25, 2012 at 10:24 AM, Johannes Weberhofer <jweberhofer(a)weberhofer.at> wrote: Forget to attach it or did mailman strip it off? Should be able to put it on a pastebin somewhere.

On Thu, Oct 25, 2012 at 10:52 AM, Johannes Weberhofer <jweberhofer(a)weberhofer.at> wrote: Do you have a memcached server defined that isn't actually there or is pointing to the wrong place? That's about the only thing I can think of given all the time is spent in MessageCache::load

On 25/10/12 18:18, Johannes Weberhofer wrote: That's odd. If you provided CACHE_ACCEL and you don't really have a cache accelerator which can be used, you should be getting an exception. So not something that "just slows things" (in old versions I think it was indeed silently skipped). Do you have any accelerator installed? Maybe you have one with very little space available and it was spending the time with many misses?

On Thu, Oct 25, 2012 at 6:18 PM, Johannes Weberhofer < jweberhofer(a)weberhofer.at&gt; wrote: Your log file: CACHES: XCacheBagOStuff[main] XCacheBagOStuff[message] XCacheBagOStuff[parser] [...] MessageCache::load: Loading en... cache is empty, loading from database, loading FAILED - cache is disabled [...] MessageCache::load 1 10002.965 10002.965 98.312% 1670 ( 10002.965 - 10002.965) [1] Shouldn't it fallback to the DB if XCache is not found? Were there changes with respect to this in 1.20? Bryan

On 26 October 2012 01:07, Platonides &lt;Platonides(a)gmail.com&gt; wrote: It tries to acquire the lock until MGS_WAIT_TIMEOUT has passed, which is 10 seconds. The lock is there to prevent multiple threads building the cache concurrently. For reasons unknown to me, there is another lock inside ->lock() ->unlock() which again fails and disables message cache. -Niklas -- Niklas Laxström

On 26/10/12 07:48, Niklas Laxström wrote: However, XCache must be working for it to wait the WAIT_TIMEOUT. If there wasn't a key, the mMemc->add would have been successful (XCacheBagOStuff::set always returns true). Maybe Johannes has XCache installed and working, but it doesn't have enough space to store the message cache, so it is always a miss, and his threads are constantly racing to rebuild it.