On Sat, 5 Nov 2016 10:44:44 +0100, Alessandro Pasotti <[email protected]> wrote:
Sorry to butt in, a few days ago I stumbled with a resource that can improve significatively the performance of the server, at least if it runs on Linux: http://www.memcached.org/ Perhaps it deserves a try. HTH. > On Sat, Nov 5, 2016 at 10:27 AM, Richard Duivenvoorde > <[email protected]> wrote: > > > On 05-11-16 10:11, Alessandro Pasotti wrote: > > > Hi > > > > > > I've been monitoring the server and I think that with the caching > > > in place the timeouts are gone, but of course we need to add the > > > new versions as they appear. > > > > > > Richard: what do you think about adding a couple of future > > > versions too, so we don't forget? > > > > Yep, done: 3.0 3.1 3.2 3.3 added > > > > Currently the nightly requests the ?qgis=3.0 version (see [0]) > > > > by the way, running the get_plugins.xml script takes 2,5 minute > > $ time ./get_plugins_xml > > real 2m27.058s > > user 0m0.040s > > sys 0m0.096s > > And runs every 10 minutes. Maybe we should bring that up in future, > > or make the script faster? > > > > > > I've already spent quite a lot of time trying to make it faster and > it's now indeed much faster: a single run (for a single QGIS version) > is ~3 seconds depending on the server load (it was sometimes over 10 > seconds before my latest changes!). > What seemed a simple addition (the trusted flag) came out to be a > performance killer, because the user "trusted" information is a > couple of tables away: > plugin_version -> plugin -> author -> group -> permission > > We are now using raw queries: > https://github.com/qgis/QGIS-Django/commit/3ffd9c3d989ce144a2cf9d7f02ea2b0dc8f5b5ad > > The problem is also that on that server there are a lot of other > processes going on (doc builds for example) that eat resources. > > I don't think we can achieve much better results without some deep > refactoring of the plugins application (adding triggers to set the > trusted flag on the plugin version row could be a good starting point) > > > BTW, I'd lower the frequency of the caching script: I don't think that > waiting 30 minutes or more to see a plugin appear in the XML will kill > anyone. > > Roxo -- ---------------- Non luctari, ludare -------------------+ WYSIWYG Fernando M. Roxo da Motta <[email protected]> | Editor? Except where explicitly stated I speak on my own behalf.| VI !! ( Usuário Linux registrado #39505 ) | I see text, ------------ Quis custodiet ipsos custodes?-------------+ I get text! _______________________________________________ Qgis-user mailing list [email protected] List info: http://lists.osgeo.org/mailman/listinfo/qgis-user Unsubscribe: http://lists.osgeo.org/mailman/listinfo/qgis-user
