Part of the gerrit slowness is that the web server is in eqiad and its database is in pmtpa. We noticed some slowness of the interface when moving the gerrit service to a new server. That said, the new server handles the load much better, so overall it was worth the extra latency.
We need to point the gerrit server at a database in eqiad at some point soon. I'll look into this when I get back from vacation. On Sat, Mar 31, 2012 at 4:02 AM, Platonides <[email protected]> wrote: > On 28/03/12 18:10, Antoine Musso wrote: >> Gerrit slowness: >> >> I do not have that much an issues. Possible root causes could be: >> - you have a slow computer >> - your network connection is bad >> - Gerrit client interface is badly programmed >> - Gerrit server is overloaded >> Or that might just be because delay between click is not good enough for >> you. Maybe we could investigate on the server side or add a layer of >> caching in front of Gerrit. >> Anyway, something worth a bug so we could investigate it. > > I also find gerrit delays annoying, so I went to > https://gerrit.wikimedia.org/r/#change,3794 and opened the file diff. It > turns out to be a well suited change for testing, since it's just a > "test 1" -> "test 2". > > The developer console showed: >> [20:41:13.563] GET >> https://gerrit.wikimedia.org/r/gerrit/deferredjs/6B3633B5038AF814D44FB7828C304DAD/1.cache.js >> [HTTP/1.1 304 Not Modified 183ms] >> [20:41:13.867] POST >> https://gerrit.wikimedia.org/r/gerrit/rpc/ChangeDetailService [HTTP/1.1 200 >> OK 966ms] >> [20:41:14.835] POST >> https://gerrit.wikimedia.org/r/gerrit/rpc/PatchDetailService [HTTP/1.1 200 >> OK 655ms] > > That's a bit more than 1.8s, although it feels slower, maybe due to the > later javascript and CSS processing of those messages. > > I suspect the first one avoidable, since it seems to already include > versioning in the url. The content could also be improved, I wouldn't be > fond of the work of the used minimizer. > > But, why does it need the other two slow AJAX messages? And just a > one-line diff! :( > > My round-trip time to manganese is ~164ms, so there's ample room for > improvement. > > > _______________________________________________ > Wikitech-l mailing list > [email protected] > https://lists.wikimedia.org/mailman/listinfo/wikitech-l _______________________________________________ Wikitech-l mailing list [email protected] https://lists.wikimedia.org/mailman/listinfo/wikitech-l
