Hi Team, It's me again and I've got some new tests results [3]. I'd got the following test environment:
- 3 controllers - 2 computes - 1 node with horizon - 60 items per page - 60 instances launched with only 3 different images. As always, any feedback is welcome [3] https://docs.google.com/spreadsheets/d/14zDpdkPUfGDR_ ZycaGUoT64kHsGi1gL9JOha8n5PVeg/edit#gid=0 Regards, Ivan Kolodyazhny, http://blog.e0ne.info/ On Wed, Dec 6, 2017 at 12:47 PM, Ivan Kolodyazhny <e...@e0ne.info> wrote: > Hi Team, > > Last week we had a hot discussion about this topic at the meeting [6]. We > still don't have an agreement how to go forward with it. As was requested, > I did some perf testing with a current approach [3]. TBH, these tests > results don't show real data but look realistic. I used the same VM with > the same code (just checked out to the previous commit of horizon). To > get better results, we need to re-test all but I don't have enough time > now:(. I hope, these results are good enoudh and we can make our final > decision on this topic. > > [3] https://docs.google.com/spreadsheets/d/14zDpdkPUfGDR_ > ZycaGUoT64kHsGi1gL9JOha8n5PVeg/edit#gid=0 > [6] http://eavesdrop.openstack.org/meetings/horizon/2017/ > horizon.2017-11-29-20.00.log.html#l-12 > > > Regards, > Ivan Kolodyazhny, > http://blog.e0ne.info/ > > On Tue, Oct 31, 2017 at 11:06 PM, Ivan Kolodyazhny <e...@e0ne.info> wrote: > >> Just forgot to mention one other option: we can use celery [6] too but I >> didn't do any performance testing with it. This solution will be more >> complex. >> >> [6] http://www.celeryproject.org/ >> >> Regards, >> Ivan Kolodyazhny, >> http://blog.e0ne.info/ >> >> On Tue, Oct 31, 2017 at 7:15 PM, Ivan Kolodyazhny <e...@e0ne.info> wrote: >> >>> Hi team, >>> >>> We all know that unfortunately Horizon's performance not good enough in >>> some cases. That's why we try to fix it on both sides: client-side and >>> server-side. >>> >>> I would like to talk mostly about server-side now. On some views, we use >>> 'futurist' library [1] to fetch resources from API's in parallel. I've >>> filed a blueprint for this effort [2]. You can find some work in progress >>> patches in the Gerrit. >>> >>> For now, we use futurist.ThreadPoolExecutor in Horizon to fetch >>> resources from APIs. It requires some specific Apache configuration changes >>> to allow WSGI app to create threads. It means, our code execution depends >>> on Apache mod_wsgi configuration. >>> >>> To avoid this, we can use futurist.GreenThreadPoolExecutor. It >>> uses eventlet green threads which can be used to speed-up I/O operations >>> like 'call external API'. >>> >>> I did very simple performance testing [3] with proposed patch [4] for >>> volumes views. My tests are not ideal but you can see how it's going with >>> green thread. I propose to switch to the futurist.GreenThreadPoolExecutor >>> and add 'eventlet.monkey_patch()' call to the wsgi app for Horizon. It will >>> speed up parallel API calls and make Horizon independent on Apache or other >>> web server configuration. >>> >>> I added this topic to the next weekly meeting agenda [5] so we can >>> discuss it there too. >>> >>> >>> [1] https://github.com/openstack/futurist/ >>> [2] https://blueprints.launchpad.net/horizon/+spec/fetch-res >>> ources-in-parallel >>> [3] https://docs.google.com/spreadsheets/d/14zDpdkPUfGDR_ZycaGUo >>> T64kHsGi1gL9JOha8n5PVeg/edit?usp=sharing >>> [4] https://review.openstack.org/#/c/426493/ >>> [5] https://wiki.openstack.org/wiki/Meetings/Horizon#Agenda_ >>> for_Next_Meeting >>> >>> Regards, >>> Ivan Kolodyazhny, >>> http://blog.e0ne.info/ >>> >> >> >
__________________________________________________________________________ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
