Horizon is very slow in 100 nodes deployment

I've tested the lab and haven't encountered any extraordinary lags. It took ~6sec from I clicked 'Login' button till the home page of Dashboard has been finally rendered. Switching between different panels took ~2-4 sec, which is also quite usual for Horizon.

That doesn't mean however that everything is perfect in Horizon. I small remark: I am located in Moscow, Russia and the Lab hardware is located somewhere in Czech Republic. I strongly believe that the lags Alexander Shaposhnikov have encountered are related to the big latency he has while sending requests to Czechia.

I've collected debugging info for the successful transition from Login page to the Home page of the dashboard: http://paste2.org/GJ14dzX3 Alexander, could you gather the same logs for the same transition? To do so, you'll need to edit /etc/openstack_dashboard/local_settings.py on all 3 controller nodes, setting DEBUG = True and PROD = False (I don't really know what the latter one means, did it just in case) and restarting apache services on all three. Then see all 3 /var/log/dashboard.log - one of the will receive log messages.

There is another thing I want to notice: somehow the same controller served all my web-requests, which doesn't seem right (why use 3 of them if no load balancing occurs?).

Timur, It looks that I've seen all the lags because mysql/rabbitmq and affected services(keystone) was overloaded and slowed down.
My main concerns are only about as example those 6 seconds for login. It's quite slow. I wouldn't insist that it should be "fast-as-lightning" but may be you have some ideas how to improve it ? I assume that optionally response shouldn't be more that 3 seconds on any action or it will be out of comfort zone for most of the people.

Aleksandr,

I have an educated guess: the first time the User logs in, Horizon has to evaluate User's permissions for each panel to decide whether it's visible in template or not. Each such evaluation implies REST-request to Keystone or some other OpenStack service. The problem is that these requests are _sequential_, thus different requests latencies sums up. I'll think what can be done here (parallelizing requests won't be an easy task).

Too hard to decrease time of loading first page in Horizon given current page architecture. The problem may be eliminated once it moves to AngularJS. Setting to Won't Fix.

I know its very old post but i am seeing slowness too with 150 compute node.. 10 to 20 second for any action.. I am using queens.