While I was working on the grenade refactor I was considering using openstack client for some resource create / testing. Doing so made me realize that osc is sluggish. From what I can tell due to the way it loads the world, there is a minimum 1.5s overhead on every command execution. For instance, "openstack server list" takes a solid extra second over "nova list" in my environment.
I wrote a little tool to figure out how much time we're spending in openstack client - https://review.openstack.org/#/c/172713/ On a randomly selected dsvm-full run from master it's about 4.5 minutes. Now, that being side, there are a bunch of REST calls it's making, so it's not all OSC's fault. However there is a lot of time lost to that reload the world issue. Especially when we are making accounts. For instance, the create accounts section of Keystone setup: https://github.com/openstack-dev/devstack/blob/master/stack.sh#L968-L1016 Now takes 3.5 minutes in master - http://logs.openstack.org/13/172713/1/check/check-tempest-dsvm-full/d3b0b8e/logs/devstacklog.txt.gz 2015-04-12 12:37:40.997 | + echo_summary 'Starting Keystone' 2015-04-12 12:41:06.833 | + echo_summary 'Configuring and starting Horizon' The same chunk in Icehouse took just over 1 minute - http://logs.openstack.org/28/165928/2/check/check-tempest-dsvm-full/f0b3e07/logs/devstacklog.txt.gz 2015-04-10 15:59:08.699 | + echo_summary 'Starting Keystone' 2015-04-10 16:00:00.313 | + echo_summary 'Configuring and starting Horizon' In master we do create a few more accounts as well, again, it's not all OSC, however OSC is definitely adding to it. A really great comparison between OSC and Keystone commands is provided by the ec2 user creation: Icehouse: http://logs.openstack.org/28/165928/2/check/check-tempest-dsvm-full/f0b3e07/logs/devstacklog.txt.gz#_2015-04-10_16_01_07_148 Master: http://logs.openstack.org/13/172713/1/check/check-tempest-dsvm-full/d3b0b8e/logs/devstacklog.txt.gz#_2015-04-12_12_43_19_655 The keystone version of the commands take ~ 500ms, the OSC versions 1700ms. So, under the current model I think we're paying a pretty high strategy tax in OSC use in devstack. It's adding minutes of time in a normal run. I don't know all the internals of OSC and what can be done to make it better. But I think that as a CLI we should be as responsive as possible. < 1s seems like it should be target for at least all the keystone operations. I do think this is one of the places (like rootwrap) where load time is something to not ignore. -Sean -- Sean Dague http://dague.net __________________________________________________________________________ OpenStack Development Mailing List (not for usage questions) Unsubscribe: [email protected]?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
