On 21.08.2015 15:57, Bogdan Dobrelya wrote: > Our deployment automation teams have been addressing the critical issues > arised for HA A/A Keystone switched to the mod_wsgi and related > backends' failure modes. > We ran numerous rally tests against different configuration layouts for > Apache2 MPM worker versus Keystone and HAProxy backend config versus > deprecated eventlet case. Raw results [0] - no pretty view, sorry. > > We made a conclusion based on the tests results [1] and there are couple > bug fixes on review [2], [3]. Another related bug fix was merged for the > apache 2 control plane [4]. This reduces an amount of undesired main > Apache process restarts at deploy stage and provides better availability > for the keystone HAProxy backends. > > As a result, the situation looks stable now and there are no issues left > for the Keystone under mod_wsgi. Thank you all who participated for a > hard work done. > > [0] https://goo.gl/Hi25QG > [1] https://review.openstack.org/212439 > [2] https://review.openstack.org/209589 > [3] https://review.openstack.org/209924 > Fixed the links [0] https://goo.gl/Hi25QG [1] https://etherpad.openstack.org/p/keystone_wsgi_measuring [2] https://review.openstack.org/212439 [3] https://review.openstack.org/209589 [4] https://review.openstack.org/209924 -- Best regards, Bogdan Dobrelya, Irc #bogdando