If an event is updated towards the end of a minute, event['time'] can be up to 60 seconds away when executing it. For example, if the end_date of a lease is changed to 14:44 at 14:44:50, event['time'] will be 14:44, but the first execution attempt would be at 14:44:50 or even 14:45:00. This makes the retry calculation in _exec_event() incorrect.
If the event execution is triggered while the update request is still in progress, it can result in API errors due to an invalid state transition. See the following traces around 14:44:50 for details.
http://logs.openstack.org/98/589898/1/check/blazar-devstack-dsvm-py35/24b6daf/logs/screen-blazar-m.txt.gz#_Aug_08_14_44_50_165340
http://logs.openstack.org/98/589898/1/check/blazar-devstack-dsvm-py35/24b6daf/ara-report/result/c4be3ab9-8416-46bc-957b-4b82d5c88249/
The best option would be to extend the event model to keep track of the number of retries of each event.