Comment 4 for bug 1674381

node-1 has this time jump (same ~40 sec) a bit earlier at 01:28:41.161:

2017-03-20 01:28:41.161 17933 DEBUG neutron.db.agents_db [req-faf604d7-11a7-4e32-a727-65171800240d - - - - -] Agent healthcheck: found 14 active agents agent_health_check /usr/lib/python2.7/dist-packages/neutron/db/agents_db.py:310
2017-03-20 01:29:26.181 17922 DEBUG oslo_messaging._drivers.amqpdriver [-] received message msg_id: c6ff2fb951264277ad4fe3a907d0644b reply to reply_0f40b8804c2d4211bfc601d4462faf2b __call__ /usr/lib/python2.7/dist-packages/oslo_messaging/_drivers/amqpdriver.py:194

And after this node-1 starts reporting dead agents.

Currently it looks like time syncing issue...