As you can see there were around 24 threads under the unicorn worker[0] alone and some of those threads could very well be holding on to ActiveRecord connections leading to the over-use of available connections.