I'm getting occasional throughput spikes in my Rails 5 app hosted on Heroku. The spikes are reported in New Relic with the following transaction name: middleware/Rack/ActionDispatch::Routing::RouteSet#call. The spikes result in ~200rmp of those transactions and render the application unresponsive. Restart of the Heroku dyno solves the problem.
What might be causing this behavior? Where should I look?