Memory leak in Ruby on Rails app as garbage collector activity spikes

Question

• Framework: Rails 5.0.0.1
• Platform: Heroku
• Server: Puma, 30 processes with 10 workers each

We're seeing an increase in memory once an hour, coinciding with the Ruby garbage collector as can be seen in the screenshot linked to below. The number of requests per time unit was almost constant throughout the memory increase (~1300rpm).

Memory seems stable except for the garbage collector runs, usually fluctuating by a few megabytes in either direction around a fairly stable average. Debugging the app locally using profiling tools such as memory profiler or dumping the heap space using Objectspace allocation tracing didn't conclusively identify any memory leaks.

Question:

How to find out if it has something to do with the garbage collector not working properly?