Would this solve any real world problems? Or is just to make sure it doesn't cause any issues?

It bit me with a small bug yesterday when using thread locals which is how I found it.

It appears that this patch helped me out a lot - thanks @hassox!

I am migrating from passenger enterprise to puma, and I have a Sinatra app that memoizes several datastore connections using Thread.current[] keys in singletons. I observed a steady growth in the thread count per cluster worker (far beyond the configured puma limits), and memory usage growth over time. My various datastores were also showing an increase in connections over time.

I'm still having a few threading issues to track down, and I'm sure I'm not properly handling a few things. But this PR has solved several of my issues.