There's no code preventing both ProfileIOData and ProfileIODataImpl from having their TransportSecurityPersisters load state at the same time.  They currently do this on random task runners, so they can both be loading state at the same time.  On Windows, generally only one object is allowed to read from a file at a time - the other will be unable to open it.

So it looks to me like ProfileIOData and ProfileIODataImpl could prevent the other from loading the TransportSecurityPersister state on Windows.  Moreover, if the ProfileIODataImpl prevents ProfileIOData from loading the state, we'll effectively clear the old cached state on shutdown, which seems less than ideal.

Admittedly, this isn't a correctness-breaking bug, since this is all best effort, but seems not great.

Or am I missing something?