If we use WeakReference, it allows the cached object to be GCed when low on memory, so the user of the cache library can aggressively cache items without worrying about the memory usage.

Can be refactored from Nebula

clay-one/nebula#37

Currently, the identifier for a cache instance is the type tuple of key and value.

This means there can't be two different caches with the same key/value types. Eg. we can't have two different set of caches of <string, string>.

The design should change in a way that allows this, as this limitation have been problematic in some use cases.

Much like the ASP.NET OWIN which had an OwinContext, There is also an IDictionary<object, object> in ASP.NET Core's HttpContext.Items property. It would be nice to have some kind of helper to ease the access of such items in order to not to inject HttpContextAccessor everywhere.

The following work of Steve Gordon might be helpful:
https://github.com/stevejgordon/CorrelationId

The code to generate API access log is redundant in many projects, and can be unified in a common library with easy configuration on the OWIN or AspNetCore pipeline with a single extension method.