MemoryDistributedCache should prefix keys to avoid conflicts with IMemoryCache

[rmja]

I am having a scenario where I first look in a local IMemoryCache for a hit, and if not found, go to a distribted IDistributedCache. If the key is found in the distributed cache, I store is locally for a short time in the IMemoryCache.
This works perfectly when IDistributedCache is e.g. a Redis cache, but when the implementation is MemoryDistributedCache, then the keyspace for the two caches is the same. This results in the first lookup in IMemoryCache to return the entry expected to be returned in IDistributedCache.

Would it be possible to e.g. prefix all keys in MemoryDistributedCache with something like "distributed" so that the two caches behaves like they are separate.

[Tratcher]

Is the data stored the same, or are they in different formats due to serialization requirements? If it's the same data this seems like a harmless side-effect for local scenarios.

You could just as easily apply the prefixes in your code.

One other mitigation would be to ensure they are using different instances of IMemoryCache. You can adjust the DI registrations for this.

[davidfowl]

I think this is a reasonable request.

[Tratcher]

Note just appending "distributed." to each key only lowers the odds of a collision. You'd need to use a unique identifier per instance of MemoryDistributedCache.

[rmja]

@Tratcher that is the problem. The local one is having the instance to cache directly, where the distributed one has a serialized version. So when finding the cache hit I get invalid cast exception.

[rmja]

@Tratcher well that depends. That would only cause a collision if my key already ended with "distributed", where when there is no pre/suffix, then I have collision in 100% of the cases.

And it seems wrong from a user perspective that the local and distributed cache must not use the same key.

[Ulriksen]

The key must be unique, having a possiblity for collision between local and distributed keys will create hard to find bugs.

Having the same key space for both local and distributed cache seems like a bad idea.

[sebastienros]

Could the default implementation for IDistributedCache have it's own local cache storage instead of resolve the registered one? I assume we don't really care about have it use the same concrete registered implementation of IMemoryCache. The contract today is to have it work locally, not that it has to use the DI one.

[muratg]

@sebastienros Do you have time to look into this one in 1.0.0?

[sebastienros]

Yes, is there an agreement on my proposal? Or should I engage more devs?

[Tratcher]

@sebastienros give this workaround a try:
services.AddSingleton<IDistributedCache>(new MemoryDistributedCache(new MemoryCache())