Generating a unique cache key based on method arguments

Question

I have a basic repository framework that eventually executes a query and maps the results back into a object:

For Example:

    public SomeEntity Get(id)
    {
        return base.GetItem<SomeEntity>
                   ("select * from SomeEntities where id = @idParam",
                    new { idParam = id}); 
    }

If this looks like Dapper, it is because under the hood GetItem is wrapping Dapper.

I'd like to add automatic caching to GetItem, I have two arguments that come in:

• A string containing the query.
• a anonymous dictionary containing any parameters.

I'm worried that doing a simple prime hash on these parameters would cause cache key collisions, and when you are pulling data from a cache, a collision can be very very bad (I.E. leaking sensitive information).

So, what techniques do I have that would generate a reasonably sized cache key, while guaranteeing uniqueness based on the input of a query and parameters?

Answer 1

I use the following extension methods to make cached versions of delegates:

    public static Func<T, TResult> AsCached<T, TResult>(this Func<T, TResult> function)
    {
        var cachedResults = new Dictionary<T, TResult>();
        return (argument) =>
        {
            TResult result;
            lock (cachedResults)
            {
                if (!cachedResults.TryGetValue(argument, out result))
                {
                    result = function(argument);
                    cachedResults.Add(argument, result);
                }
            }
            return result;
        };
    }

    public static Func<T1, T2, TResult> AsCached<T1, T2, TResult>(this Func<T1, T2, TResult> function)
    {
        var cachedResults = new Dictionary<Tuple<T1, T2>, TResult>();
        return (value1, value2) =>
        {
            TResult result;
            var paramsTuple = new Tuple<T1, T2>(value1, value2);
            lock(cachedResults)
            {
                if (!cachedResults.TryGetValue(paramsTuple, out result))
                {
                    result = function(value1, value2);
                    cachedResults.Add(paramsTuple, result);
                }
            }
            return result;
        };
    }

    public static Func<T1, T2, T3, TResult> AsCached<T1, T2, T3, TResult>(this Func<T1, T2, T3, TResult> function)
    {
        var cachedResults = new Dictionary<Tuple<T1, T2, T3>, TResult>();
        return (value1, value2, value3) =>
        {
            TResult result;
            var paramsTuple = new Tuple<T1, T2, T3>(value1, value2, value3);
            lock(cachedResults)
            {
                if (!cachedResults.TryGetValue(paramsTuple, out result))
                {
                    result = function(value1, value2, value3);
                    cachedResults.Add(paramsTuple, result);
                }
            }
            return result;
        };
    }

And so on for N parameters...

In case it's not clear from the code, I create a tuple with the arguments, and use the tuple as a key to a dictionary that holds the return values for each set of arguments. Note that every time you call AsCached, you create a separate cache.

You can use these methods as follows:

private Func<int, SomeEntity> _getCached;

public SomeEntity Get(int id)
{
    if (_getCached == null)
    {
        Func<int, SomeEntity> func = GetImpl;
        _getCached = func.AsCached();
    }
    return _getCached(id);
}

private SomeEntity GetImpl(int id)
{
    return base.GetItem<SomeEntity>
               ("select * from SomeEntities where id = @idParam",
                new { idParam = id}); 
}

Answer 2

I see a few options

1. Pack the data into a class, use a BinaryFormatter to serialize the class and perform a SHA1 hash on the serialized data to to give you a hash key.

2. Pack the data into a class, implement IEqualityComparer which can then be stored in a Dictionary. By implementing IEqualityComparer you will control the generation of the Hash and the data comparison performed to identify the unique data when collisions occur.