Design pattern for memcached data caching

Question

It\'s easy to wrap optional memcached caching around your existing database queries. For example:

Old (DB-only):

function getX
    x = get from db
    r         


        

Answer 1

Read about the Identity Map pattern. This is a way to make sure you only keep one copy of a given row in your application space. Whether you store it in memcached or just plain objects, this is a way to handle what you want. I would guess that Identity Map is best used when you typically fetch one row at a time.

When you fetch whole subsets of a table, then you have to process each row individually. You might frequently have the dilemma of whether you're getting the best use out of your cache, because if 99% of your rows are in the cache but one requires fetching from the database, you have to run the SQL query anyway (at least once).

You could transform the SQL query to fetch only rows that aren't in the cache, but it's nontrivial to perform this transformation automatically without making the SQL query more costly.