Consider a M:M relation that needs to be represented in a Cassandra data store.
What M:M modeling options are available? For each alternative, when is it to prefer? What M:M modeling choices have you made in your Cassandra powered projects?
Instead of using a join table the way you would with an rdbms, you would have one ColumnFamily containing a row for each X and a list of Ys associated with it, then a CF containing a row for each Y and a list of each X associated with it.
If it turns out you don't really care about querying one of those directions then only keep the CF that you do care about.
Cassandra by design is Key value database, so to achieve M:M there are two ways to do it.
De-normalize your data so every relation ship should duplicate data.
ie. x->y(value) and x->z(value) and a->y(value)
y should be saved for x and a
This is how it should be done as it's give you strength of database
Save reference for relational key as value.
x->y(key) and x->z(Key) and a->y(Key)
So if you need x with value of y it should be two operation, get x which will give you value of y. Then get y itself in a separate operation.
Cassandra is not RDBMS so don't wrap you mind around traditional way of doing it by dropping values and define relationship.
来源：https://stackoverflow.com/questions/2573106/what-are-the-alternative-ways-to-model-mm-relations-in-cassandra