This will probably change as clustering gets out of beta, but the docs today say that MERGE is the fix.

Over time, as more and more operations modify a table, the degree to which the data is sorted begins to weaken, and the table becomes partially sorted. In a partially sorted table, queries that use the clustering columns may need to scan more blocks compared to a table that is fully sorted. You can re-cluster the data in the entire table by running a SELECT * query that selects from and overwrites the table (or any specific partition in it). In addition, any arbitrary portion of the table can be re-clustered using a DML MERGE statement.