in a heavy transaction PROD env, it is risk, considering c* has a lot of bugs.
the DDL has to be replicated to all nodes, use nodetool describecluster to check schema version same on all nodes, if not, you may restart that node which DDL not replicated.
in new version, DDL is none or all, you may not get it success.
It is similar to rdbms, alter table in a heavy transaction PROD env, may get resource busy error.
in non-prod, we always apply new DDL without stop applications, never had issue.