It appears that adding a column to an existing table in a production Cassandra cluster is pretty common. Under what conditions is it considered acceptable/safe to do so, in terms of availability and performance?
If it’s difficult or impossible to provide a definitive or clear-cut answer, then how would the expected impact be characterized? Impact might be expressed in terms of e.g. simply “no expected availability impact;” Big-O notation; or empirical measurements like duration of add operation, change in request latencies for impacted Cassandra cluster or (end to end for) client service, change in CPU or memory utilization of impacted cluster. Citations of a good resource(s) on the topic would be helpful.