This seems reasonable to me, but it raises a question of roadmap. My understanding is that we are deprecating compact storage, and will remove it in a future release (or have already partially removed it? I forget). Do these issues then constitute a blocking issue for GA, or do we modify our roadmap, or do we stipulate that users must upgrade to a future patch version of 4.0 before going to 4.next/5.0?
From: Benjamin Lerer <ble...@apache.org> Date: Friday, 4 June 2021 at 09:53 To: dev@cassandra.apache.org <dev@cassandra.apache.org> Subject: [DISCUSSION] Should we mark DROP COMPACT STORAGE as experimental Hi everybody, There are a significant amount of issues with DROP COMPACT STORAGE that can be pretty surprising for users. To name a few: * Some hidden columns will show up changing the resultset returned for wildcard queries * As COMPACT tables did not have primary key liveness there empty rows inserted AFTER the ALTER will be returned whereas the one inserted before the ALTER will not. * Also due to the lack of primary key liveness the amount of SSTables being read will increase resulting in slower queries * After DROP COMPACT it becomes possible to ALTER the table in a way that makes all the row disappears * There is a loss of functionality around null clustering when dropping compact storage (CASSANDRA-16069) In my opinion DROP COMPACT STORAGE is not ready for production use unless users fully understand what they are doing. By consequence, I am wondering if we should not mark it as experimental as we did for the Materialized Views (CASSANDRA-13959). What is your opinion?
