I had thought that updating with IF and using an appropriate write consistency level would ensure consistency even in the event of a network partition (receiving and exception if it isn't possible), but I'm told that isn't so.
If two lightweight transactions occur on the same row while Cassandra is in split-brain mode, how are they resolved once the network partition disappears?
If two lightweight transactions occur on the same row while Cassandra is in split-brain mode, how are they resolved once the network partition disappears?
Every cell value has a timestamp. Cassandra configuration requieres ntp configuration on every server. So if your consistency level would allow to acknowledge operations in both split parts, then cell timestamps would be compared and newer value should win.
