> Say I have two processes on separate machines, and a Cassandra cluster over > several machines. If the first process writes (insert) to a column while the > second process reads (get / get_slice / get_range_slices / others?) from that > column (say the consistency level is QUORUM if that makes a difference), is > the > write "atomic", or could the second process get corrupt (half-written) data? > What if Cassandra is not a cluster, but runs on a single machine? I'm > currently > using version 0.6.3, but the question is general.
You will never see an "in-between" value of a single column. You'd either see the old value or the new value. However, given a batch mutation will multiple changes in it your readers may see the results of said batch mutation being partially applied; so there is no snapshot consistency type semantics with respect to multiple values in a batch mutation. -- / Peter Schuller