I know that Cassandra offers atomicity and isolation for batch mutations on row-level basis since version 1.1 (http://www.datastax.com/dev/blog/row-level-isolation)
But do these guarantees also hold for all the repair mechanisms (hinted handoff, read repair and node repair)?
I'd guess that these operations also use batch mutations and one can therefore state that these guarantees hold. But I don't know the code and therefore can't prove or disprove that point.
I neither have found a source which tells me about atomicity and isolation guarantees for the aforementioned repair mechanisms.
So maybe someone of you can give me a source or can justify if these guarantees also hold in these cases?
Kind Regards
Stefan
Repair works at the sstable level, not individual mutations. But since any individual mutation will not be split across multiple sstables, you do get the same isolation guarantee during repair.