On Mon, Jun 7, 2010 at 12:06 AM, David Boxenhorn <da...@lookin2.com> wrote: > I wonder if there is a robust algorithm for maintaining b-trees that doesn't > require atomicity? How about if you create the three new super columns > first, then attach them to the parent, then delete the old super column? If > it fails, it would leave junk, but that could be cleaned up every once in a > while.
I think the challenge is that beyond ordering of pieces of a single transaction, there is the problem of concurrent ones, which may well see incomplete states. That is where multi-step-operations would help of course, except that there are still read-modify parts of operations that would have to be handled by server itself (only split if super column is 'full' etc). -+ Tatu +-
