It's not a fantastic fit because as you say you need to read the contents to be able to see which if any need to be deleted. (On the bright side if you are sorting by time uuid, for instance, you won't need an extra sort step.)
I would predict it would perform better than doing the same thing in a SQL database though. -Jonathan On Wed, Oct 7, 2009 at 4:18 PM, Igor Katkov <ikat...@gmail.com> wrote: > Hi, > > I was wondering if it's a good idea to use Cassandra for persist-able > circular buffer? > http://en.wikipedia.org/wiki/Circular_buffer > > The goal is to have at most X columns per key, where new columns wipe old > columns out. > There will be much more writes than reads. What bothers me is how to handle > "old" columns. > I understand that there there is no multi-remove, but even if there was that > would not help much. > > Perhaps Cassandra is ill-suited for the task... What'd you recommend? > > It could be done in SQL one way or another but it would be pretty as well: > 1. always insert and purge once in a while - this approach fails under any > significant load > 2. purge on insert - still requires unnecessary reads >
