On Fri, Nov 13, 2015 at 1:47 PM, Peddi, Praveen <pe...@amazon.com> wrote:
> We do not currently run repairs because we know our deployment time for > each cassandra node is very short. I do understand we have to run repairs > but would repair be in the picture here when no nodes in the cluster were > down for last 2 weeks? > The only mechanism Cassandra provides that *ensures* that data doesn't undelete itself after gc_grace_seconds is periodic repair. To expand slightly on what rustyrazorblade says down-thread, you might have : 1) dropped a mutation 2) stored a hint 3) failed to deliver that hint If that hint was a DELETE, you will unmask the deleted data once gc_grace_seconds has passed and the tombstone has been compacted away on other nodes. =Rob