There's more burden on the user, but with a benefit– client supplied timestamps mean that client retries are idempotent as long as they use the same timestamp.*
-ryan * doesn't apply to counters On Fri, Jun 17, 2011 at 8:01 AM, Tim Wisniewski <tjw3...@gmail.com> wrote: > it's slightly less burden on the user > > On Fri, Jun 17, 2011 at 11:00 AM, Gary Dusbabek <gdusba...@gmail.com> wrote: > >> That problem is already solved externally. I don't see the value of >> reimplementing it inside of cassandra. >> >> Gary. >> >> On Fri, Jun 17, 2011 at 09:58, Tim Wisniewski <tjw3...@gmail.com> wrote: >> > Hi, >> > >> > My understanding is data corruption can occur if the node clocks are too >> far >> > out of sync. What if Cassandra nodes maintained an internal time sync >> via >> > ntp or something? >> > >> > -Tim >> > >> >
