considering: row size large or not update a lot or not ----- update is insert actually read heavy or not overall read performance
if row size large , you may consider table:user_detail , add column id in all tables. In application side, merge/join by id. But paid read price, 2nd query to user_detail. Just my 2 cents. hope helpful. Thanks, James On Sun, Jul 29, 2018 at 11:20 PM, onmstester onmstester <onmstes...@zoho.com > wrote: > > How many rows in average per partition? > > around 10K. > > > Let me get this straight : You are bifurcating your partitions on either > email or username , essentially potentially doubling the data because you > don’t have a way to manage a central system of record of users ? > > We are just analyzing output logs of a "perfectly" running application!, > so no one let me change its data design, i thought maybe it would be a more > general problem for cassandra users that someone both > 1. needed to access a identical set of columns by multiple keys (all the > keys should be present in rows) > 2. there was a storage limit (due to TTL * input rate would be some TBs) > I know that there is a strict rule in cassandra data modeling : "never use > foreign keys and sacrifice disk instead", but anyone ever been forced to do > such a thing and How? > >