In a TTL only use case with no explicit deletes, if read CL + write CL > RF you can likely avoid repairs with a few huge caveats:
1) read repair may mess up your ttl expiration if you’re using TWCS 2) if you lose a host you probably need to run repairs or you may not see some data after replacement (true in general) -- Jeff Jirsa > On May 12, 2018, at 5:27 AM, onmstester onmstester <onmstes...@zoho.com> > wrote: > > Thank you Nitan, > That's exactly my case (RF > CL). But as long as there is no node outage, > shouldn't the hinted handoff handle data consistency? > > Sent using Zoho Mail > > > > ---- On Sat, 12 May 2018 16:26:13 +0430 Nitan Kainth <nitankai...@gmail.com> > wrote ---- > > If you have RF>CL then Repair needs to be run to make sure data is in sync. > > Sent from my iPhone > > On May 12, 2018, at 3:54 AM, onmstester onmstester <onmstes...@zoho.com> > wrote: > > > In an insert-only use case with TTL (6 months), should i run this command, > every 5-7 days on all the nodes of production cluster (according to this: > http://cassandra.apache.org/doc/latest/operating/repair.html )? > nodetool repair -pr --full > When none of the nodes was down in 4 months (ever since the cluster was > launched) and none of the rows been deleted, why should i run nodetool repair? > >