Yes. There really isn't a good answer for this at the moment. Unless we implement some kind of staged node addition process where you can add an entire set of nodes at once (we discussed some ideas at NGCC two years ago, but I don't think it has gone anywhere) you run into problems with any strategy that wants to automatically start replicating to new DCs.
> On Dec 3, 2016, at 10:22 AM, Ben Bromhead <b...@instaclustr.com> wrote: > > Yup the core issue is that system_auth is not great from a HA perspective > at the moment for new users. > > Given the general opinion is that RF=N is not appropriate / too rife for > abuse. Would something similar to min(rf=N, X per DC) be more appropriate? > The only thing with that is implementing that replication strategy so it is > aware of all current DCs sounds tricky (without looking into it) + I can > imagine some gnarly corner cases / complexity with adding / removing DCs. > >> On Sat, 3 Dec 2016 at 07:02 Aleksey Yeschenko <alek...@apache.org> wrote: >> >> It isn’t, but you are supposed to change it. >> >> The reason it cannot be set higher by default is that out of the box >> single-node clusters should still work, >> and setting the default RF to higher than 1 would break this, as it >> performs some queries at quorum CL. >> >> -- >> AY >> >> On 3 December 2016 at 06:47:12, sankalp kohli (kohlisank...@gmail.com) >> wrote: >> >> The point Ben is saying is that for auth keyspace, default of RF=1 is not >> good for any type of cluster whether it is small or large. >> > -- > Ben Bromhead > CTO | Instaclustr <https://www.instaclustr.com/> > +1 650 284 9692 > Managed Cassandra / Spark on AWS, Azure and Softlayer
