On 2017-06-16 10:31 (-0700), John Hughes <johnthug...@gmail.com> wrote: 
> Hi Affan,
> 
> Others can likely speak to this more authoritatively I am sure, but with a
> RF of 1x, I would not expect it to rebalance. Now if you were 4 nodes and a
> RF of 2x I would expect it to.
> 

Even with an RF of 1, any token range that moves to the new (joining) node will 
result in data transfer.

What won't happen automatically, however, is data being removed from the 
source. "nodetool cleanup" is provided to do that.  Until you run "cleanup", no 
data will be removed from the original hosts.

Note, however, that if you're not sure if things are properly 
bootstrapped/repaired, don't run cleanup until you're happy. If you had RF=1 
and you joined a node with auto_bootstrap=false, the node would immediately 
join the ring and be responsible/own data that it never received, and with RF=1 
there's no way to run repair. You can either remove that node (nodetool 
decommission) and re-add it with auto_bootstrap=true to recover most of the old 
data (this isn't PERFECTLY safe, but it's better than not having the data at 
all).  Alternatively, you could use 'sstableloader' to re-stream the data into 
your cluster.

Again, that's only an issue if you had RF=1 and joined a node with 
auto_bootstrap=false.




---------------------------------------------------------------------
To unsubscribe, e-mail: user-unsubscr...@cassandra.apache.org
For additional commands, e-mail: user-h...@cassandra.apache.org

Reply via email to