You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@cassandra.apache.org by Yang <te...@gmail.com> on 2011/09/21 08:38:04 UTC
-DCassandra.join_ring=false;
CASSANDRA-526 provides this ability, I just want to make sure,
let's say I have one node, with token 0, and now I want to add 2 new
ones, with initial_token set at 1/3 and 2/3 of the full range.
now I start nodes 2 and 3 with the -DCassandra.join_ring=false option,
and later use JMX joinRing() to let 2 and 3 join.
if there is a slight lag between the time 2 and 3 talk to 1, would it
be possible that 2/3 of the keyspace is shipped out to node 2, and
then when node 3 joins, it's shuffled out to node 3 again
(unnecessarily) ? basically I want everyone to come together, agree
on
the current global membership, then start carving out the existing load.
Thanks
Yang
Re: -DCassandra.join_ring=false;
Posted by aaron morton <aa...@thelastpickle.com>.
If you have node 1 with token 0 and a total token range to 10 then…
* add node 2 with token 3 and it will only take ownership of the range 0 to 3. Node 1 will own 3 to 0
* add node 3 with token 6 and it will take ownership of the range 3 to 6. Node 1 will now own 6 to 0
Unless you have an urgent need I would one node at a time and let the bootstrap finish.
Cheers
-----------------
Aaron Morton
Freelance Cassandra Developer
@aaronmorton
http://www.thelastpickle.com
On 21/09/2011, at 6:38 PM, Yang wrote:
> CASSANDRA-526 provides this ability, I just want to make sure,
>
> let's say I have one node, with token 0, and now I want to add 2 new
> ones, with initial_token set at 1/3 and 2/3 of the full range.
>
> now I start nodes 2 and 3 with the -DCassandra.join_ring=false option,
> and later use JMX joinRing() to let 2 and 3 join.
>
> if there is a slight lag between the time 2 and 3 talk to 1, would it
> be possible that 2/3 of the keyspace is shipped out to node 2, and
> then when node 3 joins, it's shuffled out to node 3 again
> (unnecessarily) ? basically I want everyone to come together, agree
> on
> the current global membership, then start carving out the existing load.
>
>
> Thanks
> Yang