Riak crashed and crashed again when recovering

Sean Cribbs sean at basho.com
Wed May 5 15:14:46 EDT 2010


We haven't often dealt with repartitioning clusters, which is why I
recommended a complete backup and restore. The behavior when trying to
dynamically change the partition quantity is undefined. Honestly I
don't know how the backend would affect it at all. As always, however,
adding nodes will increase performance and fault tolerance. If you
intend to grow your cluster, a larger partition count will ease the
transition and provide a more even spread.

On Wednesday, May 5, 2010, Garrett Smith <g at rre.tt> wrote:
> Sean,
>
> I apologize if this is readily available in the docs, faq, etc. (I
> obviously missed if it is).
>
> When a cluster is repartitioned, e.g. as a result of adding or
> removing vnodes, is one backend preferable to another? Does keeping
> node size down help/hurt in terms of performance, reliability, etc?
>
> Garrett
>
> On Wed, May 5, 2010 at 7:05 AM, Sean Cribbs <sean at basho.com> wrote:
>> Germain,
>>
>> It looks like you're filling up the dets tables -- which have a 2GB limit per file, although Riak uses multiple files, one per vnode.  Have you tried the innostore backend?  Also if you continue to use dets, try increasing the number of partitions, which will make more, smaller files.
>>
>> Sean Cribbs <sean at basho.com>
>> Developer Advocate
>> Basho Technologies, Inc.
>> http://basho.com/
>



More information about the riak-users mailing list