Hitting per-node limitations

Dean Cookson cookson at basho.com
Tue Jul 13 11:30:17 EDT 2010


Michael,
The answer to your first question is basically "yes".  The sysadmin really should be monitoring memory usage and disk space and adding nodes to the cluster when/if needed.  The SNMP hooks in the EnterpriseDS version are very handy for that type of thing and I'd be happy to have that conversation off list, if you're interested.  That being said, the system does automatically balance objects across the cluster evenly and adding a new node will trigger a re-balance that does not interrupt cluster function.  Currently, there is not really a way to force an un-even distribution across the cluster.

On the best-practices front, there is a section of the wiki that addresses exactly that at: https://wiki.basho.com/display/RIAK/Best+Practices

Beyond that, I'd add that Riak is generally happier with more nodes rather than fewer (e.g. If N=3, 5 nodes is a good place to start).

Best regards,
Dean

--
Dean Cookson
VP Business Development
Basho Technologies, Inc.
Google Voice: +1 415 692 1775
cookson at basho.com


On Jul 12, 2010, at 2:59 PM, Michael Russo wrote:

> Hi all,
> 
> I'm currently evaluating Riak and am looking to understand the system in more depth.  One of the best ways I know of doing this is to examine the edge cases.  So, without further ado:
> 
> There are several per-node limitations that, if hit, could cause a Riak node to stop functioning properly.  For example,
> 
> - disk: each node has a file system of finite size and cannot indefinitely accept writes
> - memory: the Bitcask keydir structure must fit entirely in memory and thus cannot indefinitely accept writes
> 
> From a management perspective, is the onus on the sysadmin to monitor disk and memory usage and to increase the number of nodes in the cluster as appropriate, or are there any built-in mechanisms to automatically re-balance data across the cluster?  
> 
> Furthermore, if a limit is inadvertently hit, is there a re-balancing mechanism available that can be manually triggered to compensate, or is it a requirement that every node take an equal share of partitions?
> 
> Are there any other related best-practices?
> 
> Thanks,
> Michael
> _______________________________________________
> riak-users mailing list
> riak-users at lists.basho.com
> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com





More information about the riak-users mailing list