redundant writers

John E. Vincent lusis.org+riak-users at gmail.com
Tue Feb 15 15:44:26 EST 2011


Les,

This is pretty much Dynamo 101 territory at this point and one of the
tradeoffs with a distributed model.

If you aren't familiar with Dynamo, Andy Gross (from Basho) gives an
AWESOME walkthrough in an episode of TheNoSQLTapes:

http://nosqltapes.com/video/understanding-dynamo-with-andy-gross

Concurrent writes really do behave the same way they would during a
network partition event. You can just stick your riak cluster behind a
load balancer and go to town.

On Tue, Feb 15, 2011 at 3:38 PM, Les Mikesell <lesmikesell at gmail.com> wrote:
> In this scenario, the behavior I would want would be to only see one copy
> for each unique key, but to always have one even if one of the feeds or
> writers fails or the cluster is partitioned, then re-joined. This sounds
> good in theory, but what about the details? Is there likely to be a big
> performance hit from the normally-colliding writes?  Will it take twice the
> disk space, the eventually clean up the duplicates? Would it be reasonable
> to do this to riak-search with something like news stories?
>
>
> On 2/15/2011 2:02 PM, Alexander Sicular wrote:
>>
>> What about siblings? http://wiki.basho.com/REST-API.html (seach for
>> sibling)
>>
>> On Tue, Feb 15, 2011 at 14:57, Dan Reverri<dan at basho.com>  wrote:
>>>
>>> Riak maintains a single value per key and provides mechanisms (vector
>>> clocks) to detect/resolve conflicting values. In the proposed use case
>>> the
>>> multiple copies would overwrite each other and Riak, by default, would
>>> return a single value for a requested key.
>>> Behind the scenes Riak determines the appropriate value per key using
>>> vector
>>> clocks. More information about vector clocks is available here:
>>> http://blog.basho.com/2010/01/29/why-vector-clocks-are-easy/
>>> http://blog.basho.com/2010/04/05/why-vector-clocks-are-hard/
>>> Thanks,
>>> Dan
>>> Daniel Reverri
>>> Developer Advocate
>>> Basho Technologies, Inc.
>>> dan at basho.com
>>>
>>>
>>> On Tue, Feb 15, 2011 at 10:11 AM, Les Mikesell<lesmikesell at gmail.com>
>>> wrote:
>>>>
>>>> Is riak suitable as a very reliable store where you have multiple feeds
>>>> of
>>>> streaming data that are at least theoretically identical?  That is, can
>>>> you
>>>> count on writing multiple copies with the same keys at the same time to
>>>> do
>>>> something reasonable regardless of cluster partitioning?  And is this a
>>>> common usage scenario?
>>>>
>>>> --
>>>>  Les Mikesell
>>>>   lesmikesell at gmail.com
>>>>
>>>> _______________________________________________
>>>> riak-users mailing list
>>>> riak-users at lists.basho.com
>>>> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
>>>
>>>
>>> _______________________________________________
>>> riak-users mailing list
>>> riak-users at lists.basho.com
>>> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
>>>
>>>
>
>
> _______________________________________________
> 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