Hi, It almost sounds like you might just want to investigate use riak_core instead of full blown riak. The riak_core component only deals with ring management and messaging. We've used it to create a completely different distributed store with a thrift frontend and custom C database backend. You could probably swap out your home-grown Dynamo software with just riak_core?
HTH, -Anthony On Tue, Dec 13, 2011 at 11:17:05AM +0100, Marc Worrell wrote: > Hi, > > We are working on the distributed version of Zotonic (the Erlang CMS), as > some know. > > We have some home-grown Dynamo like software for the ring-management. > Still I want to do a reality check, as I much prefer to go with more > "standard" software. > > For Riak we bumped into the following problems with some Zotonic use-cases: > > 1. It must be safe out of the box. This means: > - Only ports open that are controlled by the Zotonic access control. > - So no Riak APIs (webmachine, protocol buffers, handoff listeners etc.) > Are there configurations for Riak to disable those ports? > > 2. A very common install is a single machine/node, growing when needed. > We don't want to store three copies of all data on a single machine when > that is so common. > Is it possible to tune the riak put/get FSMs to group requests to vnodes > on the same node? > Do the vnodes > In that case we might be able to use a special storage engine per node. > > 3. We have many sites with many tables in a single install. > Think of 20+ sites with 30+ tables. > It must be possible to: > - keeps these tables/data separated > - move all data for one site between servers > How can this be done without having a too big gossip state and how can we > keep the data partitioned? > > 4. We need ad-hoc queries. > I have the impression that the full text and 2i indexing of Riak is > problematic for us because: > - we have document based queries (Riak full text is term based) > - many documents have the same terms (think categories etc) > - > As a solution I think of implementing a SQL backend (PostgreSQL and > Sqlite3) with its own commands. > We are doing that already in our own Dynamo trials. > What problems should I expect when following that route? > > 5. We have services per site, not only per node. > - Is it possible to use more complex service names than just an atom? > > 6. We are using a Webmachine clone, which is much faster for our use case. > I understand to use our version I need to clone the whole Riak repo and > modify rebar.config files, is that correct? > > > Guess I will have more questions when we do decide to take the Riak route. > > > Marc Worrell > > > > > _______________________________________________ > riak-users mailing list > riak-users@lists.basho.com > http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com -- ------------------------------------------------------------------------ Anthony Molinaro <antho...@alumni.caltech.edu> _______________________________________________ riak-users mailing list riak-users@lists.basho.com http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
