It's not just for testing. We found that loading/updating keys was somewhat 
slow, even with a low replication count. So we're exploring routes of taking a 
node down in an existing cluster, loading a large batch of keys, and then 
bringing it up. It's ok if it takes the data a while to rebalance after we 
rejoin the node.

It sounds like this isn't as explored as I first thought, so we'll do our own 
analysis.

-Nate

On Oct 10, 2011, at 9:18 AM, Ryan Zezeski wrote:

> Nate,
> 
> You could do this but if you're just doing it for testing then it seems like 
> more work than necessary.  If you insist on doing it this way you'll want to 
> write it in Erlang so you can make use of the preflist calculation, 
> riak_object, and eleveldb driver.
> 
> You could do the following:
> 
> 1. stand up a fresh cluster
> 
> 2. use one of the clients to load the data
> 
> 3. archive the data dir on all nodes
> 
> Next time you want to stand up a cluster with the same data simply unarchive 
> the data dirs before starting the cluster.
> 
> -Ryan
> 
> 
> On Fri, Oct 7, 2011 at 9:29 PM, Nate Lawson <[email protected]> wrote:
> We're working on a test where we want to add batches of keys to Riak. We're 
> using the LevelDB backend with 1.0. One suggestion we heard was to take down 
> a node, batch insert directly into the backing store, and then bring the node 
> up again.
> 
> Could someone give us some more details on this? How would Riak know which 
> keys in LevelDB had not yet been redistributed to the rest of the cluster?
> 
> Thanks,
> Nate
> 


_______________________________________________
riak-users mailing list
[email protected]
http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com

Reply via email to