Rebalancing (newbie alert)
sean at basho.com
Tue May 18 19:23:42 EDT 2010
> 1. I vaguely recall one of Riak authors replying to someone here that it is not possible to know where particular bucket resides, i.e. at which vnode.
> If so, how can one say after one physical node crashed & burned (say, its hard drive failed totally) or another physical node was added that key/bucket pairs have indeed rebalanced across entire logical "ring"?
Riak automatically replicates your data across the cluster, so if a node goes down, the data should still be available (assuming you are making more than one replica). To the application you build that uses Riak, where the data is stored in the cluster is opaque. You can treat one node the same as any other. In addition, hinted handoff and sloppy quorums ensure that Riak can still function despite node failures.
When adding or removing nodes from the cluster, it starts a state called "handoff" in which ownership/claim of the various partitions is transferred to the new node (or from the leaving node to the remaining ones). This may take time depending on the amount of data in the cluster.
> Is it possible to watch such process of rebalancing via some interface?
> 2. is it possible to obtain information how many "copies" particular bucket has in the ring? I don't mean configuration information, but in fact the situation when one physical node has crashed.
No, you cannot directly. In general, you should not worry about how many replicas are available, unless you're in a situation where you have massive failures. Riak's standard operating mode is to remain available for reads and writes, even in the case of node outages. If you receive a "not found" response where you don't expect one (in the case of a failure), lower the R value (read quorum) to get a successful read.
> 3. Backups. Is it safe to just dump a live filesystem (e.g. using dump/restore) of a physical riak node and that means that I have a reliable backup? Or is like files of an opened relational database - might work but this is not the way it should be done?
> Is there some backup & restore utility for a physical riak node that allows to quickly dump its buckets into, say, one huge backup file?
Quickly is relative, of course. You can dump the data from a node or an entire cluster via `riak-admin backup`.
Sean Cribbs <sean at basho.com>
Basho Technologies, Inc.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the riak-users