riak | Riak is a decentralized datastore from Basho Technologies

From the above, I can imagine this being implemented either via maps, possibly containing sets with map reduce/secondary indexes or by using Yokozuna (Solr).

Despite this, the above data set looks more like it should be in a relational database than a non-SQL database. For something in the middle, you might want to consider Riak TS (download) which is similar to KV but has a thin SQL layer on top. The current version (1.5.1) is a bit old but we plan to launch an improved version later this year with additional SQL features available.

I'd suggest using bucket-types as an additional namespace https://www.tiot.jp/riak-docs/riak/kv/2.2.6/using/reference/bucket-types/#buckets-as-namespaces as this would allow you to give each tenant a namespace bucket-type and then they can create buckets and keys inside that bucket type namespace.

To avoid people fishing on bucket-type to see if they can gain access to a different tenant's data, you can then extend this by creating users and granting exclusive permissions by bucket or bucket type https://www.tiot.jp/riak-docs/riak/kv/2.2.6/using/security/basics/#managing-permissions

Using these two features together should allow a multi-tenant environment within the same Riak KV instance.

On macOS Catalina (and prior versions, and most likely subsequent versions too), there are two aspects to the problem and some suggested solutions.

The reality is that there are no guarantees when using $bucket or $key, especially if you combine with Map/Reduce, that they will not impact the stability of the cluster. However, if you know there to be a relatively small number of keys in a bucket, or if you use max_results in your query - then $bucket should be relatively safe (certainly a lot better than using list keys).

There are safer ways of erasing masses of keys in Riak KV 2.9.1, which are available regardless of backend (assuming the use of Tictac AAE).

As for $bucket or $key queries returning deleted keys, I suspect that this may well be true due to the fact that this is an "internal" leveldb query and within leveldb it wan't be able to distinguish between a Riak object and a riak tombstone. The improvements in Riak 2.9.x handle this situation, and won't return deleted keys (unless you're looking for tombstones to reap).

When the client sends a get it goes to a single node that coordinates the get. The node_gets stat at that node gets incremented.

The node hashes the requested key, looks up the has in the ring, and gets the n_val(default 3) vnodes that should hold the value. It then forwards the request to the node that owns each of those vnodes. The vnode_gets stat at each of those nodes is then updated.

So each get from the client should equate to 1 node_get and n_val vnode_gets.