A Dynamic, Distributed R-Tree (DDRT) library written in Elixir. The 'dynamic' part of the title refers to the fact that this implementation is optimized for a high volume of update operations. Put another way, this is an R-tree best suited for use with spatial data in constant movement. The 'distributed' part refers to the fact that this library is designed to maintain a spatial index (rtree) across a cluster of distributed elixir nodes.

The library uses @derekkraan's MerkleMap and CRDT implementations to ensure reliable, "eventually consistent" distributed behavior.

The complete documentation is available on hexdocs. You can find the hex package here.

The package can be installed by adding ddrt to your list of dependencies in mix.exs:

Start up a DDRT process with default values

Or add it to your supervision tree:

Otherwise if you're just looking to use the standalone R-tree functionality on a single machine (not a cluster of machines), you would instead use the DDRT.DynamicRtree module:

Note: all configuration parameters and public API methods are exactly the same between the DDRT and DDRT.DynamicRtree modules.

Available configuration parameters are:

• name: The name of the DDRT process. Defaults to DDRT
• width: The max number of children a node may have. Defaults to 6
• verbose: allows Logger to report console logs. (Also decreases performance). Defaults to false.
• seed: Sets the seed value for the pseudo-random number generator which generates the unique IDs for each node in the tree. This is a deterministic process; so the same seed value will guarantee the same pseudo-random unique IDs being generated for your tree in the same order each time. Defaults to 0

First it's important to understand that distributed networking capabilities come built-in with Erlang. To get Elixir processes communicating amongst themselves over a network in general, we first have to use that fundamental Erlang networking magic to make all of the running Erlang Virtual Machines "aware" of eachother's existence on the network. In Elixir, these concepts are expressed in the Node module. One can use Node.connect/2 to make two Erlang VM nodes aware of eachother, and then Elixir processes are able to send messages to eachother on those nodes.

Connecting up the Erlang VMs in your cluster is outside of the scope of this package. There are already other libraries in Elixir designed to do exactly this. Possibly the best example is bitwalker/libcluster.

A very simple libcluster configuration for quick and easy development might look like:

Then you would have to pass in those same node names to iex when you start your application, like:

Finally, after starting DDRT on each node you would use DDRT.set_members/2 to begin communication between DDRT processes like this:

From here, everything done on either tree will be reflected in the other tree.

Note: it's important that you have the same configuration parameters for each DDRT process running on each connected node in your cluster.

Starts a local DDRT named :peter

Insert "Griffin" into the :peter DDRT

Insert "Parker" on into the :peter DDRT

Query which leaves in the :peter R-tree overlap with box [{0,7},{4,8}]

Updates "Griffin" bounding box in the :peter R-tree

Repeat the last query again. (This time "Griffin" is no longer within the query bounding box.)

Now lets delete both "Griffin" and "Parker" keys from the tree.

[{x_min,x_max}, {y_min,y_max}]

If you're only interested in using an R-tree on a single machine in Elixir, you should be using the DDRT.DynamicRtree module. This module is optimized to run on a single machine, and as such the r-tree is significantly faster without the distribution overhead.

The DDRT.DynamicRtree module shares the same API and initialization options as the main DDRT module.