Rust port of the anna key-value store, using zenoh for communication.

Note: This project is still in a prototype state, so don't use it in production!

This crate started out as a line-by-line port of the original anna project developed in the RISE Lab at UC Berkeley. We recommend to check out their ICDE 2018 paper for more information.

You need the latest version of Rust for building. The build commands are cargo build for a debug build and cargo build --release for an optimized release build. After building, you can find the resulting executables under ../target/debug (for debug builds) or ../target/release (for release builds).

To run the test suite, execute cargo test. The API documentation can be generated through cargo doc --open.

This project contains several executables:

• kvs: The main key-value store node.
• routing: Routing nodes are responsible for coordinating kvs nodes and keeping track of which key is mapped to which node.
• client: A client proxy that provides an interactive prompt where you can send GET/PUT commands.
• logger: Subscribes to all local zenoh messages and logs them to stdout.
• benchmark: The benchmark node provides different ways to evaluate the performance of the key-value store. To trigger benchmark commands, the send-benchmark-commands executable can be used.
• TODO: monitoring, management

Each executable can be started by running the built executable from target. Alternatively, it's also possible to combine the build and run steps using cargo run --bin <bin-name>, where <bin-name> is the name of the executable. Since all nodes expect that an routing node is available, it is recommended to start the routing executable first.

For routing and kvs executables, several environment variables can be set to control the behavior:

• routing
  • ANNA_PUBLIC_IP: The public IP address to listen on so that other nodes can connect TCP to this node. If not set, TCP mode won't be available.
  • ANNA_TCP_PORT_BASE: The base of TCP ports to listen on. The real port used will be ANNA_TCP_PORT_BASE + thread_id. If not set, the ports will be allocated by system.
• kvs
  • ANNA_PUBLIC_IP: Same as for the routing executable.

All executables (with the exception of the logger) expect the path to a config file as argument. An example for this config file can be found in example-config.yml.

With cargo run, this argument can be passed in the following way:

cargo run --bin <bin-name> -- <path>

Note the additional space between -- and <path>. The cargo run command uses a -- argument as separator, so everything after it is passed to the executable.

Open four terminal windows and run the following commands in them (one per terminal window):

1. cargo run --bin logger to start the zenoh logger, so that we can see the messages that are sent. This step is optional.
2. cargo run --bin routing -- example-config.yml to start the routing node.
3. cargo run --bin kvs -- example-config.yml to start the key-value store node.
4. cargo run --bin client -- example-config.yml to start the client proxy.

The client proxy executable will show a kvs> prompt, in which you can use the following commands:

All operations can also be written in lowercase, e.g. put_set instead of PUT_SET.

For a description of the mentioned lattice types (e.g. LastWriterWinsLattice), see the lattice module in the API docs of this crate, which you can generate using cargo doc --open.

For benchmarking, the benchmark executable can be used. It spawns a proxy node that listens for incoming benchmark commands and executes them. It supports a variety of different commands, for example for executing a specific number of random PUT requests. To send commands, the send-benchmark-commands executable can be used.

Run the routing and kvs executables as described above under Example, but in --release mode. Then open two additional terminal windows. In the first, spawn the benchmark node by running cargo run --bin benchmark --release -- example-config.yml. From the second window, run cargo run --bin send-benchmark-commands -- <command> to send trigger the benchmark command <command>.

The following commands are supported:

• cache <num_keys>: Determines the responsible KVS nodes for keys 0..num_keys and caches this information for subsequent commands.

• warm <num_keys> <value_len> <total_threads>: Generates num_keys PUT requests to the key-value store. The total_threads argument must be the configured number of benchmark threads (the requests are split between the threads). The stored values are of the form aaaaaaa..., with a length determined by the value_len argument.

• load <ty> <num_keys> <value_len> <report_period> <time> <zipf>: Performs a random set of operations with the given zipf coefficient. If zipf is 0, a uniform random distribution is used. The time argument specifies how long the benchmark should run (in seconds) and the report_period specifies the interval (in seconds) of throughput reports.

  The num_keys argument specifies the set of possible keys: the random keys will be selected from the range 0..num_keys. The value_len parameter specifies the number of a characters in the stored values (all values are of the form aaaaaaaa...).

  The ty parameter specifies what kind of operations should be performed on the random set of keys. The following options are available:

  • G: Performs GET requests.
  • P: Performs PUT requests.
  • M: Perfomrs a PUT request and then a GET request on the same key; also reports the latency of such a PUT/GET pair.

The results are reported via zenoh messages. A subset of results is also reported as log messages in the stdout of the benchmark executable (not the send-benchmark-commands executable) and in the benchmark-thread-<thread_num>.log file.

Make sure to compile all nodes in --release mode when benchmarking!

Licensed under the Apache License, Version 2.0 (LICENSE or http://www.apache.org/licenses/LICENSE-2.0). Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be licensed as above, without any additional terms or conditions.