Key-Value Store

This is an academic reference implementation of a distributed in-memory data store. It was developed as an undergraduate project in distributed systems by Spencer Peterson, Adam Askari, Vineet Ramareddi, and Will Asper.

Key features include

• High availability: Designed to accept writes in the event of failed nodes.
• Eventual causal consistency: Database interactions will never violate the happens-before relationship, and replicas gossip new writes.
• Sharded: Data is partitioned among shards for scalability.
• Replicated: Data is replicated within shards for durability.
• Proxying: Nodes transparently proxy requests as needed to load balance traffic.

Build and Run

The binary can be compiled with a Go toolchain and

go install github.com/spencer-p/key-value-store

Alternatively, a Docker image can be built with

docker build path/to/key-value-store -t key-value-store:latest

To run, the binary must be supplied with a few environment variables:

1. ADDRESS. The IP address this node is running on.
2. PORT. The TCP port to bind to.
3. VIEW. A comma-separted list of addresses including in the cluster.
4. REPL_FACTOR. The number of replicas to assign per shard (integer).

API

The key value store exposes a CRUD API over HTTP.

Every request may supply a causal-context which the key value store will use to appropriately apply requests. Every response will return the most up-to-date causal-context which should be supplied in subsequent requests.

Create & Update

PUT /kv-store/keys/x HTTP/1.1
Host: 127.0.0.1
Content-type: application/json
Content-length: 36

{"value": "1", "causal-context": {}}

Delete

DELETE /kv-store/keys/x HTTP/1.1
Host: 127.0.0.1
Content-length: ???
{"causal-context": {insert-context-here}}

Read

GET /kv-store/keys/x HTTP/1.1
Host: 127.0.0.1
Content-length: ???
{"causal-context": {insert-context-here}}