This is a masterless, quorumless distributed database, hardcoded for a specific task (a twitter clone).
The frontend assumes at least one member of the cluster is sitting on localhost:3004. You can change this by chainging the known_hosts array in frontend/db.py to point to your favorite DB host.
It chooses A and P over C, so try not to make conflicting queries :)
The clustering works via a ring topology. Basically, each node picks an ID on startup, and all nodes know the full set of nodes on the network. They then send to the node with the first id greater than theirs (with the highest node sending to the lowest node). When a node gets an update that it initiated, it drops it, otherwise it applys the update and passes it on to the next node in the sequence.
When passing an update, if a server cannot be reached, it's next partner is then tried. This is repeated until a living node is reached, at which point an update will first be generated informing the cluster of the now dead nodes.
Some useful scripts for cluster manipulation are in /scripts
Query structure:
- A full query structure BNF is included in backend/parser/parser.h
- Queries are responded to with a list of newline-separated replies, followed by `DONE\n"
- Multiple queries can be made in a single connection, but this is not currently used by the frontend
Debug messages can be turned on by defining the environment variable DEBUG before compiling the backend.
