Graph/Database

• Graph database

  • Overview

    • A graph database stores data as nodes and relationships (edges), with properties on both, so the primary question is what connects to what rather than which row matches this key.
    • It matters when the interesting fact is the link: a knowledge garden's wikilinks, a social graph, a bill of materials, or a fraud ring are cheaper to traverse as edges than to reconstruct with joins.

  • Context

    • Contrasts with a table store (rows and foreign keys) and with a document store (nested trees). A graph engine treats adjacency as a first-class index.
    • This garden already uses the metaphor: journaling is Chron/ological/Order; knowledge gardening clusters related pages the way a graph clusters neighbors — what goes where more than what happened.

  • Key Principles

    • Nodes and relationships are data — the edge is not an afterthought join; it has a type and can carry properties.
    • Local traversal — queries walk neighborhoods (friends-of-friends, citation chains) without scanning the whole store.
    • Index-free adjacency (ideal) — each node knows its neighbors directly; cost grows with the walk, not with table size.

  • Mechanism

    • Two common models: labeled-property graphs (nodes/edges with labels and maps of properties) and RDF triples (subject–predicate–object).
    • Query languages include Cypher / GQL, Gremlin, SPARQL, and vendor languages such as AQL.

  • Examples

    • Neo4j — native property graph; Cypher; the engine most cited in this garden.
    • ArangoDB — multi-model (document + graph + key-value).
    • Amazon Neptune — managed graph on AWS (property graph and RDF).
    • JanusGraph — distributed open-source graph over Cassandra/HBase, queried with Gremlin.

  • Misconceptions

    • "A graph database is a knowledge graph" — False; a knowledge graph is a modeled web of meaning; a graph database is one way to store it.
    • "Graphs replace SQL" — Oversimplified; they win on relationship-heavy queries and lose on simple tabular aggregates unless the product is multi-model.