数据模型

tinkerpop3

1. 数据模型设计总则

• Vertex采用PARTITION模式，均匀分布在各个ignite节点上。
• 尽可能保证Vertex的属性存储在所属的vertex所在的节点上。
• 尽可能保证Edge的属性存储在所属的Edge所在的节点上。
• 尽可能少的跨cache操作。
• 将Vertex和Edge用不同的cache存储。
  
  • 因为gremlin中产生GraphTraversalSource产生GraphTraversal时有两种方式g.V()和g.E()两种方式，也就是说，极可能需遍历所有的点也可能遍历所有的边。
  • 当获取一个vertex时，能很快的找到vertex对应的出边和入边（如g.V().outE()）；获取一个edge时，也能很快的找到edge对应的出点和入点(如g.E().outV())。
• 对外部索引有很好的支持。
• 便于OLTP的实现及可扩展性。

2. OLTP需要实现的接口

OLTP需要实现的接口无外乎包含以下几种类型：

• 对vertex、edge、property的增删改查操作。
• 获取vertex、edge或property的集合的Iterator。

2.1 Edge

    /**
     * Retrieve the vertex (or vertices) associated with this edge as defined by the direction.
     * If the direction is {@link Direction#BOTH} then the iterator order is: {@link Direction#OUT} then {@link Direction#IN}.
     *
     * @param direction Get the incoming vertex, outgoing vertex, or both vertices
     * @return An iterator with 1 or 2 vertices
     */
    public Iterator<Vertex> vertices(final Direction direction);

    @Override
    public <V> Iterator<Property<V>> properties(final String... propertyKeys);

2.2 Vertex

    /**
     * Add an outgoing edge to the vertex with provided label and edge properties as key/value pairs.
     * These key/values must be provided in an even number where the odd numbered arguments are {@link String}
     * property keys and the even numbered arguments are the related property values.
     *
     * @param label     The label of the edge
     * @param inVertex  The vertex to receive an incoming edge from the current vertex
     * @param keyValues The key/value pairs to turn into edge properties
     * @return the newly created edge
     */
    public Edge addEdge(final String label, final Vertex inVertex, final Object... keyValues);
    /**
     * Create a new vertex property. If the cardinality is {@link VertexProperty.Cardinality#single}, then set the key
     * to the value. If the cardinality is {@link VertexProperty.Cardinality#list}, then add a new value to the key.
     * If the cardinality is {@link VertexProperty.Cardinality#set}, then only add a new value if that value doesn't
     * already exist for the key. If the value already exists for the key, add the provided key value vertex property
     * properties to it.
     *
     * @param cardinality the desired cardinality of the property key
     * @param key         the key of the vertex property
     * @param value       The value of the vertex property
     * @param keyValues   the key/value pairs to turn into vertex property properties
     * @param <V>         the type of the value of the vertex property
     * @return the newly created vertex property
     */
    public <V> VertexProperty<V> property(final VertexProperty.Cardinality cardinality, final String key, final V value, final Object... keyValues);
    /**
     * Gets an {@link Iterator} of incident edges.
     *
     * @param direction  The incident direction of the edges to retrieve off this vertex
     * @param edgeLabels The labels of the edges to retrieve. If no labels are provided, then get all edges.
     * @return An iterator of edges meeting the provided specification
     */
    public Iterator<Edge> edges(final Direction direction, final String... edgeLabels);
    /**
     * Gets an {@link Iterator} of adjacent vertices.
     *
     * @param direction  The adjacency direction of the vertices to retrieve off this vertex
     * @param edgeLabels The labels of the edges associated with the vertices to retrieve. If no labels are provided,
     *                   then get all edges.
     * @return An iterator of vertices meeting the provided specification
     */
    public Iterator<Vertex> vertices(final Direction direction, final String... edgeLabels);
    /**
     * {@inheritDoc}
     */
    @Override
    public <V> Iterator<VertexProperty<V>> properties(final String... propertyKeys);

2.3 VertexProperty

    /**
     * {@inheritDoc}
     */
    @Override
    public <U> Iterator<Property<U>> properties(final String... propertyKeys);

2.4 Element

    /**
     * Gets the unique identifier for the graph {@code Element}.
     *
     * @return The id of the element
     */
    public Object id();
    /**
     * Gets the label for the graph {@code Element} which helps categorize it.
     *
     * @return The label of the element
     */
    public String label();
    /**
     * Get the graph that this element is within.
     *
     * @return the graph of this element
     */
    public Graph graph();
    /**
     * Add or set a property value for the {@code Element} given its key.
     */
    public <V> Property<V> property(final String key, final V value);
    /**
     * Get an {@link Iterator} of properties.
     */
    public <V> Iterator<? extends Property<V>> properties(final String... propertyKeys);
    /**
     * Removes the {@code Element} from the graph.
     */
    public void remove();

3. 数据模型采用Object

KuazVertex

public class KuazVertex {
    private Object id;
    private String label;
    private List<Object> outEdgeIds;
    private List<Object> inEdgeIds;
}

KuazEdge

public class KuazEdge {
    private Object id;
    private String label;
    private Object outVertexId;
    private Object inVertexId;
}

KuazProperty

public class KuazProperty {
    private static String value = "value";
    private static String property = "property";
    /**
     * 属性所属的点或边的id.
     */
    private Object elementId;
    /**
     * 属性id.
     */
    private String propertyId;
    /**
     * 属性的类型是key/value还是属性中有属性。
     */
    private String propertyType = value;
    /**
     * 属性名
     */
    private String propertyKey;
    /**
     * 属性值
     */
    private Object propertyValue;
    /**
     * 用于存多值属性和属性中的属性的id。
     */
    private List<Object> proerties;
}

4. 数据模型采用Map

5. 个人建议的数据模型设计

• 采用 Object + Map 的设计设计方案。
• 只建立vertexCache和edgeCache两个Cache。
• vertex和edge的属性分别存在对应的Object中的Map中。

5.1 优点

• 满足g.V()和g.E()开始的两种GraphTraversal。
• 保证Vertex和Edge的属性存储在所属的vertex或Edge所在的节点上。
• 比较少的跨cache操作。
• 方便建立属性对点或边的索引。
• 解决了Object临时添加属性的缺陷。
• 对边affinity到点所在的节点上，可以采用ignite原生的Ojbect的affinity特性。

5.2 数据模型

5.2.1 第一版

public class KuazEdge {
    private Object id;
    @QuerySqlField(index = true)
    private String label;
    @AffinityKeyMapped
    private Object outVertexId;
    @AffinityKeyMapped
    private Object inVertexId;
    private Map<String, Object> properties;
}

public class KuazVertex {
    private Object id;
    @QuerySqlField(index = true)
    private String label;
    private List<Object> outEdgeIds;
    private List<Object> inEdgeIds;
    private Map<String, Map<String, Object>> properties;
}

5.2.2 第二版

public class KuazEdge {
    private Object id;
    private String label;
    private Object outVid;
    private Object inVid;
    private List<Property> properties;
}

public class KuazVertex {
    private Object id;
    private String label;
    private Object outEid;
    private Object inEid;
    private List<VertexProperty> property;
}

public class Property {
    protected Object id;
    protected String key;
    protected String value;
}

public class VertexProperty extends Property {
    private List<Property> properties;
}

5.2.3 第三版

public class KuazEdge {
    private Object id;
    private String label;
    private Object outVid;
    private Object inVid;
    private List<Property> properties;
}

public class KuazVertex {
    private Object id;
    private String label;
    private KuazEdge outEdge;
    private KuazEdge inEdge;
    private List<VertexProperty> property;
}

public class Property {
    protected Object id;
    protected String key;
    protected String value;
}

public class VertexProperty extends Property {
    private List<Property> properties;
}

5.2.3 第四版

public class KuazEdge {
    private Object id;
    private String label;
    private Object outVid;
    private Object inVid;
    private List<Property> properties;
}

public class KuazVertex {
    private Object id;
    private String label;
    private KuazEdge outEdge;
    private KuazEdge inEdge;
    private List<VertexProperty> property;
}

public class Property {
    protected Object id;
    protected String key;
    protected String value;
}

public class VertexProperty extends Property {
    private List<Property> properties;
}

说明：采用两个edgeCache来存储KuzaEdge，一个cache跟出点做并置，另一个cache跟入点做并置。此设计是针对优化computer进行优化的。

6. 索引

• 对所有label建立索引可以对hasLabel或需要通过label来查询的指令，如：

g.V().hasLabel('person').count()

• 对定义的schema中的属性需要添加索引的需建立索引。
• 需定义index API