gephi / graphstore

Fast in-memory graph structure, powering Gephi

GraphStore

GraphStore is an in-memory graph structure implementation written in Java. It's designed to be powerful, efficient and robust. It's powering the Gephi software and supports large graphs in intensive applications.

Features Highlight

• Blazing fast graph data structure optimized for reading and writing
• Comprehensive APIs to read and modify the graph structure
• Low memory footprint - reduced usage of Java objects and collections optimized for caching
• Supports directed, undirected and mixed graphs
• Supports parallel edges (i.e. edges can have a label)
• Any number of attributes can be associated with nodes or edges
• Thread-safe - Implements read-write locking mechanism to allow multiple reading threads
• Supports dynamic graphs (graphs over time)
• Built-in index on attribute values
• Fast and compact binary serialization
• Spatial indexing based on a quadtree

Download

Stable releases can be found on Maven central.

Development builds can be found on Sonatype's Snapshot Repository.

Documentation

API Documentation is available here.

Follow this QuickStart to get started.

Usage

From a Maven project

<dependency>
    <groupId>org.gephi</groupId>
    <artifactId>graphstore</artifactId>
    <version>0.7.2</version>
</dependency>

From a Gradle project

compile 'org.gephi:graphstore:0.7.2'

Dependencies

GraphStore is built for JRE 11+ and depends on FastUtil and Colt.

For a complete list of dependencies, consult the pom.xml file.

Developers

How to build

GraphStore uses Maven for building.

> mvn clean install

Note that code formatting is automatically applied at that time.

How to test

> mvn test

How to obtain code coverage report

> mvn jacoco:report

Contribute

The source code is available under the Apache 2.0 license. Contributions are welcome.

graphstore's Issues

Element.add/removeTimestamp() should update column version

Add TableDiff capability on TableObserver

Indices aren't correctly maintained in views

Attribute and timestamp indices should be maintained when the graph or data changes. That should also be true for views.

EdgeImpl.setWeight(double, double) has no effect

TimestampMap's put and remove methods should return boolean

Add way to create new GraphModel instances directly through the interface

Serialization should also serialize the Configuration

Make TimestampInternalMap track objects and values count

Adding after removal in TimestampValueSets isn't working

Remove open ability to Interval class

Not really needed at this point

AttributeUtils.isNumberType() fails with array types

Return false instead of true with int[], float[], Integer[] etc...

Timestamp index isn't automatically created for views on getIndex()

IndexImpl.size() is wrong when columns have been removed

Default column type for node/edge is Object.class but it is unsupported by AttributeUtils

For example, this causes that when AttributeUtils.isNumberType(column.getTypeClass()) is called on an id column, an exception is thrown.

Graph.getEdge() not working within views

Index.get(column, value) should return null when value is missing

Throws an NPE within LockableIterator

Graphstore currently can't be used with the graph-streaming plugin

Hi,

I've been working on some software that constructs graphs using graphstore. I also want to
use the graph-streaming plugin/libraries in my software, to be able to stream the graphs I'm building to Gephi clients. It seems like it is possible for me to swap out graphstore with the org.graph packages in gephi-toolkit. However, it still seems to be an issue that there are naming conflicts by having duplicate "org.graph" package names in projects from the gephi group. It also feels like it is the intention to have graphstore 'replace' org.graph in the toolkit, but hasn't been finished yet. I'm just generally confused because I don't know how efficient org.graph from the toolkit might be compared to graphstore and so on and so forth; there are differences, such as not having GraphListener and GraphEvent, which made me unable to use it in my project.

Make GraphFactory internal counters increment with self-entered ids

Add width and height in text properties

Add properties so the text dimensions can be processed by algorithms such as label adjust.

Element.setLabel() should increment the column version

Unable to remove attribute where attribute key is a string and attribute value is a boolean

Hi,

I'm able to set attributes where the attribute value is a string and the attribute key is a boolean, but then when I try to remove it I get a RunTimeException telling me that I'm not supposed to do that. However when I try to remove attributes with other value data types I don't seem to be having any problems.

Thanks!

Refactor TimestampValueSet

Rename timestamp value sets into timestamp value maps

Add Edge.removeWeight() when using timestamp weights

Add serialization hook in GraphModel

TimestampValueSets removal breaks underlying array consistency

Boolean index doesn't behave correctly with null values

The null set isn't taken in account so countValues(), values() and getValueSet() return wrong results.

Remove timestamp index indirection in TimestampSet and TimestampMap

This system should be kept in the Timestamp indices so keys can remain array indices but it adds to much overhead in the TimestampSet and TimestampMap classes. We shall directly use the (double) timestamp instead of the timestamp index. This will simplify the implementation and remove the index lookup in many places. It will also allow to keep the set/map sorted by timestamp rather than sorted by index, which will speed up some operations (e.g. estimators). Using double keys instead of integers will use twice more memory but I think it's worth for the speed and simplicity gain.

NPE in GraphViewDecorator when fetching elements that don't exist

Int array serialization isn't consistent with short values

ClassCastException when using Element.getAttribute(Column, GraphView) with the main graph view

Improve LongPacker with additional methods and documentation

Add similar methods with byte[] input.
Improve documentation.

Add a way to obtain columns from an Element (node/edge)

Now elements have a getAttributeKeys() method, but you cannot obtain the type class of each attribute directly from the element.

Add text bounds properties in TextProperties

This is necessary for algorithms such as LabelAdjust

Move Table.getEstimator() and Table.setEstimator() to the Column interface

Makes more sense in the Column

TimestampValueSet.getTimestamps() is wrong after the removal of the last element

Column's table is null after deserialization

Add default estimator in GraphConfiguration instead of hardcoding Estimator.FIRST

Set default estimator in dynamic column

It was null so far, keep it at null for static columns but Estimator.FIRST for dynamic columns

Introduce deepEquals/deepHashCode paradigm to guarantee immutability

Multiple internal implementations (TableImpl, NodeStore, GraphViewImpl) override equals() and hashcode() but shouldn't as these objects aren't immutable and could potentially be used as keys in hash tables. In most cases, this is only needed for testing purposes (e.g. serialization tests) and could easily be replaced by deepEquals/deepHashCode.

Add Graph.getModel()

Utility method to retrieve the model the graph belongs to

Reorganize packages

Simplify packages so only the following five remains:

• org.gephi.graph.api
• org.gephi.graph.api.types
• org.gephi.graph.spi
• org.gephi.graph.impl
• org.gephi.graph.impl.utils

TextPropertiesImpl color handling return wrong values

Wrong values after using setR/setG/setB on TextPropertiesImpl

Table.hasColumn(String id) is not consistent with added column ids regarding case insensitivity

If you try to add a column with "Id" as id, an exception is thrown because a column with "id" as id exists (lower case is applied), but hasColumn("Id") returns false (lower case is not applied).

TimestampIndexImpl.get() isn't unlocking properly

Add Element.removeAttribute() method for dynamic attributes

Create graph bridge interface so columns, configurations and elements can be exchanged between graph models

Edge's dynamic weight should use its dedicated column timestamp map

Better handling in EdgeImpl with static/dynamic getWeight() usage

Edges can either have a static or a dynamic weight. A static weight is simply a double number and has a default value of 1.0. Dynamic weights are associated to timestamps and set by using setWeight(double, double). We enforce the usage of one or the other so it shouldn't be possible to get the static weight after a dynamic weight has been set and vice-versa.

NPE on IndexImpl.destroy()

Remove TimestampInternalMap usage in ColumnStore

This was originally intended to track timestamps per column but I don't think it's really needed at this point.