Configuration for different versions of the Java SDK-MaxCompute(MaxCompute)-阿里云帮助中心

The MaxCompute Graph SDK for Java provides the classes you need to build, configure, and submit graph processing jobs on MaxCompute. Each class plays a specific role and is registered on a central GraphJob object before the job is submitted.

Add the Maven dependency

Search for odps-sdk-graph in the Maven repository to find the latest version, then declare the dependency in your project:

<dependency>
    <groupId>com.aliyun.odps</groupId>
    <artifactId>odps-sdk-graph</artifactId>
    <version>0.20.7-public</version>
</dependency>

SDK classes

Class	Description	GraphJob configuration method
`GraphJob`	Defines, submits, and manages a graph job. Inherits `JobConf`.	— (entry point)
`Vertex`	Defines a vertex. Properties: `id`, `value`, `halted`, `edges`.	`setVertexClass`
`Edge`	Defines an edge. Properties: `destVertexId`, `value`. MaxCompute Graph uses the adjacency list — a vertex's outbound edges are stored in its `edges` property.	Defined as part of `Vertex`
`GraphLoader`	Loads a graph.	`setGraphLoaderClass`
`VertexResolver`	Handles conflicts during graph topology modification. Register separate instances for graph loading and iterative computing.	`setLoadingVertexResolverClass`, `setComputingVertexResolverClass`
`Partitioner`	Distributes vertices across workers to enable parallel computing. The default `HashPartitioner` assigns each vertex by computing `hash(vertexId) % numWorkers`.	`setPartitionerClass`
`WorkerComputer`	Runs custom logic when a worker starts and stops — useful for initializing shared state or flushing results.	`setWorkerComputerClass`
`Aggregator`	Processes and summarizes global information across all workers. You can register multiple `Aggregator` classes in the same job.	`setAggregatorClass(Class…)`
`Combiner`	Summarizes the output records with the same key.	`setCombinerClass`
`Counters`	Tracks counts across workers. Get a `Counters` instance from `WorkerContext` and increment it; the framework sums values across all workers automatically.	Accessed via `WorkerContext`
`WorkerContext`	Provides access to all framework features within a superstep: modify graph topology, send messages, write results, and read resources.	Injected by the framework