1.自定义背景

dubbo2.7.5之后的版本，对线程池做了优化。具体实现在DefaultExecutorRepository类中。

  1. public class DefaultExecutorRepository implements ExecutorRepository {
  2.     private static final Logger logger = LoggerFactory.getLogger(DefaultExecutorRepository.class);
  4.     private int DEFAULT_SCHEDULER_SIZE = Runtime.getRuntime().availableProcessors();
  6.     private final ExecutorService SHARED_EXECUTOR = Executors.newCachedThreadPool(new NamedThreadFactory("DubboSharedHandler", true));
  8.     private Ring<ScheduledExecutorService> scheduledExecutors = new Ring<>();
  10.     private ScheduledExecutorService serviceExporterExecutor;
  12.     private ScheduledExecutorService reconnectScheduledExecutor;
  14.     private ConcurrentMap<String, ConcurrentMap<Integer, ExecutorService>> data = new ConcurrentHashMap<>();
  16.     public DefaultExecutorRepository() {
  17. //        for (int i = 0; i < DEFAULT_SCHEDULER_SIZE; i++) {
  18. //            ScheduledExecutorService scheduler = Executors.newSingleThreadScheduledExecutor(new NamedThreadFactory("Dubbo-framework-scheduler"));
  19. //            scheduledExecutors.addItem(scheduler);
  20. //        }
  21. //
  22. //        reconnectScheduledExecutor = Executors.newSingleThreadScheduledExecutor(new NamedThreadFactory("Dubbo-reconnect-scheduler"));
  23.         serviceExporterExecutor = Executors.newScheduledThreadPool(1, new NamedThreadFactory("Dubbo-exporter-scheduler"));
  24.     }
  25. }

线程池维护在私有成员变量data中，但是整个DefaultExecutorRepository并没有对外暴露data。网上提供的方案，基本都是利用反射。

2.自定义XyDefaultExecutorRepository

代码与DefaultExecutorRepository相同，添加一个自定义的暴露data的方法。

  1. public ConcurrentMap<String, ConcurrentMap<Integer, ExecutorService>> getExecutors() {
  2.     return data;
  3. }

�3.配置dubboSPI加载自定义ExecutorRepository

在resources资源目录下，添加META-INF/dubbo/目录。

将原来dubbo的扩展文件org.apache.dubbo.common.threadpool.manager.ExecutorRepository拷贝过来，设置自己的自定义ExecutorRepository。

  1. default=com.xy.rbac.api.config.XyDefaultExecutorRepository

实际包名用自己真是包名。