RpcServerFactoryPBImpl force yarn users to define a protocol in a entailed namespace

[coderplay]

It's not a bug of Dragon, but is of Hadoop Yarn.

We defined wire protocol use protobuf with the java package name "org.apache.hadoop.realtime.proto"

option java_package = "org.apache.hadoop.realtime.proto";

Such definition would cause a ClassNotFoundException when starting an application master of Dragon.

12/05/29 14:45:33 ERROR app.DragonAppMaster: Error starting DragonAppMaster
org.apache.hadoop.yarn.YarnException: Failed to Start org.apache.hadoop.realtime.client.app.DragonAppMaster
    at org.apache.hadoop.yarn.service.CompositeService.start(CompositeService.java:78)
    at org.apache.hadoop.realtime.client.app.DragonAppMaster.start(DragonAppMaster.java:155)
    at org.apache.hadoop.realtime.client.app.DragonAppMaster$1.run(DragonAppMaster.java:218)
    at java.security.AccessController.doPrivileged(Native Method)
    at javax.security.auth.Subject.doAs(Subject.java:396)
    at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1177)
    at org.apache.hadoop.realtime.client.app.DragonAppMaster.initAndStartAppMaster(DragonAppMaster.java:214)
    at org.apache.hadoop.realtime.client.app.DragonAppMaster.main(DragonAppMaster.java:200)
Caused by: org.apache.hadoop.yarn.YarnException: Failed to load class: [org.apache.hadoop.yarn.proto.DragonClientProtocol$DragonClientProtocolService]
    at org.apache.hadoop.yarn.factories.impl.pb.RpcServerFactoryPBImpl.getServer(RpcServerFactoryPBImpl.java:105)
    at org.apache.hadoop.yarn.ipc.HadoopYarnProtoRPC.getServer(HadoopYarnProtoRPC.java:63)
    at org.apache.hadoop.realtime.client.DragonClientService.start(DragonClientService.java:134)
    at org.apache.hadoop.yarn.service.CompositeService.start(CompositeService.java:68)
    ... 7 more
Caused by: java.lang.ClassNotFoundException: org.apache.hadoop.yarn.proto.DragonClientProtocol$DragonClientProtocolService
    at java.net.URLClassLoader$1.run(URLClassLoader.java:202)
    at java.security.AccessController.doPrivileged(Native Method)
    at java.net.URLClassLoader.findClass(URLClassLoader.java:190)
    at java.lang.ClassLoader.loadClass(ClassLoader.java:307)
    at sun.misc.Launcher$AppClassLoader.loadClass(Launcher.java:301)
    at java.lang.ClassLoader.loadClass(ClassLoader.java:248)
    at java.lang.Class.forName0(Native Method)
    at java.lang.Class.forName(Class.java:247)
    at org.apache.hadoop.conf.Configuration.getClassByName(Configuration.java:1162)
    at org.apache.hadoop.yarn.factories.impl.pb.RpcServerFactoryPBImpl.getServer(RpcServerFactoryPBImpl.java:103)
    ... 10 more

RpcServerFactoryPBImpl hard coded namespace suffix and class suffix of every protocol we defined. It force yarn users to define a protocol in a entailed namespace, below is some lines of RpcServerFactoryPBImpl.java that issues in the bug.

  private static final String PROTO_GEN_PACKAGE_NAME = "org.apache.hadoop.yarn.proto";
  private static final String PROTO_GEN_CLASS_SUFFIX = "Service";
  private static final String PB_IMPL_PACKAGE_SUFFIX = "impl.pb.service";
  private static final String PB_IMPL_CLASS_SUFFIX = "PBServiceImpl";

//...

  private String getProtoClassName(Class<?> clazz) {
    String srcClassName = getClassName(clazz);
    return PROTO_GEN_PACKAGE_NAME + "." + srcClassName + "$" + srcClassName + PROTO_GEN_CLASS_SUFFIX;  
  }