[YARN-8664] ApplicationMasterProtocolPBServiceImpl#allocate throws NPE when NM is lost

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.8.2
• Fix Version/s: 2.8.6
• Component/s: resourcemanager
• Hadoop Flags: Reviewed
• Assignee: Jiandan Yang
• Reporter: Jiandan Yang
• Created: 14/Aug/18 14:20
• Updated: 12/Sep/18 07:21
• Resolved: 12/Sep/18 07:21

Description

ResourceManager logs about exception is:

2018-08-09 00:52:30,746 WARN [IPC Server handler 5 on 8030] org.apache.hadoop.ipc.Server: IPC Server handler 5 on 8030, call Call#305638 Retry#0 org.apache.hadoop.yarn.api.ApplicationMasterProtocolPB.allocate from 11.13.73.101:51083
java.lang.NullPointerException
        at org.apache.hadoop.yarn.proto.YarnProtos$ResourceProto.isInitialized(YarnProtos.java:6402)
        at org.apache.hadoop.yarn.proto.YarnProtos$ResourceProto$Builder.build(YarnProtos.java:6642)
        at org.apache.hadoop.yarn.api.records.impl.pb.ResourcePBImpl.mergeLocalToProto(ResourcePBImpl.java:254)
        at org.apache.hadoop.yarn.api.records.impl.pb.ResourcePBImpl.getProto(ResourcePBImpl.java:61)
        at org.apache.hadoop.yarn.api.records.impl.pb.NodeReportPBImpl.convertToProtoFormat(NodeReportPBImpl.java:313)
        at org.apache.hadoop.yarn.api.records.impl.pb.NodeReportPBImpl.mergeLocalToBuilder(NodeReportPBImpl.java:264)
        at org.apache.hadoop.yarn.api.records.impl.pb.NodeReportPBImpl.mergeLocalToProto(NodeReportPBImpl.java:287)
        at org.apache.hadoop.yarn.api.records.impl.pb.NodeReportPBImpl.getProto(NodeReportPBImpl.java:224)
        at org.apache.hadoop.yarn.api.protocolrecords.impl.pb.AllocateResponsePBImpl.convertToProtoFormat(AllocateResponsePBImpl.java:714)
        at org.apache.hadoop.yarn.api.protocolrecords.impl.pb.AllocateResponsePBImpl.access$400(AllocateResponsePBImpl.java:69)
        at org.apache.hadoop.yarn.api.protocolrecords.impl.pb.AllocateResponsePBImpl$6$1.next(AllocateResponsePBImpl.java:680)
        at org.apache.hadoop.yarn.api.protocolrecords.impl.pb.AllocateResponsePBImpl$6$1.next(AllocateResponsePBImpl.java:669)
        at com.google.protobuf.AbstractMessageLite$Builder.checkForNullValues(AbstractMessageLite.java:336)
        at com.google.protobuf.AbstractMessageLite$Builder.addAll(AbstractMessageLite.java:323)
        at org.apache.hadoop.yarn.proto.YarnServiceProtos$AllocateResponseProto$Builder.addAllUpdatedNodes(YarnServiceProtos.java:12846)
        at org.apache.hadoop.yarn.api.protocolrecords.impl.pb.AllocateResponsePBImpl.mergeLocalToBuilder(AllocateResponsePBImpl.java:145)
        at org.apache.hadoop.yarn.api.protocolrecords.impl.pb.AllocateResponsePBImpl.mergeLocalToProto(AllocateResponsePBImpl.java:176)
        at org.apache.hadoop.yarn.api.protocolrecords.impl.pb.AllocateResponsePBImpl.getProto(AllocateResponsePBImpl.java:97)
        at org.apache.hadoop.yarn.api.impl.pb.service.ApplicationMasterProtocolPBServiceImpl.allocate(ApplicationMasterProtocolPBServiceImpl.java:61)
        at org.apache.hadoop.yarn.proto.ApplicationMasterProtocol$ApplicationMasterProtocolService$2.callBlockingMethod(ApplicationMasterProtocol.java:99)
        at org.apache.hadoop.ipc.ProtobufRpcEngine$Server$ProtoBufRpcInvoker.call(ProtobufRpcEngine.java:447)
        at org.apache.hadoop.ipc.RPC$Server.call(RPC.java:989)
        at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:846)
        at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:789)
        at java.security.AccessController.doPrivileged(Native Method)
        at javax.security.auth.Subject.doAs(Subject.java:422)
        at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1804)
        at org.apache.hadoop.ipc.Server$Handler.run(Server.java:2457)

ApplicationMasterService#allocate will call AllocateResponse#setUpdatedNodes when NM losting, and AllocateResponse#getProto will call ResourceBPImpl#getProto to transform NodeReportPBImpl#capacity into format of PB . Because ResourcePBImpl is not thread safe and
multiple AM will call allocate at the same time, ResourcePBImpl#getProto may throw NullPointerException or UnsupportedOperationException.
I wrote a test code which can reproduce exception.

@Test
  public void testResource1() throws InterruptedException {
    ResourcePBImpl resource = (ResourcePBImpl) Resource.newInstance(1, 1);
    for (int i =0;i<10;i++ ) {
      Thread thread = new PBThread(resource);
      thread.setName("t"+i);
      thread.start();
    }
    Thread.sleep(100000000);
  }

  class PBThread extends Thread {
    ResourcePBImpl resourcePB;

    public PBThread(ResourcePBImpl resourcePB) {
      this.resourcePB = resourcePB;
    }

    @Override 
    public void run() {
      while(true) {
        this.resourcePB.getProto();
      }
    }
  }