Uploaded image for project: 'Apache Avro'
  1. Apache Avro
  2. AVRO-2780

ProtobufData and ThriftData Can Get Into Endless Loop

    XMLWordPrintableJSON

    Details

    • Type: Bug
    • Status: Closed
    • Priority: Critical
    • Resolution: Fixed
    • Affects Version/s: 1.9.2
    • Fix Version/s: 1.10.0
    • Component/s: java
    • Labels:
      None

      Description

      We have found a bug related, but probably not limited to ProtobufDatumReader. In Protobuf, if the ".proto" file has the same name as the message, in the generated ".java" file, "OuterClass" will be appended to the outer class's name. This is documented here: https://developers.google.com/protocol-buffers/docs/reference/java-generated#invocation

      Now we have a test protobuf class defined like so:

       

      syntax = "proto2";
      
      package test.avrobug;
      
      option java_package = "com.test.protos.avrobug";
      option java_generic_services = true;
      
      enum State {
        BAD = 0;
        VERY_BAD = 1;
        SUPER_BAD = 2;
      }
      
      message Dummy {
        optional string token = 1;
        optional string description = 2;
        optional int64 count = 3;
        optional State state = 4;
      }
      

       

       

      If we first create a Protobuf object, write it to a file as Avro through ProtobufDatumReader, then read it into a Protobuf with ProtobufDatumReader, a stack overflow exception will happen. Code to reproduce is as follows:

       

      @Test
      public void ProtoToAvroOuterClassBug() throws Exception {
        DummyOuterClass.Dummy dummy = DummyOuterClass.Dummy.newBuilder()
            .setCount(50)
            .setDescription("hehe")
            .setToken("abc123")
            .setState(DummyOuterClass.State.BAD)
            .build();
        
        ProtobufDatumWriter<DummyOuterClass.Dummy> pbWriter =
            new ProtobufDatumWriter<>(DummyOuterClass.Dummy.class);
        DataFileWriter<DummyOuterClass.Dummy> dataFileWriter = new DataFileWriter<>(pbWriter);
        Schema schema = ProtobufData.get().getSchema(DummyOuterClass.Dummy.class);
        dataFileWriter.create(schema,
            new File("/Users/me/Documents/DummyAvroNoDefaultValues.avro"));
        dataFileWriter.append(dummy);
        dataFileWriter.close();
      
        ProtobufDatumReader<DummyOuterClass.Dummy> pbReader =
            new ProtobufDatumReader<>(DummyOuterClass.Dummy.class);
        DataFileReader<DummyOuterClass.Dummy> dataFileReader =
            new DataFileReader<>(new File("/Users/me/Documents/DummyAvroNoDefaultValues.avro"),
                pbReader);
        while(dataFileReader.hasNext()) {
          DummyOuterClass.Dummy record = dataFileReader.next();
          String recordStr = record.toString();
          String originStr = dummy.toString();
          System.out.println(recordStr);
          System.out.println(originStr);
        }
      }
      

       

       

      When this is run, a stack overflow exception will happen with the following stack trace:

       

      java.lang.StackOverflowErrorjava.lang.StackOverflowError                      at java.base/java.util.concurrent.ConcurrentHashMap.get(ConcurrentHashMap.java:936)                      at org.apache.avro.specific.SpecificData.getClass(SpecificData.java:250)                      at org.apache.avro.protobuf.ProtobufData.newRecord(ProtobufData.java:141)                      at org.apache.avro.protobuf.ProtobufData.newRecord(ProtobufData.java:143)                      at org.apache.avro.protobuf.ProtobufData.newRecord(ProtobufData.java:143)                      at 
      ..................

      newRecord() is recursed infinitely until stack explodes

      I did a bit code tracing and I found out when Avro tries to load the corresponding Java class from the schema, it tries to load the class "com.test.protos.avrobug.Dummy.Dummy". I suspect the double "Dummy" at the end is related to the stack overflow, but I don't have enough knowledge on Proto or Avro to know the exact reason.

       

        Attachments

          Issue Links

            Activity

              People

              • Assignee:
                belugabehr David Mollitor
                Reporter:
                tlang1991 Tianyu Lang
              • Votes:
                0 Vote for this issue
                Watchers:
                6 Start watching this issue

                Dates

                • Created:
                  Updated:
                  Resolved: