[PIG-5115] Builtin AvroStorage generates incorrect avro schema when the same pig field name appears in the alias

• Type: Bug
• Status: Patch Available
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 0.17.0
• Fix Version/s: 0.18.0
• Patch Info: Patch Available
• Flags: Patch
• Assignee: Anyi Li
• Reporter: Anyi Li
• Created: 25/Jan/17 21:57
• Updated: 26/May/17 22:36

Description

Pig ResourceSchema allows to use same field names but different types when they are not in the same level. The pig schema like

data: {col1: (col2: (col1_data: chararray)),col2: {col2: (col2_data: chararray)}}

Although col2 has been redefined, they are not appeared in the same level, it is a totally valid pig schema.

However, once it is translated by AvroStorage, it will throw exception

Can't redefine: col2
        at org.apache.pig.newplan.logical.visitor.InputOutputFileValidatorVisitor.visit(InputOutputFileValidatorVisitor.java:64)
        at org.apache.pig.newplan.logical.relational.LOStore.accept(LOStore.java:66)
        at org.apache.pig.newplan.DepthFirstWalker.depthFirst(DepthFirstWalker.java:64)
        at org.apache.pig.newplan.DepthFirstWalker.depthFirst(DepthFirstWalker.java:66)
        at org.apache.pig.newplan.DepthFirstWalker.depthFirst(DepthFirstWalker.java:66)
        at org.apache.pig.newplan.DepthFirstWalker.walk(DepthFirstWalker.java:53)
        at org.apache.pig.newplan.PlanVisitor.visit(PlanVisitor.java:52)
        at org.apache.pig.newplan.logical.relational.LogicalPlan.validate(LogicalPlan.java:212)
        at org.apache.pig.PigServer$Graph.compile(PigServer.java:1767)
        at org.apache.pig.PigServer$Graph.access$300(PigServer.java:1443)
        at org.apache.pig.PigServer.execute(PigServer.java:1356)
        at org.apache.pig.PigServer.executeBatch(PigServer.java:415)
        at org.apache.pig.PigServer.executeBatch(PigServer.java:398)
        at org.apache.pig.tools.grunt.GruntParser.executeBatch(GruntParser.java:171)
        at org.apache.pig.tools.grunt.GruntParser.parseStopOnError(GruntParser.java:234)
        at org.apache.pig.tools.grunt.GruntParser.parseStopOnError(GruntParser.java:205)
        at org.apache.pig.tools.grunt.Grunt.exec(Grunt.java:81)
        at org.apache.pig.Main.run(Main.java:631)
        at org.apache.pig.Main.main(Main.java:177)
Caused by: org.apache.avro.SchemaParseException: Can't redefine: col2
        at org.apache.avro.Schema$Names.put(Schema.java:1042)
        at org.apache.avro.Schema$NamedSchema.writeNameRef(Schema.java:511)
        at org.apache.avro.Schema$RecordSchema.toJson(Schema.java:626)
        at org.apache.avro.Schema$ArraySchema.toJson(Schema.java:737)
        at org.apache.avro.Schema$UnionSchema.toJson(Schema.java:814)
        at org.apache.avro.Schema$RecordSchema.fieldsToJson(Schema.java:648)
        at org.apache.avro.Schema$RecordSchema.toJson(Schema.java:635)
        at org.apache.avro.Schema.toString(Schema.java:297)
        at org.apache.avro.Schema.toString(Schema.java:287)
        at org.apache.pig.builtin.AvroStorage.setOutputAvroSchema(AvroStorage.java:442)
        at org.apache.pig.builtin.AvroStorage.checkSchema(AvroStorage.java:433)
        at org.apache.pig.newplan.logical.visitor.InputOutputFileValidatorVisitor.visit(InputOutputFileValidatorVisitor.java:54)
        ... 18 more

It is caused by a bug in AvroStorageSchemaConversionUtilities class which uses tuple name as GenericRecord name as well as the fieldname that wraps the record.

So it would like to produces the avro schema like the following

{
  "type": "record",
  "name": "data",
  "fields": [
    {
      "name": "col1",
      "type": {
        "type": "record",
        "name": "col1_1",
        "fields": [
          {
            "name": "col2",
            "type": {
              "type": "record",
              "name": "col2",
              "fields": [
                {
                  "name": "col1_data",
                  "type": "string"
                }
              ]
            }
          }
        ]
      }
    },
    {
      "name": "col2",
      "type": {
        "type": "array",
        "items": {
          "type": "record",
          "name": "col2",
          "fields": [
            {
              "name": "col2_data",
              "type": "string"
            }
          ]
        }
      }
    }
  ]
}

But according to the avro 1.7.7 specs (https://avro.apache.org/docs/1.7.7/spec.html#Names), col2 has been defined as record and redefined as array later, it is an invalid schema, unless the fullname (namespace + name) is unique.

Since AvroStorageSchemaConversionUtilities will generate avro record if the pig schema is a tuple, we need a way to generate unique recordName.

public static Schema resourceSchemaToAvroSchema(final ResourceSchema rs,
      String recordName, final String recordNameSpace,
      final Map<String, List<Schema>> definedRecordNames,
      final Boolean doubleColonsToDoubleUnderscores) throws IOException {

    if (rs == null) {
      return null;
    }

    recordName = toAvroName(recordName, doubleColonsToDoubleUnderscores);

    List<Schema.Field> fields = new ArrayList<Schema.Field>();
    Schema newSchema = Schema.createRecord(
            recordName, null, recordNameSpace, false);

The AvroStorage class from piggybank solved this problem by defining a static method and generate unique recordName. We can implement the similar method for the builtin AvroStorage