Details
-
New Feature
-
Status: Closed
-
Major
-
Resolution: Incomplete
-
None
-
None
-
None
-
None
Description
Issue:
Avro objects are not serialized by Kryo causing the Avro GenericRecord to not be available to downstream operators if users don't explicitly mark the stream locality at container_local or thread_local.
Solution:
This JIRA is used to create a Module on top of AvroFileInputOperator and AvroToPojo operators such that downstream operators will access POJO instead of Avro GenericRecord. It, therefore, removes the exposure of GenericRecord to downstream operators and instead exposes the created POJO to downstream operators.
In this Module, the stream between the two encapsulated operators (AvroFileInputOperator and AvroToPojo) is set to CONTAINER_LOCAL.
Along with this new module, existing avro support files are moved from contrib module to a new 'avro' module.
---------------------------------------------------------------------------------------------------------------------------------------------------------------------
Unit Test
Unit test for this Avro module has been added in malhar-avro package.
Move to new package and Backward compatibility
Additionally, this module is part of a new package 'malhar-avro' and the operator files/tests are all moved from contrib package to the new package.
Old operator files are marked deprecated and made to extend from new operator files for backward compatibility.
Creating a new maven module for Avro is in accordance with the JIRA "https://issues.apache.org/jira/browse/APEXMALHAR-1843."
Git history of all the moved files is maintained
Application Level Testing
- To test the module, I created a sample StreamingApplication and a POJO class. This application adds the new AvroToPojoModule, and ConsoleOperator to the DAG. ConsoleOperator received and displayed POJO from the module
To test backward compatibility, I created sample application which adds AvroFileInputOperator and AvroToPojo from the old package to the DAG. It also adds ConsoleOperator to the DAG. ConsoleOperator received and displayed POJO from the module