[BEAM-13158] "Dead letter" handling for problem rows in BigQueryIO Storage Write API

• Type: New Feature
• Status: Open
• Priority: P2
• Resolution: Unresolved
• Component/s: io-java-gcp
• Assignee: Unassigned
• Reporter: Sergei Lilichenko
• Created: 01/Nov/21 16:06
• Updated: 04/Jun/22 22:26

Description

A single invalid row causes the BigQueryIO transform and the whole pipeline to fail. The desired behavior would be to allow control of the error handling - either fail on any validation failure (current behavior) or return the list of failed records through the WriteResult. 

There are two places where the exception occurs - Json to protobuf conversion and the BigQuery backend. 

Example of the exception caused by the conversion:

io.grpc.StatusRuntimeException: INVALID_ARGUMENT: The proto field mismatched with BigQuery field at D586b3f9a_1543_4dbe_87ff_ef786d6803c2.bytes_sent, the proto field type string, BigQuery field type INTEGER Entity: projects/event-processing-demo/datasets/bigquery_io/tables/events/streams/Cic2MzUyMTYxYy0wMDAwLTI2MjktOGVjYy1mNDAzMDQ1ZWY5Y2U6czI

Example of the exception caused by the BigQuery backend: 

io.grpc.StatusRuntimeException: INVALID_ARGUMENT: Field dst_ip: STRING(15) has maximum length 15 but got a value with length 54 Entity: projects/event-processing-demo/datasets/bigquery_io/tables/events/streams/CiQ2MzRkOGM5Mi0wMDAwLTI2MjktOGVjYy1mNDAzMDQ1ZWY5Y2U