Tom Hardman created BEAM-10233:
----------------------------------
Summary: Return error message/information with existing FAILED_ROW
data from BigQueryWriteFn (Python SDK)
Key: BEAM-10233
URL: https://issues.apache.org/jira/browse/BEAM-10233
Project: Beam
Issue Type: Improvement
Components: io-py-gcp
Reporter: Tom Hardman
A user may call `apache_beam.io.gcp.bigquery.WriteToBigQuery` to write their
streamed data to BQ. I any rows fail to write, this will return a tagged
pcollection `BigQueryWriteFn.FAILED_ROWS`. This data includes a tuple
`(destination_table, failed_row_payload)`.
My suggestion is to include the error information in the `FAILED_ROWS`
pcollection. From the source code we can see that we have access to the error
information, e.g. that the row failed because field `id` was `invalid` because
`this field is not a record`. I think we should surface this to the user.
I'm happy to open a PR this myself (as I've already had to overwrite the
original code in several projects), but it looks like we'd need a breaking
change by either extending the tuple which would cause unpacking issues in
existing code, or by returning a different data structure entirely.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
