[ https://issues.apache.org/jira/browse/BEAM-1834?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15948060#comment-15948060 ]
Kevin Peterson commented on BEAM-1834: -------------------------------------- Gotcha. Yes, I'd like to request the ability to set data dependent schemas. OR just turn off the [validation|https://github.com/apache/beam/blob/master/sdks/java/io/google-cloud-platform/src/main/java/org/apache/beam/sdk/io/gcp/bigquery/BigQueryIO.java#L952] that prevents me from using CREATE_NEVER - I generate all of the tables before starting the pipeline, so I never actually need to create them. > Bigquery Write validation doesn't work well with ValueInSingleWindow > -------------------------------------------------------------------- > > Key: BEAM-1834 > URL: https://issues.apache.org/jira/browse/BEAM-1834 > Project: Beam > Issue Type: Bug > Components: sdk-java-gcp > Reporter: Kevin Peterson > > I am using the new {{Write<T> to(SerializableFunction<ValueInSingleWindow<T>, > String> tableSpecFunction)}} function to write data to different Bigquery > tables depending on the values. I'm my case, the values can have a different > schema (it starts as an {{Any}} encoded protobuf, which I parse and expand to > a {{TableRow}} object). > Since the tables have different schemas, the existing implementation of > {{withSchema}} doesn't work. > Some options: > # Allow {{CreateDisposition.CREATE_NEVER}} in this situation. Failed inserts > from a missing table just fail (and eventually pass through via BEAM-190). > # Add a new {{withSchema(SerializableFunction<ValueInSingleWindow<T>, > TableSchema>}} function. > I think eventually both of the above should be allowable configurations, but > just one will unblock my current error. Happy to implement, given some > guidance on design preferences. -- This message was sent by Atlassian JIRA (v6.3.15#6346)