[jira] [Created] (BEAM-3067) BigQueryIO.Write fails on empty PCollection with DirectRunner (batch job)

["Dmitry Bigunyak (JIRA)" <ji...@apache.org>]

Dmitry Bigunyak created BEAM-3067:
-------------------------------------

             Summary: BigQueryIO.Write fails on empty PCollection with DirectRunner (batch job)
                 Key: BEAM-3067
                 URL: https://issues.apache.org/jira/browse/BEAM-3067
             Project: Beam
          Issue Type: Bug
          Components: runner-direct, sdk-java-gcp
    Affects Versions: 2.1.0
         Environment: Arch Linux, Java 1.8.0_144
            Reporter: Dmitry Bigunyak
            Assignee: Thomas Groh


I'm using side output feature to filter out malformatted events (errors) from a stream of valid events. Then I save valid events into one BigQuery table and errors go into another dedicated table.
Here is the code for outputting error rows:

{code:java}
invalidEventRows.apply("WriteErrors", BigQueryIO.writeTableRows()
        .to(errorTableRef)
        .withSchema(ProcessEvents.getErrorSchema())
        .withCreateDisposition(BigQueryIO.Write.CreateDisposition.CREATE_IF_NEEDED)
        .withWriteDisposition(BigQueryIO.Write.WriteDisposition.WRITE_APPEND));
{code}

The problem is that when running on DirectRunner in a batch mode (reading input from a file) and {{invalidEventRows}} PCollection ends up being empty (all events are valid -- no errors), I get the following error:
{code}
[ERROR]   "status" : {
[ERROR]     "errorResult" : {
[ERROR]       "message" : "No schema specified on job or table.",
[ERROR]       "reason" : "invalid"
[ERROR]     },
[ERROR]     "errors" : [ {
[ERROR]       "message" : "No schema specified on job or table.",
[ERROR]       "reason" : "invalid"
[ERROR]     } ],
[ERROR]     "state" : "DONE"
[ERROR]   },
{code}

There are no errors when executing the same code and {{invalidEventRows}} PCollection is not empty, the BigQuery table is created and the data are correctly inserted.
Also everything seems to be working fine in a streaming mode (reading from Pub/Sub) on both DirectRunner and DataflowRunner.

Looks like a bug?
Or should I open an issue in GoogleCloudPlatform/DataflowJavaSDK github project?



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)