Daniel Halperin commented on BEAM-670:

The bug is actually at L618. We're calling nextBackoffMillis twice in a row 
instead of using the cached value.


Good catch. How did tests not get this?

> BigQuery TableRow inserter incorrectly handles nextBackOff millis
> -----------------------------------------------------------------
>                 Key: BEAM-670
>                 URL: https://issues.apache.org/jira/browse/BEAM-670
>             Project: Beam
>          Issue Type: Bug
>          Components: sdk-java-gcp
>            Reporter: Luke Cwik
>            Assignee: Daniel Halperin
>            Priority: Minor
> From:
> https://github.com/GoogleCloudPlatform/DataflowJavaSDK/commit/94d57207924ed8650cf3c97fccb2a45f27bcc6a3#commitcomment-19135952
> Also present in:
> https://github.com/apache/incubator-beam/pull/888/files#diff-f6d45f28c12083c9556bb410bde8b109R614
> The check is inverted, it should be nextBackoffMillis != BackOff.STOP
> Otherwise it causes Thread.sleep() to be called with value -1 which causes an 
> IllegalArgumentException exception to be thrown.

This message was sent by Atlassian JIRA

Reply via email to