[BEAM-3798] Performance tests flaky due to Dataflow transient errors

• Type: Sub-task
• Status: Resolved
• Priority: P2
• Resolution: Fixed
• Fix Version/s: Not applicable
• Component/s: runner-dataflow
• Assignee: Lukasz Gajowy
• Reporter: Lukasz Gajowy
• Created: 07/Mar/18 15:22
• Updated: 24/Jul/20 20:15
• Resolved: 29/Mar/18 06:57

Description

Performance tests are flaky due to transient errors that happened during data processing (eg. SocketTimeoutException while connecting to DB). Currently exceptions that happen on Dataflow runner but are retried successfully, fail the test regardless of the final job state (giving a false-negative result).

Possible solution for batch scenarios:
We could "rethrow" exceptions that happened due to transient errors only if the job status is other than DONE.

Possible solution for streaming scenarios:
(don't know yet)

Link to discussion on dev list

Issue Links

• links to: GitHub Pull Request #4871