[
https://issues.apache.org/jira/browse/BEAM-3245?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17249495#comment-17249495
]
Boyuan Zhang commented on BEAM-3245:
------------------------------------
I'm looking into this issue in portable execution. It seems like this test
suite is disabled on all portable runners now. Based on current FnHarness
implementation, java SDK harness invokes teardown when exiting the instruction
loop. This requires the runner to wait for the Java SDK harness to terminate
from main() instead of killing the process directly. At lease for Dataflow, I
don't think Dataflow waits for the SDK harness to exit.
Should we revisit the API or the SDK harness implementation?
> Dataflow runner does not respect ParDo's lifecycle on case of exceptions
> ------------------------------------------------------------------------
>
> Key: BEAM-3245
> URL: https://issues.apache.org/jira/browse/BEAM-3245
> Project: Beam
> Issue Type: Bug
> Components: runner-dataflow
> Affects Versions: 2.2.0, 2.3.0
> Reporter: Ismaël Mejía
> Priority: P3
>
> The lifecycle of the DoFn is not respected in case of exception in any of the
> lifecycle methods after setup.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
