Eugene Kirpichov created BEAM-3261:
--------------------------------------

             Summary: Apex runner does not detect pipeline failure
                 Key: BEAM-3261
                 URL: https://issues.apache.org/jira/browse/BEAM-3261
             Project: Beam
          Issue Type: Bug
          Components: runner-apex
            Reporter: Eugene Kirpichov
            Assignee: Thomas Weise
            Priority: Blocker


I was looking at https://github.com/apache/beam/pull/4074/files and asked 
myself "Don't we already have a ValidatesRunner test for this?"

Turns out we do: 
https://github.com/apache/beam/blob/3b79b6298e84711528b5ad1302200cb8acbac07e/sdks/java/core/src/test/java/org/apache/beam/sdk/transforms/ViewTest.java#L793

I ran this test with TestApexRunner, and I observed the same exception as in 
https://stackoverflow.com/questions/46982827/error-when-using-side-input-withsideinputs-method-not-accepting-kv-type-as-inp?noredirect=1#comment81040223_46982827
 , however the test passed.

It seems that it is passing because ApexRunnerResult.waitUntilFinish() detects 
only PAssert assertion errors, but not any other errors: 
https://github.com/apache/beam/blob/3b79b6298e84711528b5ad1302200cb8acbac07e/runners/apex/src/main/java/org/apache/beam/runners/apex/ApexRunnerResult.java#L62

This seems very problematic, as it means that 1) the status of Apex 
ValidatesRunner tests cannot be trusted: if they fail in any other way other 
than a PAssert error, the failure will be undetected, and 2) in general, a user 
can not trust a successful pipeline.run() from Apex runner.

For tests in particular, some other TestXRunner's guard against such a failure 
mode by verifying not only that there were no assertion failures, but also that 
all assertions succeeded - using metrics: e.g. 
https://github.com/apache/beam/blob/3b79b6298e84711528b5ad1302200cb8acbac07e/runners/google-cloud-dataflow-java/src/main/java/org/apache/beam/runners/dataflow/TestDataflowRunner.java#L248
 . Fixing this would be optimal, but meanwhile, the runner should at least fail 
the pipeline in case of error.



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

Reply via email to