[Travis issue] TestJsonStringToHoodieRecordMapFunction.testMapFunction failed

[yanghua]

[ERROR] TestJsonStringToHoodieRecordMapFunction.testMapFunction:82 expected: <10> but was: <8>

Travis log detail: https://api.travis-ci.com/v3/job/473125571/log.txt

@garyli1019 Does this issue triggered by this line? Can you check it?

hudi/hudi-flink/src/test/java/org/apache/hudi/source/TestJsonStringToHoodieRecordMapFunction.java

Line 73 in c931dc5

env.setParallelism(2);

[garyli1019]

@yanghua Not sure why this happened. This test was running fine for a while. Maybe related to the unstable Travis service today? Can we retry when Travis back to normal?

[vburenin]

@garyli1019 it is running totally fine in my environment, but I would still be a bit concerned about the potentially different outcome. If it differs to the point that it must not ever happen, then we potentially have a bug in Hudi.

[garyli1019]

@vburenin this test was just converting json to HoodieRecord, the workflow was not related to Hoodie. The sink function and mock environment were set up just for testing. My guess was the test environment was unstable on travis so the results become flaky when the parallelism is more than 1. If we want consistent results, we can change test parallelism to 1.

[vinothchandar]

This typically happens if the test was written in "data dependent" way and the random test data generator produces a different set of data on CI. I would audit the test a bit and see if we have some leads. It failed again on my land as well.
https://api.travis-ci.com/v3/job/473304345/log.txt

[nsivabalan]

Have created a tracking jira https://issues.apache.org/jira/browse/HUDI-1547

[nsivabalan]

Closing the issue as have a tracking jira. Do open a new issue if you find anything else. As always, thanks for helping improve Hudi :)