[
https://issues.apache.org/jira/browse/FLINK-33443?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17789859#comment-17789859
]
huiyang chi edited comment on FLINK-33443 at 11/27/23 3:35 PM:
---------------------------------------------------------------
Hi Martijn : I can understand that currently this problem is not exposed in the
CI, because this kind of 80% problem we detected exposed mainly because the
change of the test environment (and more dangerous for lower-version JDK),
while for the CI the test environment won't change. In detail, ~50% of problem
we address is because the non-deterministic of the Hash (Set and Map), this is
one example Apache repo
[wicket|https://github.com/apache/wicket/commit/ed64e166dcba6715eafcbb7ca460d2b87e84cffc]
had encountered and addressed, this is kind of problem resolved for the
long-term gain :) . Here is a list of this kind of this problem :
https://github.com/TestingResearchIllinois/idoft
was (Author: JIRAUSER303032):
Hi Martijn : I can understand that currently this problem is not exposed in the
CI, because this kind of 80% problem we detected exposed mainly because the
change of the test environment (and more dangerous for lower-version JDK),
while for the CI the test environment won't change. In detail, ~50% of problem
we address is because the non-deterministic of the Hash (Set and Map), this is
one example Apache repo
[wicket|https://github.com/apache/wicket/commit/ed64e166dcba6715eafcbb7ca460d2b87e84cffc]
had encountered and addressed, this is kind of problem resolved for the
long-term gain :) . I also have some tests detected fixed:
https://github.com/MyEnthusiastic/flink/pulls
> Make the test "testWriteComplexType" stable
> -------------------------------------------
>
> Key: FLINK-33443
> URL: https://issues.apache.org/jira/browse/FLINK-33443
> Project: Flink
> Issue Type: Bug
> Components: Runtime / State Backends
> Reporter: Krishna Anandan Ganesan
> Priority: Minor
>
> We are proposing to make the following test stable:
> {code:java}
> org.apache.flink.connectors.hive.HiveRunnerITCase.testWriteComplexType{code}
> *STEPS TO REPRODUCE THE ISSUE:*
> * The following command can be run to execute the test with the
> [NonDex|https://github.com/TestingResearchIllinois/NonDex] plugin:
> {code:java}
> mvn -pl flink-connectors/flink-connector-hive
> edu.illinois:nondex-maven-plugin:2.1.1:nondex
> -Dtest=org.apache.flink.connectors.hive.HiveRunnerITCase#testWriteComplexType
> {code}
> * The following error will be encountered:
> {code:java}
> [ERROR] Failures:
> [ERROR] HiveRunnerITCase.testWriteComplexType:166
> expected: "[1,2,3] {1:"a",2:"b"} {"f1":3,"f2":"c"}"
> but was: "[1,2,3] {2:"b",1:"a"} {"f1":3,"f2":"c"}"
> [INFO]
> [ERROR] Tests run: 1, Failures: 1, Errors: 0, Skipped: 0{code}
> *ROOT CAUSE ANALYSIS:*
> The test is currently flaky because of the assumption that the order of
> elements received in the _result_ variable will be consistent. There are
> currently two versions of query output that can be stored in _result._
> # The actual order that is expected where the output of the map attribute is
> \{1: "a", 2: "b"}.
> # The other order is the one shown in the error extract above where the
> ordering of the map attribute from the table is \{2: "b", 1: "a"}.
> *POTENTIAL FIX:*
> * The fix that I can suggest/have ready to raise a PR for is introducing
> another assertion on the second variant of the query output.
> * By asserting on whether the contents in _result_ are in one of the two
> orders, we can ascertain that the expected attributes with their contents are
> received as expected should the order in which they are received, not matter.
> Please share your thoughts on this finding and let me know if any other
> potential fix is possible for this test.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
