[
https://issues.apache.org/jira/browse/HADOOP-6276?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12758415#action_12758415
]
Chris Douglas commented on HADOOP-6276:
---------------------------------------
The intent is to avoid writing tests with race conditions affecting success. A
\-1 from test-patch requires that one demonstrate that the sleep duration is
orthogonal to the outcome, e.g. that the slow writer is simply slow, and the
test is measuring the system's response to it. While writing flaky tests
doesn't require Thread.sleep, it's a lot easier with it. Polling should be OK
where we don't offer callbacks.
> test-patch script should -1 patches that increase the number of
> Thread.sleep() calls in tests
> ---------------------------------------------------------------------------------------------
>
> Key: HADOOP-6276
> URL: https://issues.apache.org/jira/browse/HADOOP-6276
> Project: Hadoop Common
> Issue Type: Improvement
> Components: test
> Reporter: Nigel Daley
> Assignee: Giridharan Kesavan
>
> Calls to Thread.sleep() in tests are likely dubious. There are rare
> occasions where these are required, but in general they lead to flaky unit
> tests that fail sporadically. We used to be much better at code reviewing
> these out of patches, but have lost ground on that in the last few years (and
> our flaky unit tests show that!).
> I propose we have the test-patch script ensure that a patch does not add to
> the number of Thread.sleep() calls in *tests* by giving a -1 to patches that
> do increase it. I suggest this is a simple grep for lines containing
> "Thread.sleep(" and a "+" character in the first column.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
