[
https://issues.apache.org/jira/browse/HBASE-7384?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13540276#comment-13540276
]
Enis Soztutar commented on HBASE-7384:
--------------------------------------
This tool is useful, but as Nick pointed out, we should really be using
CountDownLatch'es for test synchronization. The problem, however, is that most
of the time, the latch release should happen from the main code, and there is
no generic injection mechanism yet to do this from the tests cleanly. For
example the patch in HBASE-5494 introduces a class called InjectionHandler for
doing this, but not sure if it is the best interface.
Above is out of the scope for this isssue, but I think we should keep that in
mind.
> Introducing waitForCondition function into test cases
> -----------------------------------------------------
>
> Key: HBASE-7384
> URL: https://issues.apache.org/jira/browse/HBASE-7384
> Project: HBase
> Issue Type: Test
> Components: test
> Reporter: Jeffrey Zhong
> Assignee: Jeffrey Zhong
> Labels: test
> Fix For: 0.96.0
>
> Attachments: hbase-7384_1.0.patch, hbase-7384.patch, Waiter.java
>
>
> Recently I'm working on flaky test cases and found we have many places using
> while loop and sleep to wait for a condition to be true. There are several
> issues in existing ways:
> 1) Many similar code doing the same thing
> 2) When time out happens, different errors are reported without explicitly
> indicating a time out situation
> 3) When we want to increase the max timeout value to verify if a test case
> fails due to a not-enough time out value, we have to recompile & redeploy code
> I propose to create a waitForCondition function as a test utility function
> like the following:
> {code}
> public interface WaitCheck {
> public boolean Check() ;
> }
> public boolean waitForCondition(int timeOutInMilliSeconds, int
> checkIntervalInMilliSeconds, WaitCheck s)
> throws InterruptedException {
> int multiplier = 1;
> String multiplierProp = System.getProperty("extremeWaitMultiplier");
> if(multiplierProp != null) {
> multiplier = Integer.parseInt(multiplierProp);
> if(multiplier < 1) {
> LOG.warn(String.format("Invalid extremeWaitMultiplier
> property value:%s. is ignored.", multiplierProp));
> multiplier = 1;
> }
> }
> int timeElapsed = 0;
> while(timeElapsed < timeOutInMilliSeconds * multiplier) {
> if(s.Check()) {
> return true;
> }
> Thread.sleep(checkIntervalInMilliSeconds);
> timeElapsed += checkIntervalInMilliSeconds;
> }
> assertTrue("WaitForCondition failed due to time out(" +
> timeOutInMilliSeconds + " milliseconds expired)",
> false);
> return false;
> }
> {code}
> By doing the above way, there are several advantages:
> 1) Clearly report time out error when such situation happens
> 2) Use System property extremeWaitMultiplier to increase max time out
> dynamically for a quick verification
> 3) Standardize current wait situations
> Pleas let me know what your thoughts on this.
> Thanks,
> -Jeffrey
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
