[ 
https://issues.apache.org/jira/browse/YARN-6547?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16014403#comment-16014403
 ] 

Carlo Curino commented on YARN-6547:
------------------------------------

Hi [~wangda], thanks for looking at this. I updated the patch (v1).

Issue #1: Does not happen for me on proper runs (was happening due to other 
failures. Since the test is parametrized things might have bled across 
instances on failures), seems ok now.
Issue #2: This is related to YARN-6111, and to what I was mentioning in my 
first comment. We should finesse a set of invariants that are tight, yet 
robust. This should be a combination of traces, timeouts, and exit invariants 
for which we have good confidence (I need your help for this). For now I loosen 
up the {{exit-invariants.txt}} so it doesn't scream at RUMEN runs, but we 
should follow up and tightening where possible (maybe after YARN-6111 is 
addressed). 
Issue #3: This is a parametrized test issue (separate runs were fine), the 
stop/restart interacts poorly with MetricsSystem registration process (as the 
jvmMetrics objects persists across, so is not re-registered). I fixed it by 
ensuring that even if JvmMetrics already exist, we check whether it should be 
re-registered to the MetricsSystem. (Please double-check that code).


> Enhance SLS-based tests leveraging invariant checker
> ----------------------------------------------------
>
>                 Key: YARN-6547
>                 URL: https://issues.apache.org/jira/browse/YARN-6547
>             Project: Hadoop YARN
>          Issue Type: Bug
>            Reporter: Carlo Curino
>            Assignee: Carlo Curino
>         Attachments: YARN-6547.v0.patch, YARN-6547.v1.patch
>
>
> We can leverage {{InvariantChecker}}s to provide a more thorough validation 
> of SLS-based tests. This patch introduces invariants checking during and at 
> the end of the run.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to