[ 
https://issues.apache.org/jira/browse/LUCENE-9818?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17292591#comment-17292591
 ] 

Robert Muir commented on LUCENE-9818:
-------------------------------------

Besides just making them faster, i'm thinking these mega-suites aren't great 
for test load balancing and so on either. For someone that has a lot of cores, 
it might be better if a few of them got split up. TestIndexWriter is a classic 
offender, in the past we started splitting it up into subclasses to help 
parallel tests:
{noformat}
TestIndexWriterCommit.java         TestIndexWriterExceptions2.java  
TestIndexWriter.java             TestIndexWriterMerging.java       
TestIndexWriterOnVMError.java             TestIndexWriterUnicode.java
TestIndexWriterConfig.java         TestIndexWriterExceptions.java   
TestIndexWriterLockRelease.java  TestIndexWriterNRTIsCurrent.java  
TestIndexWriterOutOfFileDescriptors.java  TestIndexWriterWithThreads.java
TestIndexWriterDeleteByQuery.java  TestIndexWriterForceMerge.java   
TestIndexWriterMaxDocs.java      TestIndexWriterOnDiskFull.java    
TestIndexWriterReader.java
TestIndexWriterDelete.java         TestIndexWriterFromReader.java   
TestIndexWriterMergePolicy.java  TestIndexWriterOnJRECrash.java    
TestIndexWriterThreadsToSegments.java
{noformat}

> add option or hack or patch to see "slowest test suites"
> --------------------------------------------------------
>
>                 Key: LUCENE-9818
>                 URL: https://issues.apache.org/jira/browse/LUCENE-9818
>             Project: Lucene - Core
>          Issue Type: Improvement
>            Reporter: Robert Muir
>            Priority: Major
>
> Currently we print the N slowest test methods at the end of the run, which is 
> great. 
> Can we add some way (even if hacky and just a patch we don't commit) to see 
> the N slowest *suites*? Ideally it would sum the times of all test methods 
> and beforeClass/afterClass stuff too.
> It would help spot test classes doing pathological indexing in beforeClass(), 
> which isn't captured today I think? Also you can spot big test classes as a 
> whole, which might mean there is something inefficient about the design of 
> the whole class (e.g. it should use beforeClass() instead of setup() if all 
> methods just use the same index). Maybe each method takes 2s but it has 50 of 
> them, for example.
> cc [~dweiss] just an idea I had while working on LUCENE-9817 



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org

Reply via email to