[ 
https://issues.apache.org/jira/browse/SOLR-5671?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13896840#comment-13896840
 ] 

Hoss Man commented on SOLR-5671:
--------------------------------

Given this...

bq. All the failures were using either Lucene41 or Lucene42 codec

... and the fact that (as far as i can tell) we haven't seen any more failures 
of this type since the codec constraints were added to the test in SOLR-5652, 
my hunch is that this is the same test bug manifesting itself in a slightly 
different way:  just as the non-deterministic behavior the docvalues w/missing 
value caused a doc to appear twice in SOLR-5652, shifting from it's originla 
position to a later position on a subsequent request, it's easily concievable 
that the same non-deterministic behavior could cause a doc to shift to an early 
position -- prior to the current cursor page -- on subsequent requests, causing 
that doc to be skipped entirely.

I think we can go ahead and resolve this -- if it does pop up again we can 
re-open (and we'll have the better assertion failure message steve added to 
help diagnose)

> Heisenbug #2 in DistribCursorPagingTest: full walk returns one fewer doc than 
> expected 
> ---------------------------------------------------------------------------------------
>
>                 Key: SOLR-5671
>                 URL: https://issues.apache.org/jira/browse/SOLR-5671
>             Project: Solr
>          Issue Type: Bug
>    Affects Versions: 4.7
>            Reporter: Steve Rowe
>             Fix For: 5.0, 4.7
>
>
> Twice on Uwe's Jenkins, DistribCursorPagingTest has paged through a small 
> number of indexed docs and retrieved one fewer doc than the number of indexed 
> docs.  Both of these failures were on trunk on Windows:
> http://jenkins.thetaphi.de/job/Lucene-Solr-trunk-Windows/3708/
> http://jenkins.thetaphi.de/job/Lucene-Solr-trunk-Windows/3713/  
> I've also seen this twice on trunk on my OS X laptop (out of 875 trials).
> None of the seeds have reproduced for me.
> All the failures were using either Lucene41 or Lucene42 codec



--
This message was sent by Atlassian JIRA
(v6.1.5#6160)

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

Reply via email to