OK the failure is because of the interaction of CMS and LogMP.
I think we should just cut this test back to SMS, since it's
specifically testing the invariants of LogMP in a controlled
setting. I'll commit that...
The problem happens because a BG merge finishes before (in time) the
merges
Turns out this is not easily reproducible. While I am able to reproduce it
*many times* on my machine, it's not *always*. I did notice a strange
behavior -- when the test fails, the stack trace points to what seems to be
an incorrect line. I don't know why, even though I 'ant clean' and 'svn up'
My machine is hot at picking seeds today :).
[junit]
[junit] junit.framework.AssertionFailedError:
[junit] at
org.apache.lucene.util.LuceneTestCase$LuceneTestCaseRunner.runChild(LuceneTestCase.java:1076)
[junit] at
FYI this is a relatively new test, and I think its a bug in IBM's jre.
I found similar problems in ICU and i know IBM JRE's intl impl is
different than sun's (maybe a newer icu version?)
I worked around these in the icu collation tests, and we might have to
do the same here... we can't use
Shai,
This has also been failing on trunk, but I can't repro it, so it's
awesome you can!
Can you turn on IW's infoStream and get the failure to happen and post
the results?
I agree this is likely a CMS thing...
Mike
http://blog.mikemccandless.com
On Mon, Mar 28, 2011 at 8:15 AM, Shai Erera
Update: I switched to the newest IBM JRE I have available, but the test
still fails with that seed.
Shai
On Mon, Mar 28, 2011 at 5:20 PM, Robert Muir rcm...@gmail.com wrote:
FYI this is a relatively new test, and I think its a bug in IBM's jre.
I found similar problems in ICU and i know IBM
I ran tests on 3x today and hit this:
[junit] Testsuite: org.apache.lucene.index.TestIndexWriterMergePolicy
[junit] Testcase:
testMaxBufferedDocsChange(org.apache.lucene.index.TestIndexWriterMergePolicy):
FAILED
[junit]
[junit] junit.framework.AssertionFailedError:
[junit]