My guess is that there is an off by 1 bug somewhere that only manifests in
very specific segment layouts. Because we are not using the
SerialMergeScheduler, every run might produce a different segment layout.

On Fri, Jun 19, 2026 at 10:30 AM Dawid Weiss <[email protected]> wrote:

>
> I suspect it's a race somewhere that only manifests itself on a slower
> machine (ci workers).
>
> D.
>
> On Fri, Jun 19, 2026 at 11:00 AM Alan Woodward <[email protected]>
> wrote:
>
>> Yeah this is weird - it’s failing in almost every nightly run in the
>> elastic CI as well and is completely unreproducible.  I’ve tried tracing
>> things through and haven’t found any non-deterministic stuff in there
>> either (there’s no threading or anything that I can see?)
>>
>> On 19 Jun 2026, at 09:49, Dawid Weiss <[email protected]> wrote:
>>
>>
>> I couldn't reproduce this.
>>
>> On Thu, Jun 18, 2026 at 10:51 PM Policeman Jenkins Server via builds <
>> [email protected]> wrote:
>>
>>> Build: https://jenkins.thetaphi.de/job/Lucene-nightly-main-Linux/1728/
>>> Java: 64bit/hotspot/jdk-26 -XX:-UseCompressedOops -XX:+UseParallelGC
>>>
>>> 1 tests failed.
>>> FAILED:
>>> org.apache.lucene.sandbox.search.TestSortedDvMultiRangeQuery.testDuelWithStandardDisjunction
>>>
>>> Error Message:
>>> java.lang.AssertionError: expected:<1645 hits> but was:<1647 hits>
>>>
>>> Stack Trace:
>>> java.lang.AssertionError: expected:<1645 hits> but was:<1647 hits>
>>>         at
>>> __randomizedtesting.SeedInfo.seed([37A90FC76975A75D:B5AAD60471E69299]:0)
>>>         at [email protected]/org.junit.Assert.fail(Assert.java:89)
>>>         at [email protected]/org.junit.Assert.failNotEquals(Assert.java:835)
>>>         at [email protected]/org.junit.Assert.assertEquals(Assert.java:120)
>>>         at [email protected]/org.junit.Assert.assertEquals(Assert.java:146)
>>>         at
>>> org.apache.lucene.sandbox.search.TestSortedDvMultiRangeQuery.testDuelWithStandardDisjunction(TestSortedDvMultiRangeQuery.java:154)
>>>
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail: [email protected]
>>> For additional commands, e-mail: [email protected]
>>
>>
>>

Reply via email to