In the Jenkins boxes, we run something like the following:

[Solr-Test-main] $
/home/jenkins/jenkins-agent/workspace/Solr/Solr-Test-main/gradlew
-Dtests.haltonfailure=false -Dtests.badapples=false --continue test


One idea is to disable (@Ignore) all the cuVS tests immediately, and once
the official cuVS-java artifacts are released via maven central, we run the
above job with a "--refresh-dependencies" param to unpoison the checksum
caches.

Hoss, WDYT ^ ?

On Sat, 11 Oct 2025 at 00:58, Ishan Chattopadhyaya <
[email protected]> wrote:

> @Vivek Narang <[email protected]>  FYI, in case you're not
> subscribed to the list. Any ideas, please?
>
> On Sat, 11 Oct, 2025, 12:57 am Ishan Chattopadhyaya, <
> [email protected]> wrote:
>
>> I'm now wondering how to ever fix this situation..
>>
>> Maybe, only when 25.10.1 (may or may not even happen) or 25.12.0
>> (December release) is released can this be fixed? I'm out of ideas. I guess
>> these machines are now poisoned with those stale checksums.
>>
>> I now feel, in hindsight, that I should've fixed that strict version
>> checking thing in cuVS and released differently versioned artifacts for
>> snapshots. But, it is too late now.
>>
>> On Sat, 11 Oct, 2025, 12:44 am Chris Hostetter, <[email protected]>
>> wrote:
>>
>>>
>>> https://issues.apache.org/jira/browse/SOLR-17917
>>>
>>> FYI: Most of the jenkins "main" builds are now failing because they have
>>> the "old" cuvs 25.10.0 jars in their gradle caches and main has been
>>> upgraded to Lucene 10.3 (which only works with the "new" cuvs 25.10.0
>>> jars)
>>>
>>>  So they either run tests and fail with something like...
>>>
>>> java.util.ServiceConfigurationError: org.apache.lucene.codecs.Codec:
>>> com.nvidia.cuvs.lucene.Lucene101AcceleratedHNSWCodec Unable to get
>>> public no-arg constructor
>>>
>>> ..or in the case of the "lint" jobs, that skip tests, they fail because
>>> the checksum in their cache doesn't match the checksum committed by
>>> https://github.com/apache/solr/pull/3731
>>>
>>>
>>>
>>> -Hoss
>>> http://www.lucidworks.com/
>>>
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail: [email protected]
>>> For additional commands, e-mail: [email protected]
>>>
>>>

Reply via email to