Big benefit of (B) is same release cycle as solr, perhaps same version, 
piggy-back on releaseWizard (perhaps not even any new steps), auto test with 
same version.. Guess this is kind of a mono-repo kind of choice. It's different 
for solr-operator, different tech, built to support many solr versions etc.

But short term my vote is still (A)

Jan

> 13. jan. 2022 kl. 16:47 skrev Houston Putman <[email protected]>:
> 
> Thanks for listing these out David.
> 
> I think for the sake of 9.0, I choose (A). For 10.0 and beyond, I think I 
> would prefer (B). You are right that if it is in it's own repo (C) it will 
> not get much looks/releases. Which could be fine, but I think it's nice that 
> it has to release with every Solr version for now (and therefore must 
> maintain compatibility).
> 
> Either way for (B) or (C) it needs its own set of artifacts & docker file, 
> but I agree with Jan, I don't think those would be particularly hard to solve.
> 
> - Houston
> 
> On Thu, Jan 13, 2022 at 6:57 AM Jan Høydahl <[email protected] 
> <mailto:[email protected]>> wrote:
> It's kind of a free-rider for sure, and a convenient one :)
> Ideally I'd like (C), but pragmatically for 9.0 I vote for (A) unless someone 
> steps up with lots of energy :)
> 
> If we did C, there'd be a separate release artifact in CDN alongside 
> solr-operator. <https://dlcdn.apache.org/solr/> The Dockerfile would be slim 
> and simple, and the image could be pushed to apache/solr-prometheus-exporter, 
> no need for "official" image?
> 
> Jan
> 
>> 13. jan. 2022 kl. 02:56 skrev David Smiley <[email protected] 
>> <mailto:[email protected]>>:
>> 
>> The inevitable rename of the "contribs" module to something else ( 
>> https://issues.apache.org/jira/browse/SOLR-15917 
>> <https://issues.apache.org/jira/browse/SOLR-15917> ) will be a time for us 
>> to move the prometheus exporter somewhere else, as it is not a 
>> module/package within Solr; it's an independent service with its own 
>> dependencies; one of which is SolrJ.  It does not depend on Solr-core or 
>> such internals as it once did.
>> 
>> I just want to list some options here; perhaps there is another or a 
>> variation.  I suppose the last option is maybe the best but ultimately 
>> someone would need to step up to the task.  If nobody steps up (not me!), 
>> then (A) will get done; it's very easy.
>> 
>> (A) The least effort is merely to move it somewhere else in our directory 
>> structure.  If it's still under /solr, like /solr/prometheus-exporter, then 
>> definitely very minor effort & impact.
>> 
>> (B) More effort is to move to the top level of our source repo to distance 
>> itself further from the Solr server.  But that probably means it would not 
>> be in our distribution, which also means it would not be in Solr's Docker 
>> image?  We could write a Dockerfile easily enough, I'm more unsure of how to 
>> publish it and how much effort that is.
>> 
>> (C) Even more effort is to outright move it to a new ASF git repo; to 
>> arrange for CI (or just rely on GitHub Automation?).  I'm unfamiliar with 
>> the effort in all that.  I could help with extracting source history in 
>> initializing the git repo so we don't lose that.  There is also the need to 
>> add a Dockerfile and to publish it.  The beauty of this is that it can have 
>> its own release cycle!  That means very few releases in practice.  Although 
>> it means extra work for actually doing these releases (voting, 
>> release-manager steps, publishing), instead of a free ride on the Solr 
>> release train.
>> 
>> ~ David Smiley
>> Apache Lucene/Solr Search Developer
>> http://www.linkedin.com/in/davidwsmiley 
>> <http://www.linkedin.com/in/davidwsmiley>

Reply via email to