2024-08-28 Sysops input on OST/ Approval of Sunflower OST

Translator

Date

Attendees 

Discussion items

TimeItemWhoNotes
1 minScribeAll

Ingolf Kuss is next, followed by Maccabee Levine

Maccabee Levine switched with Ingolf Kuss since he is heavily involved in the conversation.

Reminder:  Please copy/paste the Zoom chat into the notes.  If you miss it, this is saved along with the meeting recording, but having it here has benefits.

*sysops input to OSTAll

Some OST decisions effect Sysops directly, how can we gather their feedback? How should decisions about system operations be included in the OST?

  • Ingolf Kuss Just discussed MinIO, Postgres, ElasticSearch on Monday.  
  • Jenn Colt Are we missing a step in the OST page process, where we should get Sysops input?
    • Ingolf Kuss Sysops should certainly review the page.  Not all is relevant.  Should Sysops make a proposal, or review a proposal from TC?
    • Jason Root Discussed at Sysops.  Need a step to make Sysops aware, for each flower release.  No review needed.  Inform Sysops via whatever mechanism.
    • Jenn Colt We have a list of channels to post to (on OST wiki page), maybe missed when Draft.  Need to send at each status.
    • Ingolf Kuss Really affects the big hosting providers.  Taras, Index Data team. 
    • Need a Sysops session with a broader audience, incl EBSCO and ID, after those notifications, so they can have a meeting focused on it.
    • Maccabee Levine Not sure if we have ever notified some of those Slack channels after an OST page change.

Finalize Sunflower OSTAll

With the feedback from sysops can we move the Sunflower OST page to approved?

  • Jenn Colt JDK decision had been to stay at 17.  Any reason to change?
    • Julian LadischTaras Spashchenko said not much work to bump to 21.  Ingolf Kuss agrees.
    • Julian Ladisch Will also need 21 for Keycloak for Eureka.
    • Craig McNally We're on Keycloak 25.  Was unaware that Keycloak had deprecated JDK 17 support.
    • Jenn Colt will leave this question open.
    • Marc Johnson The work for JDK 21 is not development work; it's tooling.  We don't know if there is community capacity for that work.  And don't know if the devops contract with Index Data includes that work; if not, would have to find folks who will.  Also not sure why we're talking about Eureka in this context, since it's not adopted by the community yet on any timeline.
    • Jenn Colt Is talking to Index Data then a missing step in the OSTs?
    • Marc Johnson Depends on what change we are making.  If one of these decisions, i.e. Java or yarn-to-npm, has a burden on how we build the system, then have to include those stakeholders.  Similarly if it had a developer impact, should talk to the developers.  There should be a step between "this would be a good idea for this release" to "can we actually achieve it".
    • Craig McNally Keycloak 25 supports but deprecates JDK 17.  Once Keycloak 26 comes out, JDK 17 support is removed in favor of JDK 21.  Keycloak only supports the current version, so would need to upgrade to Keycloak 26 and JDK 21 at least for Keycloak.  If running FOLIO in Docker containers there is not much tooling, just use a different base image.  No code to be compiled.
    • Marc Johnson Agree with Craig.  Building from source would be different.  But Keycloak is not currently in the scope of OST.  If we want Eureka supported, then Keycloak and other tools need to be in the OST.  At that point it's worth talking about Java version conflicts.  Talking several steps ahead.
    • Julian Ladisch By keeping 17, we indirectly make a decision to not allow Keycloak / Eureka on Sunflower.  There is a group of German sysops who want to do a test installation of Eureka.  But that's not the main point.  Developers wanted to move to JDK 21.  Should have a good argument on why not to switch to 21.  Jenkins support should be possible.  Docker container completed yesterday for JDK 21.  No reason to reject 21 for Sunflower.
    • Marc Johnson If folks in the community are willing to commit to those build infrastructure changes on the schedule for Sunflower, then nothing stops 21.  That's the constraint.
    • Jenn Colt Community hasn't decided on Eureka, but to support experimentation with it is a positive.
    • Jenn Colt and Craig McNally will follow up with Index Data.
    • Marc Johnson Given the confusion happening here, we need something that guides folks making changes, when and how to talk to the TC about it.
    • Tod Olson The time to raise this is "as early as possible".  Pushing OST earlier as we did is good.  Gap may be not reaching out to the stakeholders.  Someone should not have to bring an upgrade issue to us, for us to reach out.  More meeting in the middle.
    • Marc Johnson We talked today about putting notifications out at various points in the process.  Messaging needs to be different at different stages.  Put out a draft for a release, announce it, invite contact to TC within X period of time (6w-2m?) if you are interested in discussion.  Gives them enough time to think, gives us enough time to get process done by when we need.  I.e. talking about X decision on a particular date.  If they don't turn up, they don't.  At later point, we announce it – barring 1st party library changes, this is the tech choices for this release.
      • Maccabee Levine That mirrors the RFC process.   And for Q, there was about 5m between DRAFT and ACCEPTED, that is certainly enough time.
  • Infrastructure
    • Ingolf Kuss Sysops wants OST page to add Helm, say version 18.2.6.  And ES should still be 8.
    • Jason Root Decision was to test on S3 API version, rather than specific MinIO version.  Whichever MinIO version supports that API.  And then whichever Helm version that requires.
    • Marc Johnson It's useful for modules to declare what they test against.  But the intention of the policy is to go in the opposite direction – module developers should check this page, comply with this version.  One goal for this page is that on every upgrade, sysops don't have to go through every module and see what infra they support.  Policy should drive what developers support; sysops have one place to check.  Not there yet.  Agree that for today we need a baseline version in there. 
    • Marc Johnson For ES, we don't need Helm in there, it's not a driver of these policies, it's a manifestation.
      • Craig McNally agrees.  Similar to not specifying Kubernetes version.  Just another thing TC would have to maintain going forward.
      • Ingolf Kuss agrees.
    • Marc Johnson We should consider adding Docker to OST as official FOLIO tooling.  In the future, not today.
  • Jenn Colt How do we finish this conversation next meeting?  Did not finish JDK.  Agreed on MinIO but have to get documentation right.
NAZoom Chat