2025 meeting notes

[MetRonnie]

Following on from #188

[oliver-sanders]

27 January 2025

highlights

• 8.4.0 released 🎉
• NIWA ops live with 8.4.0 🎉

Dec meeting follow-up

• ? JR to follow-up on BOM Issues (see list in prev meeting notes 2024 meetings #188 (comment), including DM comments) and try to assess if they're still needed
• ? DS to investigate two claimed cases of tasks disappearing from the pool on reload
• ? (HO) Momentum Partnership Board feedback on Cylc Plan: strategic plan

discussion

• Review Roadmap.
  • OS: Current estimate > 5 months worth of work on the roadmap ATM. We could admit a couple more items.
  • HO: exposing scheduler warnings etc. in the UI
    • Not presently on the roadmap. Shall we add it?
    • Open issue: Implement log message stream.
    • Implement in cylc-flow (easy, add a new log handler) or cylc-uiserver (requires log file polling/streaming but would work offline).
  • HO: Multiple select in UI.
    • Already on the roadmap, @hjoliver pressure increasing?
  • OS: Automatic restart and upgrade
    • Makes it easier to keep up to date with Cylc version upgrades.
    • Visual nag for users sticking with older versions.
    • A priority?
    • Easier upgrade for users cylc-flow#5209
  • HO: Cylc 9 items!
• Manylinux 1 compatibility
  • This is still required but we can't really test for it without investing a bit of time setting up a docker container.
  • Suggest deleting this action in March and falling back to manual testing until we can retire our manylinux 1 requirement.

• new: hard-wired scheduler host self-identification broken cylc-flow#6575
• 8.5.0
  • High priorities:
    • group trigger
      • needs out-of-pool task matching too? (e.g. to do Cylc 8 "warm start")
    • broadcast expiration fix
    • release by March!
  • Other:
    • GraphQL stack overhaul if it is ready in time.
    • [GraphQL array deltas](GraphQL: updating arrays with delta updates cylc-flow#6307
    • global config reload?
• outstanding questions
  • exit-script does not have access to task env (Can we close this after documenting it for users?)
  • can we ditch dummy mode? (low priority, already run by Discourse,
  • ui: don't show DEBUG messages?
    • Quick!
    • The tree view shows a summary of recent job messages.
    • Should this exclude DEBUG messages.
    • These would still be visible elsewhere in the GUI (currently the tree view job-info dropdown and info view).

Minutes

Present: HO, DM, OS, TP, RD, DS, MD

NIWA ops: 57 Cylc 8 workflows, no problems so far 🎉

DS has not seen the reported "tasks disappear from the pool on reload" problem again yet.

Several of BOM issues now closed. Global reload agreed by awaiting review. Need to follow up on the metomi/rose one (HO action)

Strategic plan: HO to do as minimally as possible and run by the team. Important to note that the first-order goal is to provide a stable tool to drive our operational systems, and as such we are primarily driven by operational requirements, and those may come up unexpectedly (e.g. bugs and intervention requirements)

Another final Cylc 7 release is needed ("cylc review" PR merged, one small job runner PR needs review) (HO action)

One back-compat issue is in the wild: 8.4 UIs can't trigger tasks via 8.3 UI Server because of the schema change for the new "trigger on resume" option - TBD

Group trigger task matching: note also cylc/cylc-flow#5416

UK:

• team availability limited due to new HPC acceptance, new Spice (upcoming), support requirements
• Cylc 8 PS is expected by July, live by Nov 2025
• Cylc Hub now successfully deployed at MO. Currently using a local spawner on a single beefy VM for UI Servers
• MD moving on to another team soon 😢
• Currently reviewing Python 3 Rose Edit

Cylc UI family collapse in final review

Start a new roadmap for longer-term (Cylc 9) items, so avoid cluttering the current list of top priorities (HO action)

UIS memory blow-out: entirely (or mainly) due to not clearing memory when a workflow stops (OS)

Exposing scheduler messages in the UI:

• as an interim measure we should take the custom log handler approach: have it generate ZMQ messages
• this is live only, so not ideal, but still a nice enhancement
• ultimately we really need high visibility of incomplete tasks and/or associated warnings

ManyLinux1 compat:

• needed for arcane XC40 support till July (old glibc based distro)
• any library dependency changes must be review carefully till then, and possibly tested manually at MO

exit-script PR: HO to comment and close, with an associated doc PR to explain and suggest workarounds