2024 meetings

[oliver-sanders]

Issue meeting notes, follows on from #163

[oliver-sanders]

2024-02-19T20:00Z

Agenda:

• 8.3.0 release 😬 https://github.com/cylc/cylc-flow/milestone/117
• migration support
• Add new 'RAW' variables for SHARE and WORK variables to ensure share/work dirs are created, not left with broken symlinks cylc-flow#5978
• Open questions

---

quick summary notes

Cylc-8.3.0

• aim for March, this is pretty urgent
• triage: drop all non-essential issues back to 8.4 or 8.3.x
• cylc set is essential, but we can push cylc remove and skip-mode back
• the remaining optional outputs work is less critical, since we took expire out of the mix

Cylc-8.2.5

• UK team still think this is worth releasing, before 8.3.0 gets out
• (NIWA side: I don't mind either way, no pressing 8.2.4 bugs are affecting us that I'm aware of)
• OS to review 8.2.x and 8.2.5 milestones tomorrow

Migration:

• high support workload at the moment, making it hard to get development time

Open questions (link above)

• we all need to put a bit of time into considering those, they're piling up

Multi-user gscan

• this is viewed as a requirement at BOM, due to a significant number of production role accounts
• in principle the UI can manage multiple subscriptions to multiple (user-specific) UI Servers, and users would appear at the top level of the gscan hierarchy, but to date we've had to concentrate on efficiency and stability of the single-user subscription
• to do multi-user properly there's quite a lot to consider, e.g. what determines which other users you can see, load on the UI for many users, etc.
  -- it may be an option for BOM to do this via a local UI extension (even a separate "gscan view" perhaps); but note that development and code review resources are currently very limited, so it might have to be maintained separately for a good while with no guarantee to be merged as-is

AWS Batch

• some MO work going is going on now with Cylc on AWS
• NIWA has a dev 1-day per week for a while, considering cloud platform support in Cylc (just starting)

[oliver-sanders]

Next Meeting

Agenda:

• 8.3.0 release 😬
  • https://github.com/cylc/cylc-flow/milestone/117
  • https://github.com/cylc/cylc-uiserver/milestones
  • https://github.com/cylc/cylc-ui/milestone/25
• partially satisfied tasks in branched graphs
• suicide triggers in the UI / datastore
• Consider creating share/cycle/<cycle> dirs in Cylc? Would need a matching env var.
• DS: NIWA operations conversion gotchas?
  • show prerequisites in the GUI
  • unset prerequisites? (n.b. comment in proposal)
  • Easily get a task to "go back to waiting on prerequisites, and run when satisfied"
  • jobs that are actually in one state, but reported to be in another by the job runner (really quite common) .. 1 move/action as opposed to several and an understanding of Cylc8 concepts
• JR bug? https://cylc.discourse.group/t/possible-race-condition-in-global-cylc-reload/949/2
• Open questions

[hjoliver]

5 June Meeting

Agenda:

Primarily: 8.3.0 release. The previous agenda still applies!

• excuses: HO workflow-state PR (and N*w*-state fubar)
• (probably get OS command validation PR in too?)

Other: ...

[oliver-sanders]

Next Meeting

Agenda:

• (quick) skip mode: consider using special "skip" task messages
  • If an output was completed in skip mode, save the message as satisfied by skip mode rather than, e.g. succeeded.
  • Question: Any objections?
• (quick) data-store: incorrect information for n>0 xtriggers
  • Xtriggers are not evaluated until at least one pre-requisite has been satisfied.
  • But xtriggers will show up e.g. in the GUI whether they have been evaluated or not.
  • This can make it look like an xtrigger (e.g. wall_clock) is unsatisfied in the GUI when it is actually unevaluated which can confuse users into thinking xtriggers aren't working.
  • Question: What should we do? (see issue for options)
• Support EmPy v4?
  • Cylc's EmPy support is currently broken, it's probably an easy fix (PR up), but do we want to continue support?
  • EmPy:
    • Has a single maintainer.
    • Is a closed-repository project (i.e. effectively closed source, you can acquire their releases, but you can't access the source VCS, raise a PR or anything like that).
    • Has an infrequent release cycle and a history of year-long gaps between releases.
    • Does not maintain their conda-forge feedstock.
  • Question: Given we currently do not have any EmPy users, do we want to continue support for this system?
• Housekeep broadcasts on task completion
  • How do we want broadcast housekeeping to work at Cylc8?
  • We used to clear broadcasts when the cycle they applied to "closed".
  • This means you will get a different result if you re-trigger a historical task which doesn't really make sense.
  • We could consider a different housekeep strategy (e.g. see discussion on issue) or opt not to housekeep broadcasts at all.
  • Ensure we consider the implications of whatever we decide for reflow use cases.
  • Question: [How] should we housekeep broadcasts?
• CYLC_SHARE_CYCLE_DIR
  • Some concern on the FS impact of per-cycle directories.
  • Is this legitimate? We are already doing this for the log directories.
    • (HO: not very legit; we can provide them, but they don't have to be used, and can be hand-rolled if not provided)
  • Should we develop the housekeeping tools for this before implementing?
  • Question: Should we press ahead with this, or stick it on ice?
• How to prevent a future sub-graph from running
  • This is a new use case made possible by SoD allowing downstream tasks to be removed implicitly.
  • At Cylc 7, only explicit task removal was possible (although users usually reset tasks to succeeded to skip them).
    • (HO: the above two points are only partly true. In Cylc 7 you have to explicity remove the first pre-spawned task of an unused branch but anything downstream of that would in fact be "implicitly removed")
  • What are the use cases for implicit task removal (as opposed to removing/skipping by task/family/cycle)?
    • (HO: primarily the classic non-clock task expire use case: external circumstances say I don't need to run task-x and the sub-graph downstream of it)
  • How should downstream impacts be handled (to avoid stall)?
• BoM corrupted DB issue
  • How are these databases getting corrupted? Need information from BoM.