cylc-flow icon indicating copy to clipboard operation
cylc-flow copied to clipboard

Start paused or held, then manually trigger individual tasks

Open hjoliver opened this issue 1 year ago • 2 comments

See https://cylc.discourse.group/t/is-it-not-possible-to-trigger-a-task-while-paused/853

At the moment I think we have to use --hold-after=P with P earlier than the initial point.

Can we:

  • allow triggering of individual tasks when the workflow is paused and/or
  • new cylc play --hold-all option, to start with all tasks held (at least all initial tasks, and possibly all future tasks kind of like the way hold-after-point works)

hjoliver avatar Jan 10 '24 01:01 hjoliver

IMO this is sufficient:

cylc play --pause
cylc hold '*'
cylc play

oliver-sanders avatar Jan 11 '24 10:01 oliver-sanders

Yeah I commented on that on the thread (before I saw your response here).

I think that's fine so long as users understand:

  • why it works
  • why that only holds the initial tasks, not future ones
  • and (relatedly) why cylc hold '*' seemingly does nothing at all most of the time if the workflow isn't started with --pause

All of which isn't exactly intuitively obvious to most users.

It's our usual problem, that the old Cylc 7 task pool, as inconsistent and un-justified as it was, happened to have enough pre-spawned tasks in it that cylc hold seemed to do what users would naively expect most of the time.

For the given use case (hold everything, and trigger a bunch of individual tasks without causing downstream action - during early workflow development) I don't think it would hurt to have a "hold all tasks" start up option.

hjoliver avatar Jan 12 '24 04:01 hjoliver

At present the best way to do this is using the "hold after cycle point" as this ensures that any tasks spawned by the triggered task are also held.

Documenting that intervention here: https://github.com/cylc/cylc-doc/pull/739

With #5827 this issue will go away, so it's a simple hold all, trigger, release all.

If we're ok with this, suggest superseding with #5827

oliver-sanders avatar Jun 14 '24 09:06 oliver-sanders

Hold-after-cycle-point is really not intuitive for this use case.

I think I'm OK with superseding this by #5827 so long was we can get cylc hold '*' done pretty soon (and documenting the current intervention is good). It's also partly superseded by #5963 pending resolution of that discussion.

hjoliver avatar Jun 17 '24 22:06 hjoliver