Group trigger

[hjoliver]

[UPDATED and refactored post informal agreement on the proposal]

---

Major change to manual trigger behaviour: instead of running all target tasks immediately despite unsatisfied prerequisites, trigger them as a group that respects in-group dependencies and ignores off-group prerequisites that would cause a stall.

This is an easier way to (re)run a sub-graph when all the tasks can easily be identified - no need to manually identify sub-graph start tasks to trigger and/or set off-flow prerequisites.

(Amazingly, this should not be a breaking change for any sensible cylc trigger use cases).

High level implementation (see the the proposal for details):

1. Examine the group to identify all off-group prerequisites
2. Remove the (original or specified) flow from group tasks, to allow rerun in the same flow
3. Satisfy the off-group prerequisites, to trigger group start tasks and prevent later stalls
4. (The resulting flow then takes care of itself to propagate through the group)

---

Example from the proposal, tweaked to run automatically via a stall handler:

# Task b fails due to an upstream problem not captured by the workflow
# so we need to rerun the graph from task a, to complete the workflow.
[scheduler]
    [[events]]
        expected task failures = 1/b
        stall handlers = \
            "cylc trigger %(workflow)s \
             //1/a //1/f_m1 //1/f_m2 //1/f_m3 //1/g_m1 //1/g_m2 //1/g_m3 //1/b"
[scheduling]
    [[graph]]
        R1 = """
            start => a
            x => f_m1
            a => f_m1 => g_m1 => b
            a => f_m2 => g_m2 => b
            a => f_m3 => g_m3 => b
            b => end
            g_m3 => y
       """
[runtime]
    [[b]]
        script = [[ "${CYLC_TASK_SUBMIT_NUMBER}" != '1' ]]
    [[FAMILY]]  # (anticipates inactive tasks globbing to specify a trigger group)
    [[a, f_m1, f_m2, f_m3, g_m1, g_m2, g_m3, b]]
        inherit = FAMILY
    [[start, x, end, y]]

---

• I have read CONTRIBUTING.md and added my name as a Code Contributor.
• Contains logically grouped changes (else tidy your branch by rebase).
• Does not contain off-topic changes (use other PRs for other changes).
• Applied any dependency changes to both setup.cfg (and conda-environment.yml if present).
• Tests are included (or explain why tests are not needed).
• Changelog entry included if this is a change that can affect users
• Cylc-Doc pull request opened if required at cylc/cylc-doc/pull/XXXX.
• If this is a bug fix, PR should be raised against the relevant ?.?.x branch.

[hjoliver]

Note on changes to reference logs: group trigger causes triggering by satisfying prerequisites, rather than by direct triggering despite unsatisfied prerequisites ... so triggering now gets logged as (e.g.) 1/b triggered off ['1/a'] instead of 1/b triggered off [ ].

[hjoliver]

Integrated xtrigger prerequisite setting. Tested with this:

[scheduling]
    [[xtriggers]]
        xoff = xrandom(0)  # never
    [[graph]]
        R1 = """
            failer? => a => b & c => cleanup
            failer:fail? => triggerer
            off => c  # off-group task prerequisite
            @xoff => b  # off-group xtrigger prerequisite
        """
[runtime]
    [[failer, off]]
        script = false
    [[a, b, c]]
    [[triggerer]]
        script = "cylc trigger $CYLC_WORKFLOW_ID //1/a //1/b //1/c"
    [[cleanup]]
        script = "cylc remove $CYLC_WORKFLOW_ID//1/*:failed"

From the log:

...
INFO - Command "force_trigger_tasks" received. ID=84ed8d79-e1f6-47c6-b55a-5cf53fb1e950
    force_trigger_tasks(flow=[], flow_wait=False, on_resume=False, tasks=['1/a', '1/b', '1/c'])
INFO - [1/b:waiting(runahead)] prerequisite force-satisfied: xoff = xrandom(0)
INFO - [1/c:waiting(runahead)] prerequisite force-satisfied: 1/off:succeeded
INFO - [1/a:waiting(runahead)] prerequisite force-satisfied: 1/failer:succeeded
INFO - [1/a:waiting(runahead)] => waiting
INFO - [1/b:waiting(runahead)] => waiting
INFO - [1/c:waiting(runahead)] => waiting
INFO - Command "force_trigger_tasks" actioned. ID=84ed8d79-e1f6-47c6-b55a-5cf53fb1e950
...

[hjoliver]

TODO:

• code tidied and TODOs addressed
• manual testing against the proposal
• test for xtrigger prequisite satisfaction in a group trigger operation
• replace tests/f/cylc-trigger/08-group-trigger with integration tests

[hjoliver]

Done.

[wxtim]

Looks like something is going consistently wrong with 3.7?

[hjoliver]

All tests passed on rerun today - maybe you just need to reload your browser to show up to date results?