Draft: Add test of execution of workflow with self-triggering task

[poautran]

In GitLab by @loichuder on Aug 30, 2024, 12:12 GMT+2:

For #16

Needs https://gitlab.esrf.fr/workflow/ewoks/ewokscore/-/merge_requests/236

Migrated from GitLab: https://gitlab.esrf.fr/workflow/ewoks/ewoksppf/-/merge_requests/101

[poautran]

In GitLab by @loichuder on Sep 10, 2024, 17:25 GMT+2:

added 2 commits

• 43b48ad5 - Add test of execution of workflow with self-triggering task
• cef244e4 - Adapt tests to new ewokscore version

Compare with previous version (link unavailable after force-push)

[poautran]

In GitLab by @loichuder on Sep 10, 2024, 17:40 GMT+2:

marked this merge request as ready

[poautran]

In GitLab by @loichuder on Sep 12, 2024, 14:05 GMT+2:

added 3 commits

• cef244e4...05a43232 - 2 commits from branch main
• f31f5b4 - Add test of execution of workflow with self-triggering task

Compare with previous version (link unavailable after force-push)

[poautran]

In GitLab by @loichuder on Sep 12, 2024, 14:20 GMT+2:

marked this merge request as draft

[poautran]

In GitLab by @loichuder on Sep 12, 2024, 14:21 GMT+2:

Drafting this until #16 is fixed.

[poautran]

In GitLab by @loichuder on Feb 2, 2026, 15:00 GMT+1:

added 63 commits

• f31f5b4...9c05e7c - 62 commits from branch main
• fc6c7a4 - Add test of execution of workflow with self-triggering task

Compare with previous version (link unavailable after force-push)

[woutdenolf]

The new workflow25 in the unit tests looks like this at the moment

The decider node has three downstream links: good, bad and self.

The link to self causes the input cache of decider to have disable=True cached. When the decider receives this, it will stop triggering downstream nodes.

However it could be that we have more than one downstream trigger to good or bad. Suppose good + self is triggered but in the mean time another metric or timeout arrives and gets scheduled before disable=True is cached.

[woutdenolf]

In fact the solution to the race condition is to not use the link-to-self but rely on the cached inputs to ensure we trigger downstream only once

Since all inputs of decider are cached the task always remember all previous calls and can ensure we trigger downstream only once (good or bad).