Skip to content

Scheduled full-matrix failures go to nobody #69

Description

@dprada

WhatCI Full Matrix runs on a schedule and fails into an empty room. It has been red for four consecutive weeks and no one noticed.

2026-08-17   failure
2026-08-10   failure
2026-08-03   failure
2026-07-27   failure

The cause of those four runs is fixed in 9da4ca7 (the adapter-contract tests did not load their backends, so they failed under pytest -n but passed sequentially). This issue is not about that defect. It is about the four weeks.

How — a scheduled workflow that fails produces no issue, no assignee, and no notification anyone acts on. Anyone running the suite locally in the usual sequential way saw green, so there was nothing to contradict.

Why — the same pattern was found on the same day in uibcdf/molsysmt, where CI weekly had failed the day before a regression was discovered by other means. Two repositories of the suite, same blind spot: the only job that runs the full matrix reports to nobody.

Proposed — have the scheduled workflow open or update an issue on failure, so a red matrix has an owner and a date. Alternatively, or in addition, run the suite with -n on push so the parallel configuration is exercised by someone who is watching.

Relateduibcdf/molsysmt#171 covers the same gap from the MolSysMT side.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions