CI: the unit job's worker is killed for memory in the four-times-size Markdown cases #830

New issue

Closed

opened 2026-10-10 14:13:33 +00:00 by tiagoagueda · 0 comments

tiagoagueda commented

2026-10-10 14:13:33 +00:00

Owner

The unit job runs two pytest workers in a job container capped at 2 GB (forgejo-runner, --memory 2g). Each worker grows towards 1 GB over the suite, and the 160,000-character hostile Markdown cases of tests/security/test_markdown.py tipped one over: the kernel OOM-killed it (ouranos dmesg, 2026-10-10 13:58), xdist reported a crashed worker, and coverage fell under 92 % on runs #804, #810 and #816.

Decided 2026-10-10: mark those cases heavy, leave them out of the default run, and run pytest -m heavy -p no:xdist as a step of its own after the suite.

The unit job runs two pytest workers in a job container capped at 2 GB (`forgejo-runner`, `--memory 2g`). Each worker grows towards 1 GB over the suite, and the 160,000-character hostile Markdown cases of tests/security/test_markdown.py tipped one over: the kernel OOM-killed it (ouranos dmesg, 2026-10-10 13:58), xdist reported a crashed worker, and coverage fell under 92 % on runs #804, #810 and #816. Decided 2026-10-10: mark those cases `heavy`, leave them out of the default run, and run `pytest -m heavy -p no:xdist` as a step of its own after the suite.

tiagoagueda referenced this issue from a commit

2026-10-10 14:14:02 +00:00

Run the four-times-size Markdown cases in a process of their own

tiagoagueda closed this issue

2026-10-10 14:14:02 +00:00

tiagoagueda referenced this issue from a commit

2026-10-10 14:27:01 +00:00

forgejo: give the runner's job containers 3g

tiagoagueda referenced this issue

2026-10-29 10:03:39 +00:00

Cache what every CI job rebuilt, and fetch no map tiles in tests (#1033) #1083

tiagoagueda referenced this issue from a commit