Files
Diego Rodrigues de Sa e Souza 84cf629bec ci(release-green): run the sweep's heavy lanes one at a time, unit as ci.yml's shards (#17156)
The scheduled full sweep ran unit, vitest, integration and the package build as one
Promise.all wave. The 2026-10-18 (omni-build) and 10-19 (omni-release) sweeps killed
unit, integration and the build at their ceilings inside it. On 10-19 the webpack build
(14-16 GB) pushed the 8-core, 20 GB omni-release slot past memory.high: the kernel
throttled every lane and the app's resourcePressure guard, reading that cgroup, answered
503 inside unit tests (three false reds that pass in isolation on the tip). Both days the
single unit process at --test-concurrency=4 had run ~29k of its ~32k tests at its 80 min
ceiling, while ci.yml runs the same suite as 8 shards of 11-18 min.

Unit now runs ci.yml's own shards (npm run test:unit:ci:shard, TEST_SHARD=i/N with N read
from ci.yml's test-unit matrix), one per four cores of availableParallelism() (which honours
the slot's cpu.max: 2 at a time on omni-release), each under a 40 min ceiling; integration,
vitest and the package build follow, one at a time. Each lane is reported as it ends, and a
failed shard is named in the gate detail, so a timed-out shard no longer surfaces a
fixture's SQLITE_CANTOPEN line as the cause.

Refs #15306
2026-10-19 09:22:17 -03:00
..
2026-06-19 06:49:01 -03:00
2026-06-23 17:06:18 -03:00
2026-07-29 15:18:55 -03:00
2026-07-13 09:12:40 -03:00
2026-07-29 15:18:55 -03:00
2026-07-13 09:12:40 -03:00