Summary

• Enable Vitest file isolation and cap the in-job test pool so each existing CI runner can use file-level parallelism without adding shared/shard jobs: Linux uses 8 workers, macOS/Windows use 3 workers to avoid resource contention in process-heavy suites.
• Keep the existing OS/Node test matrix and restore the original coverage behavior for the main test job (ut run ci on every matrix entry, Codecov upload on non-Windows) so Codecov project coverage remains comparable to next.
• Keep project-level serialization only where tests are known to mutate shared process/global/external-service state: cluster, development, logrotator, mock, onerror, redis, tegg/plugin/tegg, and tegg/plugin/orm.
• Fix parallel-test fallout found during verification: mock cluster internal port allocation, long Unix-socket paths in the current workspace, and schedule/log timing assumptions.

Local verification

• Docker Redis 7 and MySQL 8 are running locally; test databases were created before full-suite verification.
• CI=true pnpm vitest run --bail 1 --retry 2 --testTimeout 20000 --hookTimeout 20000 --coverage --maxWorkers=4 --isolate --reporter=json passed in a clean short-path copy: 1897/1897 suites, 3218 passed, 257 skipped, elapsed 368.36s.
• CI=true pnpm vitest run --bail 1 --retry 2 --testTimeout 20000 --hookTimeout 20000 --coverage --maxWorkers=8 --isolate --reporter=json passed in a clean short-path copy: 1897/1897 suites, 3218 passed, 257 skipped, elapsed 234.28s.
• CI=true pnpm vitest run plugins/schedule --isolate --maxWorkers=8 --reporter=dot --silent passed: 23 files passed, 1 skipped, 28 passed, 13 skipped, elapsed 34.72s.
• CI=true pnpm vitest run plugins/mock/test/cluster.test.ts packages/cluster/test/app_worker.test.ts --isolate --maxWorkers=8 --reporter=dot --silent passed: 2 files passed, 21 passed, 16 skipped, elapsed 80.76s.
• CI=true pnpm vitest run packages/cluster/test/master/start-master.test.ts --isolate --reporter=dot --silent passed: 1 file passed, 2 passed, 4 skipped, elapsed 1.28s.
• CI=true pnpm vitest run tegg/plugin/tegg/test --isolate --maxWorkers=8 --reporter=dot passed: 18 files passed, 2 skipped, 51 passed, 2 skipped, elapsed 37.92s.
• VITEST_MAX_WORKERS=4 CI=true pnpm vitest run ...failed macOS suites... --isolate --reporter=dot --silent passed outside the sandbox: 8 files passed, 83 passed, 22 skipped, elapsed 22.15s.
• VITEST_MAX_WORKERS=3 CI=true pnpm vitest run plugins/onerror/test/onerror.test.ts plugins/redis/test/redis.test.ts tegg/plugin/orm/test --isolate --reporter=dot --silent passed: 4 files passed, 60 passed, 5 skipped, elapsed 13.34s.
• CI=true pnpm vitest run --bail 1 --retry 2 --testTimeout 20000 --hookTimeout 20000 --coverage --isolate --maxWorkers=8 --reporter=dot --silent passed in this workspace: 508 files passed, 19 skipped, 3218 passed, 257 skipped, elapsed 547.99s.
• pnpm oxfmt --check ... passed on changed files.
• pnpm oxlint --type-aware --type-check --quiet passed for the full repo: 0 errors, 276 existing warnings.

Notes

The latest full-suite run was in the actual Multica workspace. It is slower than the earlier short-path JSON run because it includes coverage and very noisy process logs, but it confirms the complete suite passes with --isolate --maxWorkers=8.