[Ariznawlll]

Scope

Split x86 UT Coverage into three complementary groups sharing the cs-large-7c14g runner pool. Each group selects its S3 configuration from the actual runner environment (MTY, MNL2, JED or RUH), uses a run/attempt/group-scoped prefix, and fails closed for unknown environments or missing configuration. Busy runners queue normally; distribution across environments is not guaranteed.

• Keep normal PR-head checkout and the existing organization-member / safe-to-test trust gate.
• Use GOMAXPROCS=7 and package parallelism 7 with the existing package/suite deadlines. No Actions module cache or builder-image cache is used.
• Check restricted-account S3 initialization, basic and streaming writes; retain the agreed exclusion of real COS/OSS provider integration tests.
• Check complete, disjoint same-source package evidence before merging all three profiles. Publish the combined profile using the existing ut-coverage-generation-...-attempt-... contract for the downstream BVT/coverage merge. Select matching latest group artifacts by run and attempt, including partial reruns.
• Preserve per-group and combined skip evidence. Comparison against a baseline is an explicit validation step; automatic discovery of a current baseline is not implemented.

Evidence and remaining rollout checks

• No-cache shared-pool experiment: success in 34m57s, 504 packages across MNL2/MNL2/MTY, S3 checks and combined profile passed, no failures/timeouts or new skip identities compared with the same-source baseline.
• Earlier shared-pool experiment: success in 35m26s across MNL2/RUH/MTY; its module-cache restore missed. These are isolated experiments, not executions of the final production workflow.
• JED shared-label probe passed. The historic RUH S3 timeout was not reproduced in a read-only probe, but its root cause and sustained reliability remain unverified.
• Local: 26 routing, grouping, status/evidence selection and package-runner tests passed; Go profile-compactor tests passed; real three-group evidence was selected and validated locally (504 packages). A pre-existing process cleanup test returned Operation not permitted on local macOS; the Linux run below passed that test.
• Linux workflow and contract checks passed for commit 58acbd4, including workflow validation, native runtime, S3 routing, three-group artifact selection and process cleanup tests.
• Multi-PR concurrent capacity, JED full Coverage using the final shared-pool workflow, production caller integration and stable RUH S3 still require validation. Do not interpret a single successful isolated run as proof of production readiness. The PR does not change main or runner capacity.

[Ariznawlll]

已补充 CS S3 接入合约测试和 JED 内部 endpoint 部署说明（a414b3e）。CS 路径按 runner 环境选择对应 secret；受信任 PR 配置不全时明确失败；真实 COS/OSS 不注入 CS 日常 UT；强制核对 CS S3 basic 和 streaming_write 实际 pass。修复 secret-free local MinIO canary 仍错误要求云凭证的检查。本地 13 项 Python 测试和 actionlint 通过；最新 Linux 合约及隔离读写清理验证运行于 https://github.com/matrixorigin/matrixone/actions/runs/43149448772 。依赖 matrixorigin/matrixone#32590；真实 S3 配置和验证未完成，保持 draft，不切换 main 或现有正式测试。