[objectstack-fleet]

Fixes #443

维护者速读

• 改了什么:企业版部署页、版本表和授权页里关于集群与高可用的说法,改成安装包今天实际能做的:只支持单节点,副本数设成大于 1 会拒绝启动,不承诺高可用。
• 为什么改:原文叫运维开启 Redis 集群驱动并设置副本数,但安装包没有接集群驱动;cloud 功能目录明确写着「不得承诺 HA」。
• 风险与回滚:只改英文文档并删除这三页的其他语言副本(能力表述被反转,按仓库规则同 PR 删除,下一轮翻译会重新生成);回滚即 revert 此 PR。
• 先看:content/docs/deploy/docker.mdx 的「Multi-node on Enterprise」整节;其次 editions/license.mdx 的表格行与「No per-node charge」一句。
• 两分钟验证:打开 /docs/deploy/docker#multi-node-on-enterprise,应看到「单节点」「大于 1 拒绝启动」,且没有「开启 Redis 集群驱动」的步骤。
• 需要你做的:无,席位落地。

What changes

• deploy/docker.mdx § Multi-node on Enterprise: replaces "turn on the Redis cluster driver, set OS_CLUSTER_REPLICAS" with: single node today, OS_CLUSTER_REPLICAS above 1 refuses the boot, do not silence with OS_ALLOW_MEMORY_CLUSTER_MULTINODE or --scale, multi-node is license-activated by contract, HA not included. The shared-secret-key and auth-secret rules stay, reworded so they no longer presuppose a driver. The Community paragraph's "Clustering and high availability are Enterprise capabilities" is reworded to match.
• deploy/index.mdx editions table and editions/license.mdx (table row, Business and Full-private bullets, scenario row): "clustering and high availability" / "Clustering / HA" become "multi-node topology (the kit runs a single node today)" or "multi-node". The line "clustering / scale is folded into Enterprise" is dropped; "no per-node charge" stays.
• Deleted the ja, zh-Hans, zh-Hant siblings of those three pages (a capability assertion is reversed; AGENTS.md rule 3).

Cloud paths each claim rests on (cloud main)

Claim	Source
Single node is the only supported shape; no cluster driver is wired into the EE kit	apps/objectos-ee/deploy/README.md § 生产建议 (line 107) and § 多节点为什么暂不支持 (lines 113-122)
OS_CLUSTER_REPLICAS above 1 refuses the boot (ClusterServicePlugin, in-process memory driver)	same README, lines 124-133
Do not use OS_ALLOW_MEMORY_CLUSTER_MULTINODE or --scale; compose ties replicas to OS_CLUSTER_REPLICAS	same README, line 135
Multi-node is activated by license contract, node cap maxNodes; a cluster driver is wired in when a contract calls for it, no date	same README, lines 117-120 and 137; docs/design/ee-feature-catalog.md line 42
No HA promise; "multi-node topology is unlocked by the Enterprise license (node cap: maxNodes)" is the allowed wording	docs/design/ee-feature-catalog.md line 42 (clustering row)

Tests

Run in the worktree at the final commit:

• pnpm install --frozen-lockfile: exit 0
• pnpm turbo run type-check --filter=@objectos/docs --force: exit 0
• node .github/scripts/check-translation-ownership.mjs --actor os-bob --files changed.txt (as translations.yml invokes it, changed list from origin/main...HEAD): exit 0
• node .github/scripts/check-translations.mjs: exit 0; gen:zh-hant --check: exit 0 (111 files match)
• pnpm turbo run build --filter=@objectos/docs --force: exit 0

Acceptance notes

Same overclaim remains outside the card's three pages and is not touched here: editions/compare.mdx line 38 ("Clustering / high availability" row), deploy/kubernetes.mdx, deploy/air-gapped.mdx (mention clustering; not read for this card). Reported for a follow-up card.

Verification

Text-only change, no UI: no browser run; the site builds.

[objectstack-fleet]

repo:cloud#1 · os-bob · session_012dn2AdARJtc4RgABPH7PHk

ACCEPT · 2026-11-07T18:01Z · head 567bf106e270f91a32c00bac0b9ed8623a28ec2b · for docs.objectos.ai#443

维护者速读

• 改了什么:部署与版本页上关于企业版「集群 / 高可用」的说法全部改成事实:企业版安装包目前只跑单节点,OS_CLUSTER_REPLICAS 设成大于 1 会按设计拒绝启动;多节点由企业版许可解锁(节点上限 maxNodes),按合同开通;不承诺高可用。涉及 docker、kubernetes、air-gapped、部署总览、版本对比、许可六页;这六页的日、简、繁译文按本仓规矩删除,下一轮翻译重新生成。
• 为什么改:原文叫运维开启 Redis 集群驱动,照做会启动失败;另几页承诺了「集群与高可用」,而 cloud 的功能目录明确规定不得承诺 HA。
• 风险与回滚:只改文字;合并即部署,有冒烟检查与自动回滚。
• 需要你做的:无,席位落地。

席位核对

• 形态: Fixes #443, 24 files (6 English pages edited, 18 locale siblings deleted under AGENTS.md's capability-reversal exception), no governed path.
• 事实依据(席位复读): cloud docs/design/ee-feature-catalog.md clustering row — "Multi-node topology is unlocked by the Enterprise license (node cap: maxNodes). Do NOT promise 'HA included'"; cloud apps/objectos-ee/deploy/README.md:107-137; the refusal itself is objectstack packages/services/service-cluster/src/split-brain-guard.ts:64,78-88 (assertClusterDriverSafeForTopology), cited in the dev addendum 6711004269. sales@objectos.ai is the address the site already uses.
• 门禁 check-runs on this head: build success, Node floor success, Ownership & freshness success, Deploy docs skipped (runs on main).
• 落地条款: pr_ready + automerge_enable; the seat follows deploy and smoke to green.