0.10.1: 安全与业务逻辑加固、新品牌与部署加固

三轮审查修复(60+ 项),相对远端 main(b536672)的关键变更:
- 安全: 数据面 SSRF 拨号防护(防 DNS rebinding)/上游凭据剥离/登录防枚举
  与锁定态统一/可信代理(X-Forwarded-For)限流加固/会话版本失效机制/
  撤销即时传播/弱密钥拒绝启动/脱敏字节级重写(保签名契约)
- 业务逻辑: 裸 body 上传 panic/bootstrap 审计管线卡死/定价通配符优先级/
  全局工具可见性/调度器停机补跑/TOTP 挑战令牌消费顺序/熔断探针语义/
  >4MB 响应 token 计量/管理员重置密码作废会话 等
- 前端: 新 logo(语枢 AI 网关主题)/Provider 凭据异常警示/删除入口/
  后端错误消息透传/localStorage 敏感数据收敛
- 部署: CREDENTIAL_MASTER_KEY 持久化与弱值拒绝/Provider DELETE 接口/
  nginx 安全头/worker 内存限制
- 新增迁移 000029(key_hash 索引)/000030(usage_daily 币种维度)
This commit is contained in:
2026-08-13 10:50:51 +08:00
parent b536672000
commit 9501751792
136 changed files with 8024 additions and 1476 deletions
+28
View File
@@ -7,6 +7,8 @@ required. Neither MinIO nor Ollama is a startup dependency: the gateway warns
and refuses file uploads until the bucket is reachable, and knowledge-base
documents are still stored (with embedding set to NULL) when Ollama is down,
with retrieval falling back to full-text search.
The scheduler is a separate stateless worker backed by PostgreSQL leases; run at
least one `scheduler-worker` replica whenever scheduled tasks are enabled.
## Prerequisites
@@ -65,6 +67,7 @@ Check status and logs:
```bash
docker compose --env-file deploy/production.env -f deploy/docker-compose.production.yml ps
docker compose --env-file deploy/production.env -f deploy/docker-compose.production.yml logs --tail=200 gateway-api
docker compose --env-file deploy/production.env -f deploy/docker-compose.production.yml logs --tail=200 scheduler-worker
curl --fail http://127.0.0.1:8080/readyz
```
@@ -77,6 +80,18 @@ The bundled database URLs use `sslmode=disable` only for the private Compose
network. When using an external PostgreSQL or Redis service, require TLS and use
`sslmode=verify-full` / `rediss://` as supported by that service.
### Scheduled tasks
- `scheduler-worker` calls the gateway over `SCHEDULER_GATEWAY_BASE_URL`; keep it
on the private application network and do not expose a scheduler port.
- Task API keys are encrypted with `CREDENTIAL_MASTER_KEY`. Back up the active
key and its historical keyring together with PostgreSQL before rotation.
- `SCHEDULER_MAX_ATTEMPTS` controls both ordinary execution retries and stale
lease recovery. Final success/failure is published through outbox and can be
delivered to the selected notification channel and the creator's inbox.
- Multiple replicas are safe because due tasks and runs are claimed with
PostgreSQL row locks and `SKIP LOCKED`.
### Vectorization and object storage
- The PostgreSQL image is `pgvector/pgvector:pg17` (data-volume compatible with
@@ -88,3 +103,16 @@ network. When using an external PostgreSQL or Redis service, require TLS and use
- Back up the `minio-data` and `ollama-models` volumes alongside PostgreSQL.
- If you previously deployed with `postgres:17-alpine`, back up the PostgreSQL
volume before switching images.
### CREDENTIAL_MASTER_KEY 持久化(重要)
- 所有加密凭据(Provider API Key、TOTP 密钥、Webhook 签名、调度任务 Key、
应用运行时 Key)都用 `CREDENTIAL_MASTER_KEY` 加密。**该密钥必须持久化**
每次部署换新 key 会让全部已存凭据无法解密。
- 本地 compose 首次启动前执行
`openssl rand -base64 32 > deploy/.env`compose 自动读取该文件),
之后重启/重建复用同一 key`.gitignore` 已排除 `deploy/.env`
- 需要轮换时使用管理端「供应商 → 凭据轮换」接口(同一 KEK 版本内重加密),
并保留历史 keyring;不要直接更换 `CREDENTIAL_MASTER_KEY` 值。
- 若误换 key:管理端 Provider 列表会降级显示「凭据无法解密」警示(不会
让整个页面报错),需重新保存各 Provider 的 API Key 恢复。