Migrate SupervisorConfig from confique derive to manual serde impl

- Replace confique-derived SupervisorConfig with manual serde Deserialize
- Add src/config/policy.rs with full restart policy, budget, and severity config
- Add src/config/audit.rs for audit channel configuration
- Rewrite src/config/state.rs with dynamic child and topology state management
- Migrate generate_supervisor binary to src/main.rs
- Update example YAML configs (local, template, supervisor) with new fields
- Sync architecture docs, manuals, and README with new config structure
- Extend test coverage for configurable schema, template, and YAML loading

Author: developerworks

Cargo.lock

@@ -1,8 +1,9 @@
 [package]
 name = "rust-tokio-supervisor"
-version = "0.1.2"
+version = "0.1.3"
 edition = "2024"
 rust-version = "1.88"
+autobins = false
 description = "A Rust tokio supervisor with declarative task supervision, restart policy, shutdown coordination, and observability."
 license = "MIT"
 repository = "https://github.com/developerworks/rust-supervisor"
@@ -32,6 +33,10 @@ include = [
 name = "rust_supervisor"
 path = "src/lib.rs"
 
+[[bin]]
+name = "rust-tokio-supervisor"
+path = "src/main.rs"
+
 [dependencies]
 clap = { version = "4.6.1", features = ["derive"] }
 confique = { version = "0.4.0", features = ["yaml"] }

Cargo.toml

@@ -38,7 +38,7 @@ Package name: `rust-tokio-supervisor`. Library crate name: `rust_supervisor`.
 - Load the primary YAML configuration from `examples/config/supervisor.yaml`.
 - Reuse `rust_supervisor::config::configurable::SupervisorConfig` for YAML loading, template generation, and JSON Schema generation.
 - Emit structured logs, tracing spans, metrics, audit events, event journal entries, and `RunSummary` diagnostics.
-- Enable target-side dashboard IPC through the optional `ipc` configuration section. The target process owns only local Unix domain socket IPC, snapshot generation, event conversion, command mapping, and shared JSON contracts.
+- Enable target-side dashboard IPC through the optional `dashboard` configuration section. The target process owns only local Unix domain socket IPC, snapshot generation, event conversion, command mapping, and shared JSON contracts.
 
 ## Platform Support
 
@@ -117,7 +117,7 @@ The supervisor dashboard feature uses three directories.
 - [rust-supervisor-relay](https://github.com/developerworks/rust-supervisor-relay) at `~/rust-supervisor-relay`: relay server, dynamic registration, `wss://`, mTLS, session gating, and command audit.
 - [rust-supervisor-ui](https://github.com/developerworks/rust-supervisor-ui) at `~/rust-supervisor-ui`: Vue, shadcn-vue, Tailwind dashboard client.
 
-The target process does not expose IPC to the network. It opens a local Unix domain socket only when `ipc.enabled=true`. A relay can read snapshots, but event and log subscriptions must be triggered by an established remote dashboard session.
+The target process does not expose IPC to the network. It opens a local Unix domain socket only when `dashboard.enabled=true`. A relay can read snapshots, but event and log subscriptions must be triggered by an established remote dashboard session.
 
 ![rust-supervisor dashboard screenshot](docs/screenshot.png)
 
@@ -130,12 +130,23 @@ The official YAML files stay single-file by default:
 - `examples/config/supervisor.yaml`: complete runnable configuration.
 - `examples/config/supervisor.template.yaml`: complete single-file template.
 
+The root configuration includes `backpressure` for observability subscriber queues:
+
+```yaml
+backpressure:
+  strategy: alert_and_block
+  warn_threshold_pct: 80
+  critical_threshold_pct: 95
+  window_secs: 30
+  audit_channel_capacity: 1024
+```
+
 This crate does not bake in `x-tree-split`. Projects that want split configuration files can wrap or reuse `SupervisorConfig` in their own crate and decide their own tree split layout.
 
 The optional dashboard IPC section has this shape:
 
 ```yaml
-ipc:
+dashboard:
   enabled: true
   target_id: payments-worker-a
   path: /run/rust-supervisor/payments-worker-a.sock
@@ -149,7 +160,7 @@ ipc:
     registration_heartbeat_interval_seconds: 15
 ```
 
-When `ipc.enabled=true`, `ipc.path` and `ipc.registration.relay_registration_path` must be absolute local paths. Registration uses dynamic registration. The relay configuration must not hard-code target lists.
+When `dashboard.enabled=true`, `dashboard.path` and `dashboard.registration.relay_registration_path` must be absolute local paths. Registration uses dynamic registration. The relay configuration must not hard-code target lists.
 
 ## Quick Start
 

README.md

@@ -36,7 +36,7 @@
 - 从 `examples/config/supervisor.yaml` 加载主 YAML(数据序列化格式) 配置.
 - 复用 `rust_supervisor::config::configurable::SupervisorConfig` 完成 YAML(数据序列化格式) 加载, template generation(模板生成) 和 JSON Schema(JSON 结构模式) 生成.
 - 发出 structured log(结构化日志), tracing span(追踪跨度), metrics(指标), audit event(审计事件), event journal entry(事件日志条目) 和 `RunSummary`(运行摘要) diagnostics(诊断信息).
-- 通过可选 `ipc` 配置启用 target-side dashboard IPC(目标侧看板进程间通信). target process(目标进程) 只拥有本机 Unix domain socket IPC(Unix 域套接字进程间通信), snapshot(快照) 生成, event conversion(事件转换), command mapping(命令映射) 和 shared JSON contract(共享 JSON 契约).
+- 通过可选 `dashboard` 配置启用 target-side dashboard IPC(目标侧看板进程间通信). target process(目标进程) 只拥有本机 Unix domain socket IPC(Unix 域套接字进程间通信), snapshot(快照) 生成, event conversion(事件转换), command mapping(命令映射) 和 shared JSON contract(共享 JSON 契约).
 
 ## 看板
 
@@ -46,7 +46,7 @@ dashboard(看板) 功能固定拆成三个目录.
 - [rust-supervisor-relay](https://github.com/developerworks/rust-supervisor-relay) 位于 `~/rust-supervisor-relay`: relay server(中继服务), dynamic registration(动态注册), `wss://`, mTLS(双向传输层安全协议认证), session gating(会话门控) 和 command audit(命令审计).
 - [rust-supervisor-ui](https://github.com/developerworks/rust-supervisor-ui) 位于 `~/rust-supervisor-ui`: Vue(网页界面框架), shadcn-vue(组件库), Tailwind(样式框架) dashboard client(看板客户端).
 
-target process(目标进程) 不把 IPC(进程间通信) 暴露到外网. 它只在 `ipc.enabled=true` 时打开本机 Unix domain socket(Unix 域套接字). relay(中继) 可以读取 snapshot(快照), 但是 event(事件) 和 log(日志) subscription(订阅) 必须由已认证 dashboard session(看板会话) 触发.
+target process(目标进程) 不把 IPC(进程间通信) 暴露到外网. 它只在 `dashboard.enabled=true` 时打开本机 Unix domain socket(Unix 域套接字). relay(中继) 可以读取 snapshot(快照), 但是 event(事件) 和 log(日志) subscription(订阅) 必须由已认证 dashboard session(看板会话) 触发.
 
 ![rust-supervisor dashboard(看板) screenshot(截图)](docs/screenshot.png)
 
@@ -59,12 +59,23 @@ target process(目标进程) 不把 IPC(进程间通信) 暴露到外网. 它只
 - `examples/config/supervisor.yaml`: 完整可运行配置.
 - `examples/config/supervisor.template.yaml`: 完整单文件模板.
 
+root configuration(根配置) 包含 `backpressure`(背压) 配置, 用来控制 observability subscriber(可观测性订阅者) 队列:
+
+```yaml
+backpressure:
+  strategy: alert_and_block
+  warn_threshold_pct: 80
+  critical_threshold_pct: 95
+  window_secs: 30
+  audit_channel_capacity: 1024
+```
+
 本 crate(包) 不默认写入 `x-tree-split`(树形拆分扩展). 如果使用者项目需要拆分配置文件, 可以在自己的项目中包装或复用 `SupervisorConfig`(监督器配置), 并自行决定 tree split layout(树形拆分布局).
 
 dashboard IPC(看板进程间通信) 的可选配置如下.
 
 ```yaml
-ipc:
+dashboard:
   enabled: true
   target_id: payments-worker-a
   path: /run/rust-supervisor/payments-worker-a.sock
@@ -78,7 +89,7 @@ ipc:
     registration_heartbeat_interval_seconds: 15
 ```
 
-当 `ipc.enabled=true` 时, `ipc.path` 和 `ipc.registration.relay_registration_path` 必须是 absolute path(绝对路径). registration(注册) 使用 dynamic registration(动态注册). relay config(中继配置) 不允许写死 target list(目标列表).
+当 `dashboard.enabled=true` 时, `dashboard.path` 和 `dashboard.registration.relay_registration_path` 必须是 absolute path(绝对路径). registration(注册) 使用 dynamic registration(动态注册). relay config(中继配置) 不允许写死 target list(目标列表).
 
 ## Quick Start(快速开始)
 

README.zh.md

@@ -14,7 +14,7 @@
 │                                                                 │
 │  ┌─────────────────────┐   ┌─────────────────────┐             │
 │  │  Core Library       │   │  Relay               │             │
-│  │  (目标进程)          │   │  (中继进程)           │             │
+│  │  (目标进程)          |   │  (中继进程)           │             │
 │  │                     │   │                      │             │
 │  │  Unix Domain Socket │◄─►│  Unix Domain Socket  │             │
 │  │  /run/.../*.sock    │   │  /run/.../relay.sock │             │
@@ -75,7 +75,7 @@
 YAML 配置
     │
     ▼
-rust-config-tree v0.1.9 加载
+rust-config-tree 加载
     │
     ▼
 ConfigState (不可变配置状态)
@@ -368,7 +368,7 @@ scripts/
 
 - 所有可调配置通过 `rust-config-tree` v0.1.9 从 YAML 加载
 - `SupervisorConfig` 同时支持 `confique::Config`, `schemars::JsonSchema`, `serde::Serialize/Deserialize`
-- `ConfigState` 加载后不可变,派生 `SupervisorSpec`、默认策略、关闭预算和可观测性配置
+- `ConfigState` 加载后不可变,派生 `SupervisorSpec`,默认策略,关闭预算,可观测性配置,背压配置,底层 policy(策略)配置,分组策略和子任务策略覆盖
 - 模块内部不得保存可调配置默认值
 
 ### 4.5 平台编译隔离
@@ -388,6 +388,10 @@ pub mod ipc;
 ```yaml
 supervisor:
   strategy: OneForAll # 监督策略
+  escalation_policy: escalate_to_parent # 默认升级策略
+  dynamic_supervisor:
+    enabled: true # 是否允许运行时增加子任务
+    child_limit: 16 # 子任务总数上限
 
 policy:
   child_restart_limit: 10 # 子任务窗口内最大重启次数
@@ -399,6 +403,27 @@ policy:
   jitter_ratio: 0.10 # 抖动比率
   heartbeat_interval_ms: 1000 # 心跳间隔
   stale_after_ms: 3000 # 心跳过期阈值
+  restart_budget:
+    window_secs: 60 # 重启预算窗口
+    max_burst: 10 # 突发重启次数上限
+    recovery_rate_per_sec: 0.50 # 令牌恢复速率
+  failure_window:
+    mode: time_sliding # 失败窗口模式
+    window_secs: 60 # 时间滑动窗口
+    max_count: 5 # 数量滑动窗口保留条数
+    threshold: 5 # 失败阈值
+  meltdown:
+    child_max_restarts: 3 # 子任务熔断阈值
+    child_window_secs: 10 # 子任务熔断窗口
+    group_max_failures: 5 # 分组熔断阈值
+    group_window_secs: 30 # 分组熔断窗口
+    supervisor_max_failures: 10 # 监督器熔断阈值
+    supervisor_window_secs: 60 # 监督器熔断窗口
+    reset_after_secs: 120 # 稳定后重置窗口
+  supervision_pipeline:
+    journal_capacity: 100 # 策略流水线事件容量
+    subscriber_capacity: 10 # 策略流水线订阅队列容量
+    concurrent_restart_limit: 5 # 并发重启上限
 
 shutdown:
   graceful_timeout_ms: 5000 # 优雅关闭超时
@@ -409,7 +434,61 @@ observability:
   metrics_enabled: true
   audit_enabled: true
 
-ipc: # 可选,仅 Unix
+backpressure:
+  strategy: alert_and_block # 背压策略
+  warn_threshold_pct: 80 # 告警阈值
+  critical_threshold_pct: 95 # 降级阈值
+  window_secs: 30 # 评估窗口
+  audit_channel_capacity: 1024 # 审计通道容量
+
+groups:
+  - name: core # 分组名称
+    children:
+      - api
+    budget:
+      window_secs: 60
+      max_burst: 10
+      recovery_rate_per_sec: 0.50
+  - name: upstream
+    children: []
+
+group_strategies:
+  - group: core
+    strategy: OneForOne
+    restart_limit:
+      max_restarts: 5
+      window_ms: 60000
+    escalation_policy: quarantine_scope
+
+group_dependencies:
+  - from_group: core
+    to_group: upstream
+    propagation: Full
+
+child_strategy_overrides:
+  - child_id: api
+    strategy: RestForOne
+    restart_limit:
+      max_restarts: 3
+      window_ms: 30000
+    escalation_policy: shutdown_tree
+
+severity_defaults:
+  - task_role: service
+    severity: Critical
+
+children:
+  - name: api
+    kind: supervisor
+    criticality: critical
+    tags:
+      - core
+    task_role: supervisor
+    severity: Critical
+    group: core
+    restart_policy: transient
+
+dashboard: # 可选,仅 Unix
   enabled: true
   target_id: payments-worker-a
   path: /run/rust-supervisor/payments-worker-a.sock
@@ -466,7 +545,7 @@ ipc: # 可选,仅 Unix
 
 ## 八、IPC 安全控制点
 
-看板 IPC 配置了 9 项安全控制点 (C1-C9):
+看板 IPC 配置了 9 项安全控制点 (C1-C9). C7 审计持久化使用顶层 `audit` 配置, 不在 `dashboard.security_config` 中重复声明:
 
 | 编号 | 控制点                     | 说明                            |
 | ---- | -------------------------- | ------------------------------- |

docs/architecture.md

@@ -16,7 +16,7 @@
 
 ## 三、IPC 安全控制点 (C1-C9)
 
-看板 IPC 配置了 9 项安全控制点, 经 `IpcSecurityPipeline` 统一编排.
+看板 IPC 配置了 9 项安全控制点, 经 `IpcSecurityPipeline` 统一编排. C7 审计持久化不在 `dashboard.security_config` 中重复配置, 它使用顶层 `audit` 配置.
 
 ### 3.1 执行顺序
 
@@ -130,7 +130,7 @@ SBOM 包含所有直接依赖和传递依赖, 每条依赖记录版本、许可
 ### 6.1 IPC 配置 (Unix only)
 
 ```yaml
-ipc:
+dashboard:
   enabled: true
   path: /run/rust-supervisor/target.sock
   permissions: "0600" # 仅所有者可读写
@@ -153,7 +153,7 @@ ipc:
 | 限制               | 说明                                          | 缓解措施                                              |
 | ------------------ | --------------------------------------------- | ----------------------------------------------------- |
 | IPC 仅 Unix        | Windows 等非 Unix 平台无法使用 dashboard      | 通过 Unix relay 间接访问                              |
-| 审计持久化默认内存 | `AuditBackend` 默认使用 ring buffer, 重启丢失 | 配置 `audit_persistence=file` 落盘                    |
+| 审计持久化默认内存 | `AuditBackend` 默认使用 ring buffer, 重启丢失 | 配置 `audit.backend: file` 和 `audit.file_path` 落盘 |
 | 配置不支持热更新   | 修改安全配置需重启 supervisor                 | 使用配置管理工具自动化重启                            |
 | 无内置 mTLS        | target 侧不处理 mTLS                          | mTLS 由 relay 侧管理, 参考 rust-supervisor-relay 文档 |