dsh-my-observability 0.1.3 → 0.1.5

package/CHANGELOG.md

@@ -2,6 +2,19 @@
 
 本文件记录 dsh-my-observability 的所有版本变更。格式遵循 [Keep a Changelog](https://keepachangelog.com/zh-CN/1.1.0/)，版本号遵循 [语义化版本](https://semver.org/lang/zh-CN/)。
 
+## [0.1.5] - 2026-09-04
+
+### 变更
+
+- feat(observability): #127 资源看门狗自动降级——写入速率/文件大小连续 ≥3 次采样超限自动暂停落盘（事件仍入内存，FIFO 有界不丢）+ 日志告警；连续 ≥3 次正常自动恢复（全量快照补齐降级窗口）；API 暴露 degraded 标记（#136）
+- 发版前功能级验证清单（issue #67）：verification/dsh-my-observability-0.1.5.md
+
+## [0.1.4] - 2026-09-03
+
+### 变更
+
+- feat(observability): #R17 资源监控看板——写放大/资源超限提前发现
+
 ## [0.1.3] - 2026-09-03
 
 ### 变更

package/README.md

@@ -26,6 +26,7 @@ Server 端只读观察 DSH 生命周期事件并记录审计日志：
 - **重启恢复**：持久化到 `$DSH_HOME/observability/audit.jsonl`（**增量追加** + 防抖批量 flush + 周期 compact 原子快照），重启后完整恢复；升级前旧格式 `audit.json` 自动迁移；
 - **防膨胀**：每会话最多 2000 条（FIFO 淘汰）、全局 20000 条（轮转淘汰）；
 - **零写放大**：落盘只写新增事件（≈事件本体字节），不会因事件流持续而反复全量重写审计文件；
+- **资源监控 + 自动降级（资源看门狗）**：面板「资源」区块展示本进程 CPU/内存 + 审计文件大小与写入速率（15s 采样），写放大/超限自动告警（告警列表展示在「资源」区块，阈值见 `resource-budget-review`）；写入速率/文件大小连续超限自动**暂停落盘**（事件仍入内存，有界不丢），回落自动恢复（全量快照补齐降级窗口），全程日志告警 + API `degraded` 标记（issue #127）。
 - **只读观察**：waterfall 事件一律透传 `next()`，绝不改变工具/模型流程。
 
 ### 2. 轨迹回放面板（时间轴）

package/lib/client.js

@@ -42,6 +42,12 @@ function isZh() {
 
 const strings = {
   replayTitle: () => (isZh() ? '轨迹回放' : 'Trajectory'),
+  resourceTitle: () => (isZh() ? '资源监控' : 'Resources'),
+  resourceLoading: () => (isZh() ? '资源采样中…' : 'Sampling…'),
+  resourceFile: () => (isZh() ? '审计文件' : 'Audit file'),
+  resourceRate: () => (isZh() ? '写入速率' : 'Write rate'),
+  resourceCpu: () => (isZh() ? 'CPU' : 'CPU'),
+  resourceMem: () => (isZh() ? '内存' : 'Memory'),
   gitTitle: () => (isZh() ? 'Git 工具' : 'Git Tools'),
   allSessions: () => (isZh() ? '全部会话' : 'All sessions'),
   filterAll: () => (isZh() ? '全部' : 'All'),
@@ -986,6 +992,7 @@ function ReplayPanel(props) {
   return createElement(
     'div',
     { className: 'dsh-my-observability-panel' },
+    createElement(ResourcePanel, { resource: s.resource }),
     createElement(ReplayToolbar, {
       sessions: s.sessions,
       selected: s.selected,
@@ -1026,6 +1033,78 @@ function ReplayPanel(props) {
   )
 }
 
+    // ── 资源监控区块（写放大/资源超限预警，见 lib/resource-monitor.js）──────
+// 依赖 replay.js 先拼接（apiJson）与 i18n.js（strings）。纯函数声明文本。
+
+const RESOURCE_POLL_MS = 15000
+
+function fmtResourceBytes(bytes) {
+  if (!Number.isFinite(bytes)) return '-'
+  return `${(bytes / 1048576).toFixed(1)} MB`
+}
+
+/** 资源采样状态：可见时每 15s 轮询 /observability/api/resources。 */
+function useResourceState(visible) {
+  const [resource, setResource] = useState(null)
+  useEffect(() => {
+    if (!visible) return undefined
+    let alive = true
+    const tick = () => {
+      if (alive) apiJson('/observability/api/resources').then(setResource).catch(() => {})
+    }
+    tick()
+    const timer = setInterval(tick, RESOURCE_POLL_MS)
+    return () => {
+      alive = false
+      clearInterval(timer)
+    }
+  }, [visible])
+  return resource
+}
+
+function ResourceMetric({ label, value }) {
+  return createElement(
+    'div',
+    { className: 'dsh-my-observability-resource-metric' },
+    createElement('span', { className: 'dsh-my-observability-resource-label' }, label),
+    createElement('span', { className: 'dsh-my-observability-resource-value' }, value),
+  )
+}
+
+/** 资源面板：四指标 + 告警列表（write-rate/file-size level=error 红色，cpu/memory warn 黄色）。 */
+function ResourcePanel({ resource }) {
+  if (resource === null || resource === undefined) {
+    return createElement('div', { className: 'dsh-my-observability-resource' }, strings.resourceLoading())
+  }
+  const alerts = Array.isArray(resource.alerts) ? resource.alerts : []
+  return createElement(
+    'div',
+    { className: 'dsh-my-observability-resource' },
+    createElement('div', { className: 'dsh-my-observability-resource-head' }, strings.resourceTitle()),
+    createElement(
+      'div',
+      { className: 'dsh-my-observability-resource-grid' },
+      createElement(ResourceMetric, { label: strings.resourceFile(), value: fmtResourceBytes(resource.fileBytes) }),
+      createElement(ResourceMetric, { label: strings.resourceRate(), value: `${fmtResourceBytes(resource.writeRateBytesPerHour)}/h` }),
+      createElement(ResourceMetric, { label: strings.resourceCpu(), value: `${Math.round(resource.cpuPercent ?? 0)}%` }),
+      createElement(ResourceMetric, { label: strings.resourceMem(), value: fmtResourceBytes(resource.memoryBytes) }),
+    ),
+    alerts.length > 0
+      ? createElement(
+          'div',
+          { className: 'dsh-my-observability-resource-alerts' },
+          alerts.map((alert) =>
+            createElement(
+              'div',
+              { className: `dsh-my-observability-resource-alert dsh-my-observability-resource-alert-${alert.level}` },
+              alert.message,
+            ),
+          ),
+        )
+      : null,
+  )
+}
+
     // ── 审计视图扩展：搜索 / 组合过滤 / 导出 / 统计 / 高亮（replay.js 拆出）────
 // 依赖 replay.js（REPLAY_POLL_MS 等常量与 loadReplayData/EventRow）与
 // audit-view.js 纯函数（applyAuditFilter 等）。始终以 function 声明提升。
@@ -1300,6 +1379,22 @@ function useReplayDataState(props) {
   const [loading, setLoading] = useState(true)
   const [error, setError] = useState('')
   const [reloadTick, setReloadTick] = useState(0)
+  const [resource, setResource] = useState(null)
+
+  // 资源采样轮询（写放大/资源超限预警；可见时 15s 一次，隐藏暂停）
+  useEffect(() => {
+    if (!visible) return undefined
+    let alive = true
+    const tick = () => {
+      if (alive) apiJson('/observability/api/resources').then(setResource).catch(() => {})
+    }
+    tick()
+    const timer = setInterval(tick, RESOURCE_POLL_MS)
+    return () => {
+      alive = false
+      clearInterval(timer)
+    }
+  }, [visible])
 
   useEffect(() => {
     if (!visible) return undefined
@@ -1324,6 +1419,7 @@ function useReplayDataState(props) {
 
   return {
     currentSession,
+    resource,
     sessions,
     selected,
     events,
@@ -1372,6 +1468,7 @@ function useReplayState(props) {
   const onExport = (format) => void runExport(format, scope, filtered, criteria, data.setError)
 
   return {
+    resource: data.resource,
     sessions: data.sessions,
     selected: data.selected,
     events: data.events,
@@ -1826,6 +1923,16 @@ const STYLES = `
 .dsh-my-observability-review-ok{font:var(--dsw-font-xxs-strong-12);color:var(--dsw-alias-state-success-primary)}
 .dsh-my-observability-ai{font:var(--dsw-font-xxs-12);color:var(--dsw-alias-label-secondary);line-height:1.5;
   border:1px dashed var(--dsw-alias-border-l2);border-radius:6px;padding:6px 8px}
+.dsh-my-observability-resource{border:1px solid var(--dsw-alias-border-l2);border-radius:8px;padding:8px;margin:0 0 8px}
+.dsh-my-observability-resource-head{font:var(--dsw-font-xxs-strong-12);color:var(--dsw-alias-label-primary);margin-bottom:6px}
+.dsh-my-observability-resource-grid{display:grid;grid-template-columns:1fr 1fr;gap:4px 10px}
+.dsh-my-observability-resource-metric{display:flex;justify-content:space-between;gap:8px;font:var(--dsw-font-xxs-12)}
+.dsh-my-observability-resource-label{color:var(--dsw-alias-label-secondary)}
+.dsh-my-observability-resource-value{color:var(--dsw-alias-label-primary);font-family:var(--dsw-font-mono-xxs)}
+.dsh-my-observability-resource-alerts{margin-top:6px;display:flex;flex-direction:column;gap:4px}
+.dsh-my-observability-resource-alert{font:var(--dsw-font-xxxs-11);border-radius:4px;padding:2px 6px}
+.dsh-my-observability-resource-alert-error{color:var(--dsw-alias-state-error-primary);background:color-mix(in srgb,var(--dsw-alias-state-error-primary) 12%,transparent)}
+.dsh-my-observability-resource-alert-warn{color:var(--dsw-alias-state-warn-primary);background:color-mix(in srgb,var(--dsw-alias-state-warn-primary) 12%,transparent)}
 `
 
 function injectStyles() {

package/lib/client.src.js

@@ -34,6 +34,7 @@ window.__ModuleLoader__.load({
     /*__PART_ICONS__*/
     /*__PART_AUDIT_VIEW__*/
     /*__PART_REPLAY__*/
+    /*__PART_RESOURCE__*/
     /*__PART_REPLAY_EXT__*/
     /*__PART_GIT__*/
     /*__PART_STYLES__*/

package/lib/index.js

@@ -22,6 +22,7 @@
 import { createStore } from './store.js'
 import { attachAuditListeners } from './audit.js'
 import { registerObservabilityRoutes } from './routes.js'
+import { createResourceMonitor } from './resource-monitor.js'
 
 export const name = 'dsh-my-observability'
 
@@ -40,9 +41,30 @@ export function apply(ctx, config) {
   // ── 事件监听（只读观察；waterfall 一律透传 next()）──────────────────
   attachAuditListeners(ctx, store.record)
 
-  // ── 路由（查询 / git 工具 / diff 审查）──────────────────────────────
-  registerObservabilityRoutes(ctx, store, options)
+  // ── 资源监控 + 降级看门狗（15s 采样 CPU/内存/审计写入速率；写放大/文件
+  //    超限连续触发时自动降级停落盘，资源回归后自动恢复——issue #127）────
+  const monitor = createResourceMonitor(ctx, {
+    intervalMs: config?.resourceIntervalMs,
+    limits: config?.resourceLimits,
+    onDegrade: () => {
+      store.setPersistEnabled(false)
+      ctx.logger.warn(
+        '[dsh-my-observability] 资源看门狗：审计写入速率/文件大小连续超限，已暂停落盘（事件仍在内存，恢复后全量快照补齐）',
+      )
+    },
+    onRecover: () => {
+      store.setPersistEnabled(true)
+      ctx.logger.warn('[dsh-my-observability] 资源看门狗：写入速率回归正常，已恢复审计落盘')
+    },
+  })
+  monitor.start()
 
-  // ── 卸载冲刷：清防抖定时器 + 立即落盘 ───────────────────────────────
-  ctx.effect(() => store.dispose, 'dsh-my-observability: persistence teardown')
+  // ── 路由（查询 / git 工具 / diff 审查 / 资源）───────────────────────
+  registerObservabilityRoutes(ctx, store, monitor, options)
+
+  // ── 卸载冲刷：清防抖定时器 + 立即落盘 + 停采样 ──────────────────────
+  ctx.effect(() => {
+    monitor.stop()
+    return store.dispose
+  }, 'dsh-my-observability: persistence teardown')
 }

package/lib/parts/i18n.js

@@ -10,6 +10,12 @@ function isZh() {
 
 const strings = {
   replayTitle: () => (isZh() ? '轨迹回放' : 'Trajectory'),
+  resourceTitle: () => (isZh() ? '资源监控' : 'Resources'),
+  resourceLoading: () => (isZh() ? '资源采样中…' : 'Sampling…'),
+  resourceFile: () => (isZh() ? '审计文件' : 'Audit file'),
+  resourceRate: () => (isZh() ? '写入速率' : 'Write rate'),
+  resourceCpu: () => (isZh() ? 'CPU' : 'CPU'),
+  resourceMem: () => (isZh() ? '内存' : 'Memory'),
   gitTitle: () => (isZh() ? 'Git 工具' : 'Git Tools'),
   allSessions: () => (isZh() ? '全部会话' : 'All sessions'),
   filterAll: () => (isZh() ? '全部' : 'All'),

package/lib/parts/replay-ext.js

@@ -272,6 +272,25 @@ function useReplayDataState(props) {
   const [loading, setLoading] = useState(true)
   const [error, setError] = useState('')
   const [reloadTick, setReloadTick] = useState(0)
+  const [resource, setResource] = useState(null)
+
+  // 资源采样轮询（写放大/资源超限预警；可见时 15s 一次，隐藏暂停）
+  useEffect(() => {
+    if (!visible) return undefined
+    let alive = true
+    const tick = () => {
+      if (alive)
+        apiJson('/observability/api/resources')
+          .then(setResource)
+          .catch(() => {})
+    }
+    tick()
+    const timer = setInterval(tick, RESOURCE_POLL_MS)
+    return () => {
+      alive = false
+      clearInterval(timer)
+    }
+  }, [visible])
 
   useEffect(() => {
     if (!visible) return undefined
@@ -296,6 +315,7 @@ function useReplayDataState(props) {
 
   return {
     currentSession,
+    resource,
     sessions,
     selected,
     events,
@@ -344,6 +364,7 @@ function useReplayState(props) {
   const onExport = (format) => void runExport(format, scope, filtered, criteria, data.setError)
 
   return {
+    resource: data.resource,
     sessions: data.sessions,
     selected: data.selected,
     events: data.events,

package/lib/parts/replay.js

@@ -279,6 +279,7 @@ function ReplayPanel(props) {
   return createElement(
     'div',
     { className: 'dsh-my-observability-panel' },
+    createElement(ResourcePanel, { resource: s.resource }),
     createElement(ReplayToolbar, {
       sessions: s.sessions,
       selected: s.selected,

package/lib/parts/resource.js

@@ -0,0 +1,80 @@
+// ── 资源监控区块（写放大/资源超限预警，见 lib/resource-monitor.js）──────
+// 依赖 replay.js 先拼接（apiJson）与 i18n.js（strings）。纯函数声明文本。
+
+const RESOURCE_POLL_MS = 15000
+
+function fmtResourceBytes(bytes) {
+  if (!Number.isFinite(bytes)) return '-'
+  return `${(bytes / 1048576).toFixed(1)} MB`
+}
+
+/** 资源采样状态：可见时每 15s 轮询 /observability/api/resources。 */
+function useResourceState(visible) {
+  const [resource, setResource] = useState(null)
+  useEffect(() => {
+    if (!visible) return undefined
+    let alive = true
+    const tick = () => {
+      if (alive)
+        apiJson('/observability/api/resources')
+          .then(setResource)
+          .catch(() => {})
+    }
+    tick()
+    const timer = setInterval(tick, RESOURCE_POLL_MS)
+    return () => {
+      alive = false
+      clearInterval(timer)
+    }
+  }, [visible])
+  return resource
+}
+
+function ResourceMetric({ label, value }) {
+  return createElement(
+    'div',
+    { className: 'dsh-my-observability-resource-metric' },
+    createElement('span', { className: 'dsh-my-observability-resource-label' }, label),
+    createElement('span', { className: 'dsh-my-observability-resource-value' }, value),
+  )
+}
+
+/** 资源面板：四指标 + 告警列表（write-rate/file-size level=error 红色，cpu/memory warn 黄色）。 */
+function ResourcePanel({ resource }) {
+  if (resource === null || resource === undefined) {
+    return createElement('div', { className: 'dsh-my-observability-resource' }, strings.resourceLoading())
+  }
+  const alerts = Array.isArray(resource.alerts) ? resource.alerts : []
+  return createElement(
+    'div',
+    { className: 'dsh-my-observability-resource' },
+    createElement('div', { className: 'dsh-my-observability-resource-head' }, strings.resourceTitle()),
+    createElement(
+      'div',
+      { className: 'dsh-my-observability-resource-grid' },
+      createElement(ResourceMetric, { label: strings.resourceFile(), value: fmtResourceBytes(resource.fileBytes) }),
+      createElement(ResourceMetric, {
+        label: strings.resourceRate(),
+        value: `${fmtResourceBytes(resource.writeRateBytesPerHour)}/h`,
+      }),
+      createElement(ResourceMetric, {
+        label: strings.resourceCpu(),
+        value: `${Math.round(resource.cpuPercent ?? 0)}%`,
+      }),
+      createElement(ResourceMetric, { label: strings.resourceMem(), value: fmtResourceBytes(resource.memoryBytes) }),
+    ),
+    alerts.length > 0
+      ? createElement(
+          'div',
+          { className: 'dsh-my-observability-resource-alerts' },
+          alerts.map((alert) =>
+            createElement(
+              'div',
+              { className: `dsh-my-observability-resource-alert dsh-my-observability-resource-alert-${alert.level}` },
+              alert.message,
+            ),
+          ),
+        )
+      : null,
+  )
+}

package/lib/parts/styles.js

@@ -123,6 +123,16 @@ const STYLES = `
 .dsh-my-observability-review-ok{font:var(--dsw-font-xxs-strong-12);color:var(--dsw-alias-state-success-primary)}
 .dsh-my-observability-ai{font:var(--dsw-font-xxs-12);color:var(--dsw-alias-label-secondary);line-height:1.5;
   border:1px dashed var(--dsw-alias-border-l2);border-radius:6px;padding:6px 8px}
+.dsh-my-observability-resource{border:1px solid var(--dsw-alias-border-l2);border-radius:8px;padding:8px;margin:0 0 8px}
+.dsh-my-observability-resource-head{font:var(--dsw-font-xxs-strong-12);color:var(--dsw-alias-label-primary);margin-bottom:6px}
+.dsh-my-observability-resource-grid{display:grid;grid-template-columns:1fr 1fr;gap:4px 10px}
+.dsh-my-observability-resource-metric{display:flex;justify-content:space-between;gap:8px;font:var(--dsw-font-xxs-12)}
+.dsh-my-observability-resource-label{color:var(--dsw-alias-label-secondary)}
+.dsh-my-observability-resource-value{color:var(--dsw-alias-label-primary);font-family:var(--dsw-font-mono-xxs)}
+.dsh-my-observability-resource-alerts{margin-top:6px;display:flex;flex-direction:column;gap:4px}
+.dsh-my-observability-resource-alert{font:var(--dsw-font-xxxs-11);border-radius:4px;padding:2px 6px}
+.dsh-my-observability-resource-alert-error{color:var(--dsw-alias-state-error-primary);background:color-mix(in srgb,var(--dsw-alias-state-error-primary) 12%,transparent)}
+.dsh-my-observability-resource-alert-warn{color:var(--dsw-alias-state-warn-primary);background:color-mix(in srgb,var(--dsw-alias-state-warn-primary) 12%,transparent)}
 `
 
 function injectStyles() {

package/lib/resource-monitor.js

@@ -0,0 +1,118 @@
+/**
+ * dsh-my-observability — 资源采样监控 + 降级看门狗。
+ *
+ * 每 intervalMs（默认 15s）采样本进程 CPU/内存与审计文件大小/写入速率，
+ * 保留最近 MAX_HISTORY 个样本（ring buffer），并做阈值评估（resource-rules）。
+ * 用途：让「写放大/资源超限」在运行期当小时可见（9/2 事故复盘结论：
+ * 15 小时 300GB 写入零监控是事故未被及时发现的主因）。
+ *
+ * 降级看门狗（issue #127 资源占用防护）：关键阈值（写放大/文件字节）连续
+ * DEGRADE_CONFIRM_COUNT 次超限 → 触发 onDegrade 回调（宿主降级：停落盘等）；
+ * 连续 RECOVER_CONFIRM_COUNT 次正常 → 触发 onRecover 回调（宿主恢复 + 全量快照）。
+ * 判定为纯函数（resource-rules.shouldEnterDegrade/shouldExitDegrade），可单测。
+ *
+ * 采样自身开销：15s 一次 process.cpuUsage/memoryUsage + fs.stat（<0.01% CPU、
+ * 零分配大对象），远低于「监控不能放大被监控对象」的护栏（resource-budget-review）。
+ */
+import { statSync } from 'node:fs'
+import { evaluateResourceAlerts, shouldEnterDegrade, shouldExitDegrade, DEFAULT_LIMITS } from './resource-rules.js'
+import { jsonlFile } from './store-persist.js'
+
+const DEFAULT_INTERVAL_MS = 15000
+const MAX_HISTORY = 60
+
+/** 创建资源监控器：{ sample, start, stop }。options: intervalMs / limits / onDegrade / onRecover。 */
+export function createResourceMonitor(ctx, options = {}) {
+  const intervalMs =
+    Number.isFinite(options.intervalMs) && options.intervalMs > 0 ? options.intervalMs : DEFAULT_INTERVAL_MS
+  const limits = { ...DEFAULT_LIMITS, ...(options.limits ?? {}) }
+  const onDegrade = typeof options.onDegrade === 'function' ? options.onDegrade : null
+  const onRecover = typeof options.onRecover === 'function' ? options.onRecover : null
+  const state = {
+    timer: null,
+    file: jsonlFile(),
+    lastSample: null,
+    lastCpu: process.cpuUsage(),
+    history: [],
+    degraded: false,
+  }
+  const monitor = {
+    sample: () => sample(state, limits, onDegrade, onRecover),
+    start: () => startMonitor(state, intervalMs, monitor),
+    stop: () => stopMonitor(state),
+    isDegraded: () => state.degraded,
+  }
+  return monitor
+}
+
+/** 采样一次：CPU 使用率（窗口内 user+sys）/RSS/审计文件字节/写入速率 + 告警 + 降级判定。 */
+function sample(state, limits, onDegrade, onRecover) {
+  const now = Date.now()
+  const cpu = process.cpuUsage()
+  const cpuDelta = cpu.user - state.lastCpu.user + (cpu.system - state.lastCpu.system) // µs
+  state.lastCpu = cpu
+  const mem = process.memoryUsage()
+  const memoryBytes = mem.rss
+  let fileBytes = 0
+  try {
+    fileBytes = statSync(state.file).size
+  } catch {
+    // 审计文件尚未创建：字节为 0
+  }
+  const prev = state.lastSample
+  if (prev !== null) {
+    const deltaMs = Math.max(now - prev.time, 1)
+    // CPU 单核折算：cpuDelta(µs) / deltaMs(ms) / 1000 → 百分比（×100）
+    const cpuPercent = (cpuDelta / 1000 / deltaMs) * 100
+    const byteDelta = fileBytes - prev.fileBytes
+    const writeRateBytesPerHour = byteDelta > 0 ? (byteDelta / deltaMs) * 3600 * 1000 : 0
+    const sample = { time: now, cpuPercent, memoryBytes, fileBytes, writeRateBytesPerHour }
+    state.history.push(sample)
+    if (state.history.length > MAX_HISTORY) state.history.splice(0, state.history.length - MAX_HISTORY)
+    state.lastSample = sample
+    updateDegradeState(state, limits, onDegrade, onRecover)
+    return {
+      ...sample,
+      history: [...state.history],
+      alerts: evaluateResourceAlerts(sample, limits),
+      degraded: state.degraded,
+    }
+  }
+  state.lastSample = { time: now, fileBytes, memoryBytes, cpuPercent: 0, writeRateBytesPerHour: 0 }
+  return { ...state.lastSample, history: [...state.history], alerts: [], degraded: state.degraded }
+}
+
+/**
+ * 降级状态机：未降级且连续超限 → 进入降级（回调）；已降级且连续正常 → 退出降级（回调）。
+ * 判定纯函数见 resource-rules.js；本函数只持有状态并触发宿主回调。
+ */
+function updateDegradeState(state, limits, onDegrade, onRecover) {
+  if (!state.degraded) {
+    if (shouldEnterDegrade(state.history, limits)) {
+      state.degraded = true
+      onDegrade?.()
+    }
+  } else if (shouldExitDegrade(state.history, limits)) {
+    state.degraded = false
+    onRecover?.()
+  }
+}
+
+/** 启动周期采样（幂等）。 */
+function startMonitor(state, intervalMs, monitor) {
+  if (state.timer === null) {
+    state.timer = setInterval(() => {
+      void monitor.sample()
+    }, intervalMs)
+    if (state.timer.unref) state.timer.unref()
+  }
+  return state.timer
+}
+
+/** 停止采样（幂等）。 */
+function stopMonitor(state) {
+  if (state.timer !== null) {
+    clearInterval(state.timer)
+    state.timer = null
+  }
+}

package/lib/resource-rules.js

@@ -0,0 +1,102 @@
+/**
+ * dsh-my-observability — 资源阈值判定（纯函数）。
+ *
+ * 阈值口径与 skill `resource-budget-review` 一致（设计/开发/运行三遍校验
+ * 的运行期部分）：写放大/CPU/内存超限在采样数据上提前暴露，告警可查询。
+ * 纯函数便于单测与配置覆盖（config.resourceLimits 浅合并）。
+ */
+
+/** 默认阈值（DSH 插件场景，来源 resource-budget-review 五维表）。 */
+export const DEFAULT_LIMITS = {
+  /** 审计文件写入速率上限：50 MB/小时。 */
+  writeRateBytesPerHour: 50 * 1024 * 1024,
+  /** 审计文件大小上限：50 MB（循环 compact 后应远小于此）。 */
+  fileBytes: 50 * 1024 * 1024,
+  /** 本进程 CPU 均值上限：10%（单核折算，采样窗口内 user+sys / 时长）。 */
+  cpuPercent: 10,
+  /** 本进程 RSS 上限：500 MB。 */
+  memoryBytes: 500 * 1024 * 1024,
+}
+
+/** 采样数据 → 告警列表（[{ rule, level, message, value, limit }]）。 */
+export function evaluateResourceAlerts(sample, limits = DEFAULT_LIMITS) {
+  const alerts = []
+  if (typeof sample.writeRateBytesPerHour === 'number' && sample.writeRateBytesPerHour > limits.writeRateBytesPerHour) {
+    alerts.push({
+      rule: 'write-rate',
+      level: 'error',
+      message: `审计写入速率 ${fmtMB(sample.writeRateBytesPerHour)}/h 超过上限 ${fmtMB(limits.writeRateBytesPerHour)}/h（写放大风险）`,
+      value: sample.writeRateBytesPerHour,
+      limit: limits.writeRateBytesPerHour,
+    })
+  }
+  if (typeof sample.fileBytes === 'number' && sample.fileBytes > limits.fileBytes) {
+    alerts.push({
+      rule: 'file-size',
+      level: 'error',
+      message: `审计文件 ${fmtMB(sample.fileBytes)} 超过上限 ${fmtMB(limits.fileBytes)}（请检查 compact/轮转）`,
+      value: sample.fileBytes,
+      limit: limits.fileBytes,
+    })
+  }
+  if (typeof sample.cpuPercent === 'number' && sample.cpuPercent > limits.cpuPercent) {
+    alerts.push({
+      rule: 'cpu',
+      level: 'warn',
+      message: `本进程 CPU ${Math.round(sample.cpuPercent)}% 超过上限 ${limits.cpuPercent}%`,
+      value: sample.cpuPercent,
+      limit: limits.cpuPercent,
+    })
+  }
+  if (typeof sample.memoryBytes === 'number' && sample.memoryBytes > limits.memoryBytes) {
+    alerts.push({
+      rule: 'memory',
+      level: 'warn',
+      message: `本进程内存 ${fmtMB(sample.memoryBytes)} 超过上限 ${fmtMB(limits.memoryBytes)}`,
+      value: sample.memoryBytes,
+      limit: limits.memoryBytes,
+    })
+  }
+  return alerts
+}
+
+/** 连续多少次采样超限（关键阈值）判定进入降级（模块内默认值）。 */
+const DEGRADE_CONFIRM_COUNT = 3
+
+/** 连续多少次采样正常判定退出降级（模块内默认值）。 */
+const RECOVER_CONFIRM_COUNT = 3
+
+/**
+ * 关键阈值判定：磁盘写放大风险主导（write-rate / file-size 任一超限即记超限）。
+ * CPU/内存超限告警但不触发落盘降级（避免误伤正常大请求峰值）。
+ */
+function isCriticalOverLimit(sample, limits = DEFAULT_LIMITS) {
+  return (
+    (sample.writeRateBytesPerHour ?? 0) > limits.writeRateBytesPerHour || (sample.fileBytes ?? 0) > limits.fileBytes
+  )
+}
+
+/**
+ * 降级判定（纯函数）：最近 confirmCount 个样本**全部**关键阈值超限 → 进入降级。
+ * 历史样本不足 confirmCount 时不降级（冷启动保护）。
+ * 返回布尔（shouldEnterDegrade）。
+ */
+export function shouldEnterDegrade(history, limits = DEFAULT_LIMITS, confirmCount = DEGRADE_CONFIRM_COUNT) {
+  const recent = history.slice(-confirmCount)
+  if (recent.length < confirmCount) return false
+  return recent.every((sample) => isCriticalOverLimit(sample, limits))
+}
+
+/**
+ * 恢复判定（纯函数）：最近 confirmCount 个样本**全部**关键阈值正常 → 退出降级。
+ * 历史样本不足 confirmCount 时不恢复（避免刚降级立即恢复抖动）。
+ */
+export function shouldExitDegrade(history, limits = DEFAULT_LIMITS, confirmCount = RECOVER_CONFIRM_COUNT) {
+  const recent = history.slice(-confirmCount)
+  if (recent.length < confirmCount) return false
+  return recent.every((sample) => !isCriticalOverLimit(sample, limits))
+}
+
+function fmtMB(bytes) {
+  return `${(bytes / 1024 / 1024).toFixed(1)} MB`
+}

package/lib/routes.js

@@ -17,7 +17,7 @@ import { reviewRules } from './review.js'
 import { runAiReview } from './ai.js'
 
 /** 注册 /observability/api 路由（effect 持有 disposer）。 */
-export function registerObservabilityRoutes(ctx, store, options) {
+export function registerObservabilityRoutes(ctx, store, monitor, options) {
   const webRuntime = ctx.get ? ctx.get('webRuntime') : undefined
   const trustedHosts =
     webRuntime !== undefined && webRuntime !== null && Array.isArray(webRuntime.trustedHosts)
@@ -30,14 +30,14 @@ export function registerObservabilityRoutes(ctx, store, options) {
       ctx.webServer.register({
         kind: 'prefix',
         path: '/observability/api',
-        handler: apiHandler(ctx, fence, store, options),
+        handler: apiHandler(ctx, fence, store, monitor, options),
       }),
     'dsh-my-observability: /observability/api routes',
   )
 }
 
 /** 统一 handler：fence → 方法分派 → 404/错误兜底。 */
-function apiHandler(ctx, fence, store, options) {
+function apiHandler(ctx, fence, store, monitor, options) {
   return async (request, response) => {
     if (!fence(request)) {
       writeJson(response, 403, { ok: false, error: { code: 'forbidden', message: 'forbidden' } })
@@ -47,7 +47,7 @@ function apiHandler(ctx, fence, store, options) {
     const pathname = url.pathname
     const method = pathname.startsWith('/observability/api/') ? pathname.slice('/observability/api/'.length) : undefined
     try {
-      const handled = await dispatchMethod(method, request, response, url, ctx, store, options)
+      const handled = await dispatchMethod(method, request, response, url, ctx, store, monitor, options)
       if (!handled) {
         writeJson(response, 404, {
           ok: false,
@@ -66,11 +66,15 @@ function isMethod(method, request, name, verb) {
 }
 
 /** 按 method 分派到具体 handler；未识别返回 false（调用方回 404）。 */
-async function dispatchMethod(method, request, response, url, ctx, store, options) {
+async function dispatchMethod(method, request, response, url, ctx, store, monitor, options) {
   if (isMethod(method, request, 'sessions', 'GET')) {
     writeJson(response, 200, { ok: true, value: store.sessions() })
     return true
   }
+  if (isMethod(method, request, 'resources', 'GET')) {
+    writeJson(response, 200, { ok: true, value: monitor.sample() })
+    return true
+  }
   if (isMethod(method, request, 'events', 'GET')) {
     writeJson(response, 200, {
       ok: true,

package/lib/store-persist.js

@@ -24,7 +24,7 @@ export const MAX_TOTAL_EVENTS = 20000
 export const COMPACT_LINES = 5000
 
 /** 数据根目录：$DSH_HOME（fallback 家目录）。 */
-export function dataHome() {
+function dataHome() {
   const home = process.env.DSH_HOME
   return typeof home === 'string' && home !== '' ? home : homedir()
 }
@@ -40,7 +40,7 @@ export function legacyFile() {
 }
 
 /** 事件结构校验（时间/会话/类型为合理形态）。 */
-export function isValidEvent(event) {
+function isValidEvent(event) {
   return (
     event !== null &&
     typeof event === 'object' &&
@@ -65,7 +65,9 @@ export function normalizeLoaded(bySession) {
   let total = 0
   for (const bucket of Object.values(sessionBuckets)) total += bucket.events.length
   if (total > MAX_TOTAL_EVENTS) {
-    const ordered = Object.keys(sessionBuckets).sort((a, b) => firstTimeOf(sessionBuckets[a]) - firstTimeOf(sessionBuckets[b]))
+    const ordered = Object.keys(sessionBuckets).sort(
+      (a, b) => firstTimeOf(sessionBuckets[a]) - firstTimeOf(sessionBuckets[b]),
+    )
     for (const sessionId of ordered) {
       if (total <= MAX_TOTAL_EVENTS) break
       total -= sessionBuckets[sessionId].events.length

package/lib/store.js

@@ -53,12 +53,14 @@ export function createStore(ctx) {
     compactTimer: null,
     compacting: false,
     migrated: false,
+    persistEnabled: true,
     dirtyChain: Promise.resolve(),
   }
   store.record = (event) => record(handle, event)
   store.events = (sessionId, type, limit) => eventsOf(handle, sessionId, type, limit)
   store.sessions = () => sessionsOf(handle)
   store.count = () => countOf(handle)
+  store.setPersistEnabled = (enabled) => setPersistEnabled(handle, enabled)
   store.dispose = () => dispose(handle)
   void loadPersisted(handle.file, handle.legacy).then((result) => onLoaded(handle, result))
   return store
@@ -76,14 +78,34 @@ function record(handle, event) {
   return item
 }
 
-/** 事件入内存桶 + 排队待落盘行（回放与运行时共用，保证回放也落盘）。 */
+/** 事件入内存桶 + 排队待落盘行（回放与运行时共用，保证回放也落盘）。
+ *  降级（persistEnabled=false）时事件照常入内存桶（有 FIFO 上限保护），
+ *  但不排队落盘行、不触发 flush——写盘完全停止，恢复时全量快照补齐。 */
 function enqueueRecord(handle, item) {
   appendEvent(handle, item)
+  if (!handle.persistEnabled) return
   handle.lineQueue.push(JSON.stringify(item))
   handle.queuedLines += 1
   if (handle.queuedLines >= COMPACT_LINES) scheduleCompact(handle)
 }
 
+/**
+ * 落盘开关（资源看门狗降级用）：
+ *  - false（降级）：停止落盘——清空待写行（内存 state 仍在，不丢），停 flush。
+ *  - true（恢复）：立即 compact 全量快照（内存=真相，补齐降级窗口事件），恢复增量。
+ */
+function setPersistEnabled(handle, enabled) {
+  if (handle.persistEnabled === enabled) return
+  handle.persistEnabled = enabled
+  if (handle.flushTimer !== null) {
+    clearTimeout(handle.flushTimer)
+    handle.flushTimer = null
+  }
+  handle.lineQueue = []
+  handle.queuedLines = 0
+  if (enabled) compactNow(handle)
+}
+
 /** 全部会话事件：合并各会话并按时间正序（sessionId='*'）。 */
 function eventsAllOf(handle, type, limit) {
   let all = []
@@ -209,12 +231,10 @@ function scheduleFlush(handle) {
 }
 
 function flushNow(handle) {
-  if (handle.lineQueue.length === 0) return
+  if (handle.lineQueue.length === 0 || !handle.persistEnabled) return
   const text = `${handle.lineQueue.join('\n')}\n`
   handle.lineQueue = []
-  handle.dirtyChain = handle.dirtyChain.then(() =>
-    appendLines(handle.file, text, handle.ctx.logger, PREFIX),
-  )
+  handle.dirtyChain = handle.dirtyChain.then(() => appendLines(handle.file, text, handle.ctx.logger, PREFIX))
 }
 
 /** 防抖紧凑调度（合并短时间多次触发）。 */

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-my-observability",
-  "version": "0.1.3",
+  "version": "0.1.5",
   "description": "DSH 可观测性 + Git 工程工具插件：会话轨迹回放（时间轴）、事件审计（agent 行为记录）、结构化 Git 提交（Conventional Commits）、增量 diff 审查（提交前审查）。DSH web plugin: trajectory replay timeline, event audit log, structured git commits, pre-commit incremental diff review.",
   "type": "module",
   "main": "lib/index.js",