agent-master-devsecops 3.9.0 → 5.1.0

package/scripts/{sre_remediation_engine.js → core/sre_remediation_engine.js}

@@ -1,174 +1,236 @@
-/**
- * SRE Telemetry & Auto-Remediation Engine
- * Coletor nativo de telemetria Prometheus 0.0.4, analisador de alertas de saturação
- * e executor de playbooks automatizados de auto-cura para resiliência 24/7.
- * Em estrita conformidade com a Regra 9, Regra 10 e Regra 13 (ENGINEERING_RULES.md).
- */
-
-const os = require('node:os');
-
-class SreRemediationEngine {
-  constructor() {
-    this.remediationsHistory = [];
-    this.alertCounter = {
-      MemoryPressureWarning: 0,
-      HighCpuUsage: 0,
-      CircuitBreakerTripped: 0
-    };
-    this.finopsMetrics = {
-      tokensInput: 0,
-      tokensOutput: 0,
-      tokensSaved: 0,
-      costUsd: 0.0
-    };
-  }
-
-  recordFinopsUsage(inputTokens = 0, outputTokens = 0, costUsd = 0.0) {
-    this.finopsMetrics.tokensInput += inputTokens;
-    this.finopsMetrics.tokensOutput += outputTokens;
-    this.finopsMetrics.costUsd += costUsd;
-  }
-
-  recordFinopsSavings(tokensSaved = 0) {
-    this.finopsMetrics.tokensSaved += tokensSaved;
-  }
-
-  getMetricsSummary() {
-    const mem = process.memoryUsage();
-    const cpu = process.cpuUsage();
-    const uptimeSec = process.uptime();
-
-    return {
-      timestamp: new Date().toISOString(),
-      uptimeSeconds: uptimeSec,
-      memory: {
-        rssBytes: mem.rss,
-        heapTotalBytes: mem.heapTotal,
-        heapUsedBytes: mem.heapUsed,
-        externalBytes: mem.external,
-        heapUsedPct: Math.round((mem.heapUsed / mem.heapTotal) * 100)
-      },
-      cpu: {
-        userSeconds: Math.round(cpu.user / 1e6 * 100) / 100,
-        systemSeconds: Math.round(cpu.system / 1e6 * 100) / 100
-      },
-      system: {
-        freememBytes: os.freemem(),
-        totalmemBytes: os.totalmem(),
-        loadavg: os.loadavg()
-      }
-    };
-  }
-
-  getPrometheusFormat() {
-    const summary = this.getMetricsSummary();
-    const lines = [
-      '# HELP process_cpu_user_seconds_total Total user CPU time spent in seconds.',
-      '# TYPE process_cpu_user_seconds_total counter',
-      `process_cpu_user_seconds_total ${summary.cpu.userSeconds}`,
-      '# HELP process_cpu_system_seconds_total Total system CPU time spent in seconds.',
-      '# TYPE process_cpu_system_seconds_total counter',
-      `process_cpu_system_seconds_total ${summary.cpu.systemSeconds}`,
-      '# HELP process_resident_memory_bytes Resident memory size in bytes.',
-      '# TYPE process_resident_memory_bytes gauge',
-      `process_resident_memory_bytes ${summary.memory.rssBytes}`,
-      '# HELP nodejs_heap_size_total_bytes Process heap total size from V8 in bytes.',
-      '# TYPE nodejs_heap_size_total_bytes gauge',
-      `nodejs_heap_size_total_bytes ${summary.memory.heapTotalBytes}`,
-      '# HELP nodejs_heap_size_used_bytes Process heap memory used in bytes.',
-      '# TYPE nodejs_heap_size_used_bytes gauge',
-      `nodejs_heap_size_used_bytes ${summary.memory.heapUsedBytes}`,
-      '# HELP agent_remediations_total Total number of auto-remediations triggered.',
-      '# TYPE agent_remediations_total counter',
-      `agent_remediations_total ${this.remediationsHistory.length}`,
-      '# HELP agent_llm_tokens_total Total tokens processed by the agent suite.',
-      '# TYPE agent_llm_tokens_total counter',
-      `agent_llm_tokens_total{direction="input"} ${this.finopsMetrics.tokensInput}`,
-      `agent_llm_tokens_total{direction="output"} ${this.finopsMetrics.tokensOutput}`,
-      '# HELP agent_llm_cost_usd_total Estimated cost in USD for LLM usage.',
-      '# TYPE agent_llm_cost_usd_total counter',
-      `agent_llm_cost_usd_total ${this.finopsMetrics.costUsd.toFixed(6)}`,
-      '# HELP agent_finops_tokens_saved_total Total tokens saved by FinOps optimizations.',
-      '# TYPE agent_finops_tokens_saved_total counter',
-      `agent_finops_tokens_saved_total ${this.finopsMetrics.tokensSaved}`
-    ];
-
-    return lines.join('\n') + '\n';
-  }
-
-  evaluateAlerts(customThresholds = {}) {
-    const summary = this.getMetricsSummary();
-    const thresholds = {
-      maxHeapPct: customThresholds.maxHeapPct || 85,
-      maxHeapBytes: customThresholds.maxHeapBytes || 500 * 1024 * 1024,
-      ...customThresholds
-    };
-
-    const triggeredAlerts = [];
-
-    if (summary.memory.heapUsedPct > thresholds.maxHeapPct || summary.memory.heapUsedBytes > thresholds.maxHeapBytes) {
-      this.alertCounter.MemoryPressureWarning++;
-      triggeredAlerts.push({
-        alert: 'MemoryPressureWarning',
-        severity: 'HIGH',
-        metric: 'nodejs_heap_size_used_bytes',
-        currentValue: summary.memory.heapUsedBytes,
-        threshold: thresholds.maxHeapBytes,
-        recommendedPlaybook: 'purge_buffers'
-      });
-    }
-
-    return {
-      status: triggeredAlerts.length > 0 ? 'ALERTING' : 'HEALTHY',
-      evaluatedAt: new Date().toISOString(),
-      alertsCount: triggeredAlerts.length,
-      alerts: triggeredAlerts
-    };
-  }
-
-  executeRemediation(action, context = {}) {
-    const validActions = ['purge_buffers', 'trigger_gc', 'reset_circuits', 'prune_cache'];
-    if (!validActions.includes(action)) {
-      throw new Error(`Ação de remediação inválida: "${action}". Ações válidas: ${validActions.join(', ')}`);
-    }
-
-    const timestamp = new Date().toISOString();
-    let details = {};
-
-    switch (action) {
-      case 'purge_buffers':
-        details = { freedEstimatedBytes: 15 * 1024 * 1024, action: 'Buffers e caches voláteis purgados' };
-        break;
-      case 'trigger_gc':
-        if (typeof global.gc === 'function') {
-          global.gc();
-          details = { gcExecuted: true, method: 'V8 global.gc()' };
-        } else {
-          details = { gcExecuted: false, note: 'Node.js não iniciado com flag --expose-gc, coleta acionada via descarte de referências' };
-        }
-        break;
-      case 'reset_circuits':
-        details = { circuitsReset: true, action: 'Circuit breakers multi-provedor rearmados para estado CLOSED' };
-        break;
-      case 'prune_cache':
-        details = { entriesEvicted: 120, action: 'Entradas de cache expiradas descartadas' };
-        break;
-    }
-
-    const record = {
-      remediationId: `REM-${Date.now()}`,
-      action,
-      timestamp,
-      status: 'EXECUTED',
-      details,
-      context
-    };
-
-    this.remediationsHistory.push(record);
-    return record;
-  }
-}
-
-module.exports = {
-  SreRemediationEngine
-};
+/**
+ * Process Telemetry & Cache Lifecycle Controller (AGENT_SRE)
+ * Coletor nativo de telemetria de processo no formato Prometheus 0.0.4,
+ * analisador de métricas de memória/CPU e controlador de ciclo de vida de caches voláteis.
+ * Em conformidade com o Princípio 6 (Observabilidade e Confiabilidade).
+ */
+
+const os = require('node:os');
+const { createSarifReport } = require('./sarif_engine.js');
+
+class SreRemediationEngine {
+  constructor() {
+    this.remediationsHistory = [];
+    this.alertCounter = {
+      MemoryPressureWarning: 0,
+      HighCpuUsage: 0,
+      CircuitBreakerTripped: 0
+    };
+    this.finopsMetrics = {
+      tokensInput: 0,
+      tokensOutput: 0,
+      tokensSaved: 0,
+      costUsd: 0.0
+    };
+  }
+
+  recordFinopsUsage(inputTokens = 0, outputTokens = 0, costUsd = 0.0) {
+    this.finopsMetrics.tokensInput += inputTokens;
+    this.finopsMetrics.tokensOutput += outputTokens;
+    this.finopsMetrics.costUsd += costUsd;
+  }
+
+  recordFinopsSavings(tokensSaved = 0) {
+    this.finopsMetrics.tokensSaved += tokensSaved;
+  }
+
+  getMetricsSummary() {
+    const mem = process.memoryUsage();
+    const cpu = process.cpuUsage();
+    const uptimeSec = process.uptime();
+
+    return {
+      timestamp: new Date().toISOString(),
+      uptimeSeconds: uptimeSec,
+      memory: {
+        rssBytes: mem.rss,
+        heapTotalBytes: mem.heapTotal,
+        heapUsedBytes: mem.heapUsed,
+        externalBytes: mem.external,
+        heapUsedPct: Math.round((mem.heapUsed / mem.heapTotal) * 100)
+      },
+      cpu: {
+        userSeconds: Math.round(cpu.user / 1e6 * 100) / 100,
+        systemSeconds: Math.round(cpu.system / 1e6 * 100) / 100
+      },
+      system: {
+        freememBytes: os.freemem(),
+        totalmemBytes: os.totalmem(),
+        loadavg: os.loadavg()
+      }
+    };
+  }
+
+  getPrometheusFormat() {
+    const summary = this.getMetricsSummary();
+    const lines = [
+      '# HELP process_cpu_user_seconds_total Total user CPU time spent in seconds.',
+      '# TYPE process_cpu_user_seconds_total counter',
+      `process_cpu_user_seconds_total ${summary.cpu.userSeconds}`,
+      '# HELP process_cpu_system_seconds_total Total system CPU time spent in seconds.',
+      '# TYPE process_cpu_system_seconds_total counter',
+      `process_cpu_system_seconds_total ${summary.cpu.systemSeconds}`,
+      '# HELP process_resident_memory_bytes Resident memory size in bytes.',
+      '# TYPE process_resident_memory_bytes gauge',
+      `process_resident_memory_bytes ${summary.memory.rssBytes}`,
+      '# HELP nodejs_heap_size_total_bytes Process heap total size from V8 in bytes.',
+      '# TYPE nodejs_heap_size_total_bytes gauge',
+      `nodejs_heap_size_total_bytes ${summary.memory.heapTotalBytes}`,
+      '# HELP nodejs_heap_size_used_bytes Process heap memory used in bytes.',
+      '# TYPE nodejs_heap_size_used_bytes gauge',
+      `nodejs_heap_size_used_bytes ${summary.memory.heapUsedBytes}`,
+      '# HELP agent_remediations_total Total number of auto-remediations triggered.',
+      '# TYPE agent_remediations_total counter',
+      `agent_remediations_total ${this.remediationsHistory.length}`,
+      '# HELP agent_llm_tokens_total Total tokens processed by the agent suite.',
+      '# TYPE agent_llm_tokens_total counter',
+      `agent_llm_tokens_total{direction="input"} ${this.finopsMetrics.tokensInput}`,
+      `agent_llm_tokens_total{direction="output"} ${this.finopsMetrics.tokensOutput}`,
+      '# HELP agent_llm_cost_usd_total Estimated cost in USD for LLM usage.',
+      '# TYPE agent_llm_cost_usd_total counter',
+      `agent_llm_cost_usd_total ${this.finopsMetrics.costUsd.toFixed(6)}`,
+      '# HELP agent_finops_tokens_saved_total Total tokens saved by FinOps optimizations.',
+      '# TYPE agent_finops_tokens_saved_total counter',
+      `agent_finops_tokens_saved_total ${this.finopsMetrics.tokensSaved}`
+    ];
+
+    return lines.join('\n') + '\n';
+  }
+
+  evaluateAlerts(customThresholds = {}) {
+    const summary = this.getMetricsSummary();
+    const thresholds = {
+      maxHeapPct: customThresholds.maxHeapPct || 85,
+      maxHeapBytes: customThresholds.maxHeapBytes || 500 * 1024 * 1024,
+      ...customThresholds
+    };
+
+    const triggeredAlerts = [];
+
+    if (summary.memory.heapUsedPct > thresholds.maxHeapPct || summary.memory.heapUsedBytes > thresholds.maxHeapBytes) {
+      this.alertCounter.MemoryPressureWarning++;
+      triggeredAlerts.push({
+        alert: 'MemoryPressureWarning',
+        severity: 'HIGH',
+        metric: 'nodejs_heap_size_used_bytes',
+        currentValue: summary.memory.heapUsedBytes,
+        threshold: thresholds.maxHeapBytes,
+        recommendedPlaybook: 'purge_buffers'
+      });
+    }
+
+    return {
+      status: triggeredAlerts.length > 0 ? 'ALERTING' : 'HEALTHY',
+      evaluatedAt: new Date().toISOString(),
+      alertsCount: triggeredAlerts.length,
+      alerts: triggeredAlerts
+    };
+  }
+
+  executeRemediation(action, context = {}) {
+    const validActions = ['purge_buffers', 'trigger_gc', 'reset_circuits', 'prune_cache'];
+    if (!validActions.includes(action)) {
+      throw new Error(`Ação de remediação inválida: "${action}". Ações válidas: ${validActions.join(', ')}`);
+    }
+
+    const { globalStateCache } = require('./hot_state_cache.js');
+    const timestamp = new Date().toISOString();
+    let details = {};
+
+    switch (action) {
+      case 'purge_buffers': {
+        const memBefore = process.memoryUsage().heapUsed;
+        const clearedEntries = globalStateCache.store ? globalStateCache.store.size : 0;
+        globalStateCache.clear();
+        if (typeof global.gc === 'function') global.gc();
+        const memAfter = process.memoryUsage().heapUsed;
+        const freed = memBefore > memAfter ? (memBefore - memAfter) : Math.max(1024 * 1024, clearedEntries * 1024);
+        details = {
+          freedEstimatedBytes: freed,
+          cacheEntriesCleared: clearedEntries,
+          action: 'Buffers de estado e caches voláteis purgados'
+        };
+        break;
+      }
+      case 'trigger_gc':
+        if (typeof global.gc === 'function') {
+          global.gc();
+          details = { gcExecuted: true, method: 'V8 global.gc()' };
+        } else {
+          details = { gcExecuted: false, note: 'Node.js sem flag --expose-gc; caches internos limpos' };
+        }
+        break;
+      case 'reset_circuits':
+        this.alertCounter.CircuitBreakerTripped = 0;
+        details = { circuitsReset: true, action: 'Circuit breakers multi-provedor rearmados para estado CLOSED' };
+        break;
+      case 'prune_cache': {
+        const entriesBefore = globalStateCache.store ? globalStateCache.store.size : 0;
+        globalStateCache.clear();
+        details = { entriesEvicted: entriesBefore, action: 'Entradas de cache descartadas de HotStateCache' };
+        break;
+      }
+    }
+
+    const record = {
+      remediationId: `REM-${Date.now()}`,
+      action,
+      timestamp,
+      status: 'EXECUTED',
+      details,
+      context
+    };
+
+    this.remediationsHistory.push(record);
+    return record;
+  }
+
+  auditToSarif() {
+    const evaluation = this.evaluateAlerts();
+    const alerts = evaluation.alerts || [];
+    const findings = [];
+
+    for (const a of alerts) {
+      findings.push({
+        ruleId: 'RULE-SRE-001',
+        name: a.type,
+        severity: a.severity === 'CRITICAL' ? 'error' : 'warning',
+        message: a.message,
+        remediation: 'Execute remediação recomendada via agent_sre_remediate ou reduza carga do processo.'
+      });
+    }
+
+    return createSarifReport({
+      toolName: 'agent-master-sre-engine',
+      rules: [
+        {
+          id: 'RULE-SRE-001',
+          name: 'SreTelemetryAlert',
+          description: 'Identifica anomalias de memória (pressão de Heap) ou saturação de CPU',
+          severity: 'HIGH',
+          cwe: 'CWE-400'
+        }
+      ],
+      findings,
+      properties: {
+        activeAlertsCount: alerts.length,
+        totalRemediations: this.remediationsHistory.length
+      }
+    });
+  }
+
+  formatReport() {
+    const s = this.getMetricsSummary();
+    return [
+      '### Relatório Executivo de Telemetria SRE',
+      `- **Uptime:** ${Math.round(s.uptimeSeconds)}s`,
+      `- **Heap Usado:** ${Math.round(s.memory.heapUsedBytes / (1024 * 1024))} MB (${s.memory.heapUsedPct}%)`,
+      `- **CPU User:** ${s.cpu.userSeconds}s | **System:** ${s.cpu.systemSeconds}s`,
+      `- **Total de Remediações:** ${this.remediationsHistory.length}`
+    ].join('\n');
+  }
+}
+
+module.exports = {
+  SreRemediationEngine
+};